From junyaoz at stanford.edu Sun Oct 3 21:14:37 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Mon, 4 Oct 2021 04:14:37 +0000
Subject: [theory-seminar] Theory Lunch 10/07: Soheil Behnezhad
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Soheil will tell us about: Time-Optimal Sublinear Algorithms for Matching and Vertex Cover
Abstract: Over the past two decades there has been a growing interest in estimating various graph parameters, such as the size of maximum matching (MM) and minimum vertex cover (MVC), in time that is sublinear in the input size. In this talk, I will start by discussing a powerful method of the literature for obtaining such sublinear time algorithms based on the randomized greedy maximal matching (RGMM) algorithm. I will then present an improved and near-tight analysis of the ?average query complexity? of RGMM, leading to algorithms for estimating the size of MM and MVC up to a factor of (almost) two in time near-linear in the number of vertices. The new algorithms turn out to be information theoretically time-optimal (up to logarithmic factors), culminating a long line of work on these two problems.
Based on https://arxiv.org/abs/2106.02942 (to appear in FOCS'21).
This week's lunch menu is as follows:
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
By the way, we still have a couple of slots available for talks (Oct 28 and Dec 2). If you would like to share your recent cool results with the group, please let me know.
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Sun Oct 3 21:14:37 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Mon, 4 Oct 2021 04:14:37 +0000
Subject: [theory-seminar] Theory Lunch 10/07: Soheil Behnezhad
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Soheil will tell us about: Time-Optimal Sublinear Algorithms for Matching and Vertex Cover
Abstract: Over the past two decades there has been a growing interest in estimating various graph parameters, such as the size of maximum matching (MM) and minimum vertex cover (MVC), in time that is sublinear in the input size. In this talk, I will start by discussing a powerful method of the literature for obtaining such sublinear time algorithms based on the randomized greedy maximal matching (RGMM) algorithm. I will then present an improved and near-tight analysis of the ?average query complexity? of RGMM, leading to algorithms for estimating the size of MM and MVC up to a factor of (almost) two in time near-linear in the number of vertices. The new algorithms turn out to be information theoretically time-optimal (up to logarithmic factors), culminating a long line of work on these two problems.
Based on https://arxiv.org/abs/2106.02942 (to appear in FOCS'21).
This week's lunch menu is as follows:
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
By the way, we still have a couple of slots available for talks (Oct 28 and Dec 2). If you would like to share your recent cool results with the group, please let me know.
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From gvaliant at cs.stanford.edu Mon Oct 4 12:03:14 2021
From: gvaliant at cs.stanford.edu (Gregory Valiant)
Date: Mon, 4 Oct 2021 12:03:14 -0700
Subject: [theory-seminar] combinatorics seminar thurs@3 w. Persi Diaconis
In-Reply-To:
References:
Message-ID:
Hi Friends,
This Thursday's combinatorics seminar might be of interest to many of you.
Info below.
Cheers,
-g
------------
This Thursday, October we have the first in-person combinatorics seminar
talk in more than a year by our very own Persi Diaconis!
What: Stanford Combinatorics Seminar
When: Thursday October, 3pm-4pm
Room: 384-H (Building 380, Fourth Floor, Room H)
Speaker: Persi Diaconis (Stanford)
Title: ISOMORPHISMS BETWEEN RANDOM GRAPHS
Abstract: Pick two Erd?s-R?nyi G(n,1/2) graphs at random. What's the chance
that they are isomorphic? Small right? How small? It's at most n!/2^(n
choose 2) so less than 10^(-1300) when n= 100. That's small. OK, now let n=
infinity. The chance that they are isomorphic is one(!). Thus we may talk
about THE random graph R. I will review it's many strange properties and
then ask if there is any finite shadow of this isomorphism. In joint work
with Sourav Chatterjee we show that the largest isomorphic induced subgraph
of two independent G(n,1/2) graphs is about 4 log n. Constraint
satisfaction problems suggest related problems. There are some surprising
results and many open problems.
On Mon, Sep 20, 2021 at 10:00 AM David Wajc wrote:
> Hello everyone,
>
> Theory lunch is back in person (outdoors)!
>
> This *Thursday, noon,* we will convene in one of the tree pits in the Engineering
> quad
>
> for an hour of socializing and catching up with our colleagues in the
> group, some of whom we haven't seen in nearly a year and a half, and some
> of whom have joined since, and so we may not have even met in person before!
>
> The plan for the rest of the quarter is to have *in-person talks*, also
> in the engineering quad, on a whiteboard. This will follow the traditional
> format: 30-minute socializing, followed by a 30-minute talk.
>
> This quarter *Junyao Zhao (cc'ed)* will be organizing theory lunches.
> (Thanks, Junyao!) If you're in town and interested in sharing some cool
> math with the group, please reach out to him to see if there are available
> slots.
>
> Wishing you a fun and productive quarter, and a great start to the
> academic year.
>
> Best,
> David
> _______________________________________________
> theory-seminar mailing list
> theory-seminar at lists.stanford.edu
> https://mailman.stanford.edu/mailman/listinfo/theory-seminar
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From gvaliant at cs.stanford.edu Mon Oct 4 12:03:14 2021
From: gvaliant at cs.stanford.edu (Gregory Valiant)
Date: Mon, 4 Oct 2021 12:03:14 -0700
Subject: [theory-seminar] combinatorics seminar thurs@3 w. Persi Diaconis
In-Reply-To:
References:
Message-ID:
Hi Friends,
This Thursday's combinatorics seminar might be of interest to many of you.
Info below.
Cheers,
-g
------------
This Thursday, October we have the first in-person combinatorics seminar
talk in more than a year by our very own Persi Diaconis!
What: Stanford Combinatorics Seminar
When: Thursday October, 3pm-4pm
Room: 384-H (Building 380, Fourth Floor, Room H)
Speaker: Persi Diaconis (Stanford)
Title: ISOMORPHISMS BETWEEN RANDOM GRAPHS
Abstract: Pick two Erd?s-R?nyi G(n,1/2) graphs at random. What's the chance
that they are isomorphic? Small right? How small? It's at most n!/2^(n
choose 2) so less than 10^(-1300) when n= 100. That's small. OK, now let n=
infinity. The chance that they are isomorphic is one(!). Thus we may talk
about THE random graph R. I will review it's many strange properties and
then ask if there is any finite shadow of this isomorphism. In joint work
with Sourav Chatterjee we show that the largest isomorphic induced subgraph
of two independent G(n,1/2) graphs is about 4 log n. Constraint
satisfaction problems suggest related problems. There are some surprising
results and many open problems.
On Mon, Sep 20, 2021 at 10:00 AM David Wajc wrote:
> Hello everyone,
>
> Theory lunch is back in person (outdoors)!
>
> This *Thursday, noon,* we will convene in one of the tree pits in the Engineering
> quad
>
> for an hour of socializing and catching up with our colleagues in the
> group, some of whom we haven't seen in nearly a year and a half, and some
> of whom have joined since, and so we may not have even met in person before!
>
> The plan for the rest of the quarter is to have *in-person talks*, also
> in the engineering quad, on a whiteboard. This will follow the traditional
> format: 30-minute socializing, followed by a 30-minute talk.
>
> This quarter *Junyao Zhao (cc'ed)* will be organizing theory lunches.
> (Thanks, Junyao!) If you're in town and interested in sharing some cool
> math with the group, please reach out to him to see if there are available
> slots.
>
> Wishing you a fun and productive quarter, and a great start to the
> academic year.
>
> Best,
> David
> _______________________________________________
> theory-seminar mailing list
> theory-seminar at lists.stanford.edu
> https://mailman.stanford.edu/mailman/listinfo/theory-seminar
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From marykw at stanford.edu Tue Oct 5 18:05:52 2021
From: marykw at stanford.edu (Mary Wootters)
Date: Tue, 5 Oct 2021 18:05:52 -0700
Subject: [theory-seminar] Volunteer needed to help coordinate mentorship
program
Message-ID:
Hello theory Ph.D. students,
Along with Dorsa Sadigh, Chelsea Finn and others, I'm helping out with a
mentoring program in the CS department that pairs undergraduates with
graduate student mentors; some of you participated in this program as
mentors last year. The goal is to give undergraduates -- especially those
from underrepresented groups -- an idea about what research and PhD life is
like, to give them someone they can talk to about getting involved in
research, etc.
We need a theory PhD student to help coordinate the program. The workload
is pretty light: this person would help recruit theory grad students to
serve as mentors, and would be in charge of matching them to
theory-interested undergrads and sending occasional reminder emails and
announcements. Saba did it last year but has now graduated -- you could
ask him about the workload if you want more information.
Please let me know if you are interested in helping out!
Thanks!
Mary
ps. If you are interested in being a mentor/mentee for this program, keep
an eye out for an email about that coming soon!
--
Mary Wootters (she/her)
Assistant Professor of Computer Science and Electrical Engineering
Stanford University
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From margalit.r.glasgow at gmail.com Thu Oct 7 10:59:58 2021
From: margalit.r.glasgow at gmail.com (Margalit Glasgow)
Date: Thu, 7 Oct 2021 10:59:58 -0700
Subject: [theory-seminar] Quals Talk Wednesday Oct 13th 1-2pm: Beyond PAC
Learning
Message-ID:
Hi theory/CS folk,
I'll be doing my theory quals talk next week on *Wednesday Oct 13th at 1pm*.
Everyone is welcome to join.
I'll be talking about some papers on extensions/modifications of the
traditional PAC learning setting that can yield better
distribution-dependent and data-dependent bounds.
It will be outside in the Engineering Quad, in whichever tree pit has the
whiteboard.
Margalit
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 7 11:53:31 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 7 Oct 2021 18:53:31 +0000
Subject: [theory-seminar] Theory Lunch 10/07: Soheil Behnezhad
In-Reply-To:
References:
Message-ID:
Gentle reminder: this is happening in 10 minutes.
On Oct 4, 2021, at 9:00 AM, Junyao Zhao wrote:
?
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Soheil will tell us about: Time-Optimal Sublinear Algorithms for Matching and Vertex Cover
Abstract: Over the past two decades there has been a growing interest in estimating various graph parameters, such as the size of maximum matching (MM) and minimum vertex cover (MVC), in time that is sublinear in the input size. In this talk, I will start by discussing a powerful method of the literature for obtaining such sublinear time algorithms based on the randomized greedy maximal matching (RGMM) algorithm. I will then present an improved and near-tight analysis of the ?average query complexity? of RGMM, leading to algorithms for estimating the size of MM and MVC up to a factor of (almost) two in time near-linear in the number of vertices. The new algorithms turn out to be information theoretically time-optimal (up to logarithmic factors), culminating a long line of work on these two problems.
Based on https://arxiv.org/abs/2106.02942 (to appear in FOCS'21).
This week's lunch menu is as follows:
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
By the way, we still have a couple of slots available for talks (Oct 28 and Dec 2). If you would like to share your recent cool results with the group, please let me know.
Cheers,
Junyao
_______________________________________________
theory-seminar mailing list
theory-seminar at lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/theory-seminar
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 7 11:53:31 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 7 Oct 2021 18:53:31 +0000
Subject: [theory-seminar] Theory Lunch 10/07: Soheil Behnezhad
In-Reply-To:
References:
Message-ID:
Gentle reminder: this is happening in 10 minutes.
On Oct 4, 2021, at 9:00 AM, Junyao Zhao wrote:
?
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Soheil will tell us about: Time-Optimal Sublinear Algorithms for Matching and Vertex Cover
Abstract: Over the past two decades there has been a growing interest in estimating various graph parameters, such as the size of maximum matching (MM) and minimum vertex cover (MVC), in time that is sublinear in the input size. In this talk, I will start by discussing a powerful method of the literature for obtaining such sublinear time algorithms based on the randomized greedy maximal matching (RGMM) algorithm. I will then present an improved and near-tight analysis of the ?average query complexity? of RGMM, leading to algorithms for estimating the size of MM and MVC up to a factor of (almost) two in time near-linear in the number of vertices. The new algorithms turn out to be information theoretically time-optimal (up to logarithmic factors), culminating a long line of work on these two problems.
Based on https://arxiv.org/abs/2106.02942 (to appear in FOCS'21).
This week's lunch menu is as follows:
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
By the way, we still have a couple of slots available for talks (Oct 28 and Dec 2). If you would like to share your recent cool results with the group, please let me know.
Cheers,
Junyao
_______________________________________________
theory-seminar mailing list
theory-seminar at lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/theory-seminar
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From marykw at stanford.edu Thu Oct 7 14:34:11 2021
From: marykw at stanford.edu (Mary Wootters)
Date: Thu, 7 Oct 2021 14:34:11 -0700
Subject: [theory-seminar] LeT-All's Graduate Application Support program
Message-ID:
Hi all,
Undergrads on this list might be interested in the following announcement
about LeT-All's program to support applications to graduate school.
Best,
Mary
===================================
Hi all,
We are pleased to announce LeT-All?s pilot Graduate Application Support
program , which is organized by
the Learning Theory Alliance (LeT-All) and Women in Machine Learning Theory
(WiML-T). The purpose of this program is to provide graduate application
support and feedback to students who are interested in applying to PhD
programs in theory of machine learning. To be considered for the program,
please apply by October 18 by filling out this form
.
The application is open to all. We especially encourage students from
marginalized communities to apply for this program. You will be
notified by October
25 whether your application has been accepted. Accepted applicants will be
matched with a senior mentor for two 45-minute mentoring sessions. The first
session will be held during November 8-12 and will include introductions,
general advice, and a review of a (rough) draft of their application
material. The second session held during November 22-24 will focus more on
the application materials and changes made since the first meeting.
For the first year of this program, we will focus on supporting students
who will be applying to PhD programs in the US and Canada, although we
intend to expand our reach in subsequent years. Participants will be
selected based on the proximity of their research interests to the theory
of machine learning, the availability of potential matches, and the
potential benefit they would receive from this program.
This program is part of our broader community-building initiative called
the Learning Theory Alliance. Check out http://let-all.com/ for more
details about our workshops and other community events. Sign up
to join our mailing
list or follow us on social media to stay up
to date on our activities.
Best,
Surbhi Goel, Nika Haghtalab and Ellen Vitercik
--
Mary Wootters (she/her)
Assistant Professor of Computer Science and Electrical Engineering
Stanford University
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From gblanc at stanford.edu Thu Oct 7 20:11:13 2021
From: gblanc at stanford.edu (Guy Blanc)
Date: Thu, 7 Oct 2021 20:11:13 -0700
Subject: [theory-seminar] Mentor undergrads interested in Theory! Low time
commitment (~30 min/month)
Message-ID:
Hi all,
We are starting up the undergraduate mentoring program again. The goal is
to mentor promising undergraduates from underrepresented groups and
ultimately have some of them involved in research in the future, either at
Stanford CS or elsewhere.
For more information, see the following doc. The time-commitment to be a
mentor is quite low, ~30 min/month.
https://docs.google.com/document/d/1tvOmMwDe7xHpcA5oYTGdGXFu_p8_BMM6kxxBebINC-s/edit?usp=sharing
If you are a grad student or postdoc and want to participate this year,
fill out this form before October 15, 2021:
https://forms.gle/EcJ8AkaV7De7hmem7
Any feedback/suggestions are welcome!
The faculty advisors for this program are: Mary Wootters, Zakir Durumeric,
Chelsea Finn, Subhasish Mitra, and Dorsa Sadigh. Please contact any of them
directly, or myself (Guy), if you have any questions!
Cheers,
Guy
(This email was mostly copy/pasted from one by Dorsa, but I strongly
endorse the message.)
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Mon Oct 11 00:35:36 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Mon, 11 Oct 2021 07:35:36 +0000
Subject: [theory-seminar] Theory Lunch 10/14: Erik Waingarten
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Erik will tell us about: Sketches for Clustering in ?pp
Abstract: We give streaming and distributed algorithms for center-based (k,p)-clustering in ?p when p ? [1, 2]. The input is a dataset x1,?, xn ? ?d and ? > 0, and using space poly(log(nd), k, 1/?), we approximate the optimal (k,p)-clustering cost. The above result was known when p = 2 (by utilizing a Johnson-Lindenstrauss projection [Cohen-Elder-Musco-Musco-Persu '15, Makarychev-Makarychev-Razenstheyn '20]); however, for p ? 2, all algorithms incurred space complexity linear in d.
Joint with Moses Charikar.
Lunch menu (both are vegetarian):
Greek salad (with pepper, cucumber, tomato, Kalamata olives and feta cheese),
Caesar salad (Oven roasted garlic croutons and parmesan cheese).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Mon Oct 11 00:35:36 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Mon, 11 Oct 2021 07:35:36 +0000
Subject: [theory-seminar] Theory Lunch 10/14: Erik Waingarten
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Erik will tell us about: Sketches for Clustering in ?pp
Abstract: We give streaming and distributed algorithms for center-based (k,p)-clustering in ?p when p ? [1, 2]. The input is a dataset x1,?, xn ? ?d and ? > 0, and using space poly(log(nd), k, 1/?), we approximate the optimal (k,p)-clustering cost. The above result was known when p = 2 (by utilizing a Johnson-Lindenstrauss projection [Cohen-Elder-Musco-Musco-Persu '15, Makarychev-Makarychev-Razenstheyn '20]); however, for p ? 2, all algorithms incurred space complexity linear in d.
Joint with Moses Charikar.
Lunch menu (both are vegetarian):
Greek salad (with pepper, cucumber, tomato, Kalamata olives and feta cheese),
Caesar salad (Oven roasted garlic croutons and parmesan cheese).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From tavorb at stanford.edu Mon Oct 11 23:54:18 2021
From: tavorb at stanford.edu (Tavor Baharav)
Date: Mon, 11 Oct 2021 23:54:18 -0700
Subject: [theory-seminar] =?utf-8?q?=22Experimentation_and_Decision-Making?=
=?utf-8?q?_in_Two-Sided_Marketplaces=3A_The_Impact_of_Interference?=
=?utf-8?q?_/_Simple_Agent=2C_Complex_Environment=3A_Efficient_Rein?=
=?utf-8?q?forcement_Learning_with_Agent_States=22_=E2=80=93_Hannah?=
=?utf-8?q?_Li_/_Shi_Dong_=28Thu=2C_14-Oct_=40_4=3A00pm=29?=
Message-ID:
Experimentation and Decision-Making in Two-Sided Marketplaces: The Impact
of Interference / Simple Agent, Complex Environment: Efficient
Reinforcement Learning with Agent StatesHannah Li / Shi Dong ? PhD
Students, Stanford
Thu, 14-Oct / 4:00pm / Packard 101
Senior graduate students Hannah Li and Shi Dong will each give a 25min talk.
Note: These talks will be held in person in Packard 101, and will start at
4pm. The talks will be streamed on Zoom for those who cannot attend:
https://stanford.zoom.us/meeting/register/tJckfuCurzkvEtKKOBvDCrPv3McapgP6HygJ.
Please join us for coffee starting at 3:30pm at the Grove outside
Packard.Experimentation
and Decision-Making in Two-Sided Marketplaces: The Impact of Interference
Abstract
Marketplace platforms use experiments (also known as ?A/B tests?) as a
method for making data-driven decisions. When platforms consider
introducing a new feature, they often first run an experiment to test the
feature on a subset of users and then use this data to decide whether to
launch the feature platform-wide. However, it is well documented that
estimates of the treatment effect arising from these experiments may be
biased, due to the presence of interference. In this talk, we survey a
collection of recent results and insights we have developed on
experimentation and decision-making in two-sided marketplaces. In
particular, we study the bias that interference creates in both the
treatment effect estimates as well as standard error estimates, and show
how both types of biases affect the platform?s ability to make decisions.
We show that for a large class of interventions (?positive interventions?),
these biases cause the platform to launch too often. Through simulations
calibrated to real-world data, we show that in many settings the treatment
effect bias impacts decision-making more than the standard error bias.
Based on joint work with Ramesh Johari, Inessa Liskovich, Gabriel
Weintraub, and Geng Zhao.
Bio
Hannah is a PhD Candidate at Stanford University, where she is advised by
Ramesh Johari and Gabriel Weintraub. She is part of the Operations Research
group in MS&E and the Research in Algorithms and Incentives in
Networks(RAIN) group. Her research uses techniques from math modeling,
optimization, and causal inference in order to analyze and design data
science methodology for marketplace platforms. Before coming to Stanford,
she graduated from Pomona College with a degree in mathematics.
Simple Agent, Complex Environment: Efficient Reinforcement Learning with
Agent StatesAbstract
In this work, we design a simple reinforcement learning (RL) agent that
implements an optimistic version of Q-learning and establish through regret
analysis that this agent can operate with some level of competence in an
arbitrarily complex environment. While we leverage concepts from the
literature on provably efficient RL, we consider a general
agent-environment interface and provide a novel agent design and analysis.
This level of generality positions our results to inform the design of
future agents for operation in complex real environments. We establish
that, as time progresses, our agent performs competitively relative to
policies that require longer times to evaluate. The time it takes to
approach asymptotic performance is polynomial in the complexity of the
agent?s state representation and the time required to evaluate the best
policy that the agent can represent. Notably, there is no dependence on the
complexity of the environment. The ultimate per-period performance loss of
the agent is bounded by a constant multiple of a measure of distortion
introduced by the agent?s state representation. This work is the first to
establish that an algorithm approaches this asymptotic condition within a
tractable time frame.
Bio
Shi Dong is a sixth-year PhD student in Electrical Engineering at Stanford
University, where he is advised by Prof. Benjamin Van Roy. Prior to
Stanford, he received his undergraduate degree from Tsinghua University. He
is interested in using mathematical tools to understand how successful
reinforcement learning agents are designed. His recent work has been
selected as a finalist in the INFORMS George Nicholson Student Paper
Competition. He is on the 2021-2022 academic job market.
*This talk is hosted by the ISL Colloquium
. To receive talk announcements, subscribe
to the mailing list isl-colloq at lists.stanford.edu
.*
------------------------------
Mailing list: https://mailman.stanford.edu/mailman/listinfo/isl-colloq
This talk: http://isl.stanford.edu/talks/talks/2021q4/hannah-li-shi-dong/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From margalit.r.glasgow at gmail.com Tue Oct 12 20:14:14 2021
From: margalit.r.glasgow at gmail.com (Margalit Glasgow)
Date: Tue, 12 Oct 2021 20:14:14 -0700
Subject: [theory-seminar] Quals Talk Wednesday Oct 13th 1-2pm: Beyond
PAC Learning
In-Reply-To:
References:
Message-ID:
Reminder that this is happening tomorrow at 1pm in the engineering quad, in
the tree pit with the white boards.
On Thu, Oct 7, 2021, 10:59 AM Margalit Glasgow
wrote:
> Hi theory/CS folk,
>
> I'll be doing my theory quals talk next week on *Wednesday Oct 13th at
> 1pm*. Everyone is welcome to join.
>
> I'll be talking about some papers on extensions/modifications of the
> traditional PAC learning setting that can yield better
> distribution-dependent and data-dependent bounds.
>
> It will be outside in the Engineering Quad, in whichever tree pit has the
> whiteboard.
>
> Margalit
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From margalit.r.glasgow at gmail.com Tue Oct 12 23:28:37 2021
From: margalit.r.glasgow at gmail.com (Margalit Glasgow)
Date: Tue, 12 Oct 2021 23:28:37 -0700
Subject: [theory-seminar] Quals Talk Wednesday Oct 13th 1-2pm: Beyond
PAC Learning
In-Reply-To:
References:
Message-ID:
One small edit I forgot: we're going to be starting at 1:10, not 1.
Margalit
On Tue, Oct 12, 2021, 8:14 PM Margalit Glasgow
wrote:
> Reminder that this is happening tomorrow at 1pm in the engineering quad,
> in the tree pit with the white boards.
>
> On Thu, Oct 7, 2021, 10:59 AM Margalit Glasgow <
> margalit.r.glasgow at gmail.com> wrote:
>
>> Hi theory/CS folk,
>>
>> I'll be doing my theory quals talk next week on *Wednesday Oct 13th at
>> 1pm*. Everyone is welcome to join.
>>
>> I'll be talking about some papers on extensions/modifications of the
>> traditional PAC learning setting that can yield better
>> distribution-dependent and data-dependent bounds.
>>
>> It will be outside in the Engineering Quad, in whichever tree pit has the
>> whiteboard.
>>
>> Margalit
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From tavorb at stanford.edu Thu Oct 14 10:22:54 2021
From: tavorb at stanford.edu (Tavor Baharav)
Date: Thu, 14 Oct 2021 10:22:54 -0700
Subject: [theory-seminar]
=?utf-8?q?=22Experimentation_and_Decision-Making?=
=?utf-8?q?_in_Two-Sided_Marketplaces=3A_The_Impact_of_Interference?=
=?utf-8?q?_/_Simple_Agent=2C_Complex_Environment=3A_Efficient_Rein?=
=?utf-8?q?forcement_Learning_with_Agent_States=22_=E2=80=93_Hannah?=
=?utf-8?q?_Li_/_Shi_Dong_=28Thu=2C_14-Oct_=40_4=3A00pm=29?=
In-Reply-To:
References:
Message-ID:
A gentle reminder that these talks are today at 4:00pm in Packard 101.
Please join us at 3:30pm at the Grove outside Packard for coffee and
snacks.
On Mon, Oct 11, 2021 at 11:54 PM Tavor Baharav wrote:
> Experimentation and Decision-Making in Two-Sided Marketplaces: The Impact
> of Interference / Simple Agent, Complex Environment: Efficient
> Reinforcement Learning with Agent StatesHannah Li / Shi Dong ? PhD
> Students, Stanford
>
> Thu, 14-Oct / 4:00pm / Packard 101
>
> Senior graduate students Hannah Li and Shi Dong will each give a 25min
> talk.
> Note: These talks will be held in person in Packard 101, and will start at
> 4pm. The talks will be streamed on Zoom for those who cannot attend:
> https://stanford.zoom.us/meeting/register/tJckfuCurzkvEtKKOBvDCrPv3McapgP6HygJ.
> Please join us for coffee starting at 3:30pm at the Grove outside Packard.Experimentation
> and Decision-Making in Two-Sided Marketplaces: The Impact of Interference
> Abstract
>
> Marketplace platforms use experiments (also known as ?A/B tests?) as a
> method for making data-driven decisions. When platforms consider
> introducing a new feature, they often first run an experiment to test the
> feature on a subset of users and then use this data to decide whether to
> launch the feature platform-wide. However, it is well documented that
> estimates of the treatment effect arising from these experiments may be
> biased, due to the presence of interference. In this talk, we survey a
> collection of recent results and insights we have developed on
> experimentation and decision-making in two-sided marketplaces. In
> particular, we study the bias that interference creates in both the
> treatment effect estimates as well as standard error estimates, and show
> how both types of biases affect the platform?s ability to make decisions.
> We show that for a large class of interventions (?positive interventions?),
> these biases cause the platform to launch too often. Through simulations
> calibrated to real-world data, we show that in many settings the treatment
> effect bias impacts decision-making more than the standard error bias.
> Based on joint work with Ramesh Johari, Inessa Liskovich, Gabriel
> Weintraub, and Geng Zhao.
> Bio
>
> Hannah is a PhD Candidate at Stanford University, where she is advised by
> Ramesh Johari and Gabriel Weintraub. She is part of the Operations Research
> group in MS&E and the Research in Algorithms and Incentives in
> Networks(RAIN) group. Her research uses techniques from math modeling,
> optimization, and causal inference in order to analyze and design data
> science methodology for marketplace platforms. Before coming to Stanford,
> she graduated from Pomona College with a degree in mathematics.
> Simple Agent, Complex Environment: Efficient Reinforcement Learning with
> Agent StatesAbstract
>
> In this work, we design a simple reinforcement learning (RL) agent that
> implements an optimistic version of Q-learning and establish through regret
> analysis that this agent can operate with some level of competence in an
> arbitrarily complex environment. While we leverage concepts from the
> literature on provably efficient RL, we consider a general
> agent-environment interface and provide a novel agent design and analysis.
> This level of generality positions our results to inform the design of
> future agents for operation in complex real environments. We establish
> that, as time progresses, our agent performs competitively relative to
> policies that require longer times to evaluate. The time it takes to
> approach asymptotic performance is polynomial in the complexity of the
> agent?s state representation and the time required to evaluate the best
> policy that the agent can represent. Notably, there is no dependence on the
> complexity of the environment. The ultimate per-period performance loss of
> the agent is bounded by a constant multiple of a measure of distortion
> introduced by the agent?s state representation. This work is the first to
> establish that an algorithm approaches this asymptotic condition within a
> tractable time frame.
> Bio
>
> Shi Dong is a sixth-year PhD student in Electrical Engineering at Stanford
> University, where he is advised by Prof. Benjamin Van Roy. Prior to
> Stanford, he received his undergraduate degree from Tsinghua University. He
> is interested in using mathematical tools to understand how successful
> reinforcement learning agents are designed. His recent work has been
> selected as a finalist in the INFORMS George Nicholson Student Paper
> Competition. He is on the 2021-2022 academic job market.
>
> *This talk is hosted by the ISL Colloquium
> . To receive talk announcements, subscribe
> to the mailing list isl-colloq at lists.stanford.edu
> .*
> ------------------------------
>
> Mailing list: https://mailman.stanford.edu/mailman/listinfo/isl-colloq
> This talk: http://isl.stanford.edu/talks/talks/2021q4/hannah-li-shi-dong/
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 14 11:51:41 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 14 Oct 2021 18:51:41 +0000
Subject: [theory-seminar] Theory Lunch 10/14: Erik Waingarten
In-Reply-To:
References:
Message-ID:
?A gentle reminder: This is happening in 10 minutes.
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 11, 2021 12:35 AM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/14: Erik Waingarten
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Erik will tell us about: Sketches for Clustering in ?pp
Abstract: We give streaming and distributed algorithms for center-based (k,p)-clustering in ?p when p ? [1, 2]. The input is a dataset x1,?, xn ? ?d and ? > 0, and using space poly(log(nd), k, 1/?), we approximate the optimal (k,p)-clustering cost. The above result was known when p = 2 (by utilizing a Johnson-Lindenstrauss projection [Cohen-Elder-Musco-Musco-Persu '15, Makarychev-Makarychev-Razenstheyn '20]); however, for p ? 2, all algorithms incurred space complexity linear in d.
Joint with Moses Charikar.
Lunch menu (both are vegetarian):
Greek salad (with pepper, cucumber, tomato, Kalamata olives and feta cheese),
Caesar salad (Oven roasted garlic croutons and parmesan cheese).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 14 11:51:41 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 14 Oct 2021 18:51:41 +0000
Subject: [theory-seminar] Theory Lunch 10/14: Erik Waingarten
In-Reply-To:
References:
Message-ID:
?A gentle reminder: This is happening in 10 minutes.
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 11, 2021 12:35 AM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/14: Erik Waingarten
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm.
Erik will tell us about: Sketches for Clustering in ?pp
Abstract: We give streaming and distributed algorithms for center-based (k,p)-clustering in ?p when p ? [1, 2]. The input is a dataset x1,?, xn ? ?d and ? > 0, and using space poly(log(nd), k, 1/?), we approximate the optimal (k,p)-clustering cost. The above result was known when p = 2 (by utilizing a Johnson-Lindenstrauss projection [Cohen-Elder-Musco-Musco-Persu '15, Makarychev-Makarychev-Razenstheyn '20]); however, for p ? 2, all algorithms incurred space complexity linear in d.
Joint with Moses Charikar.
Lunch menu (both are vegetarian):
Greek salad (with pepper, cucumber, tomato, Kalamata olives and feta cheese),
Caesar salad (Oven roasted garlic croutons and parmesan cheese).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From gvaliant at cs.stanford.edu Fri Oct 15 09:51:47 2021
From: gvaliant at cs.stanford.edu (Gregory Valiant)
Date: Fri, 15 Oct 2021 09:51:47 -0700
Subject: [theory-seminar] Fwd: MIT Norbert Wiener Postdoctoral Fellowship
In-Reply-To: <4268488C-5F43-4B8C-8A15-3240420F054D@math.mit.edu>
References: <4268488C-5F43-4B8C-8A15-3240420F054D@math.mit.edu>
Message-ID:
For those of you graduating and interested in postdoc opportunities at MIT:
---------- Forwarded message ---------
From: Philippe Rigollet
Date: Fri, Oct 15, 2021 at 9:48 AM
Subject: MIT Norbert Wiener Postdoctoral Fellowship
To: Philippe Rigollet
Hi,
MIT is looking for its next Wiener postdoctoral fellow! I would be very
grateful if you could share this announcement to your students and
networks.
Applications are here:
https://academicjobsonline.org/ajo?action=joblist&id=19967&send=Go
Deadline: December 15, 2021.
Thanks!
Philippe
--
Philippe Rigollet
www-math.mit.edu/~rigollet
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
-------------- next part --------------
A non-text attachment was scrubbed...
Name: IDSS-Stats Postdoc Flyer.pdf
Type: application/pdf
Size: 190529 bytes
Desc: not available
URL:
From tavorb at stanford.edu Mon Oct 18 10:15:58 2021
From: tavorb at stanford.edu (Tavor Baharav)
Date: Mon, 18 Oct 2021 13:15:58 -0400
Subject: [theory-seminar] =?utf-8?q?=22AI_for_clinical_trials_and_clinical?=
=?utf-8?q?_trials_for_AI=22_=E2=80=93_James_Zou_=28Thu=2C_21-Oct_?=
=?utf-8?b?QCA0OjAwcG0p?=
Message-ID:
AI for clinical trials and clinical trials for AIJames Zou ? Professor,
Stanford
Thu, 21-Oct / 4:00pm / Packard 101
Note: this talk will be held in person in Packard 101, and will start at
4pm. The talk will be streamed on Zoom for those who cannot attend:
https://stanford.zoom.us/meeting/register/tJckfuCurzkvEtKKOBvDCrPv3McapgP6HygJ.
Please join us for coffee starting at 3:30pm at the Grove outside Packard.
Abstract
Clinical trials are both the gate-keeper and bottleneck of medicine. They
can be very costly and challenging to conduct. This talk explores how AI
can make trials more efficient and, on the flip side, how to use trials to
evaluate AI rigorously. I will first discuss Trial Pathfinder, a
computational framework that generates synthetic patient cohorts from
medical records to optimize cancer trial designs (Liu et al. Nature 2021).
Trial Pathfinder enables clinical trials to be more inclusive, benefiting
diverse patients and trial sponsors. In the 2nd part, I will discuss
insights that we learned from conducting some of the first trials testing
real-time AIs at Stanford and analyzing data from >100 FDA-approved medical
AIs (Wu et al. Nature Medicine 2021). These analyses raise new technical
questions and approaches to audit ML models and understand why it makes
certain mistakes, which are critical to making ML more trustworthy.
Bio
James Zou is an assistant professor of biomedical data science and, by
courtesy, of CS and EE at Stanford University. He works on making AI more
trustworthy, reliable and fair. He is particularly interested in developing
AI to improve health outcomes, enable biotech discoveries, and make medical
care more accessible. His group develops theoretical foundations and new
algorithms and also deploys these new methods in hospitals and the biotech
industry. James has received a Sloan Fellowship, NSF CAREER Award,
Chan-Zuckerberg Investigator award, faculty awards from Google, Tencent and
Amazon, and multiple best paper awards.
*This talk is hosted by the ISL Colloquium
. To receive talk announcements, subscribe
to the mailing list isl-colloq at lists.stanford.edu
.*
------------------------------
Mailing list: https://mailman.stanford.edu/mailman/listinfo/isl-colloq
This talk: http://isl.stanford.edu/talks/talks/2021q4/james-zou/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Mon Oct 18 23:56:29 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Tue, 19 Oct 2021 06:56:29 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Mon Oct 18 23:56:29 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Tue, 19 Oct 2021 06:56:29 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Wed Oct 20 14:56:12 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Wed, 20 Oct 2021 21:56:12 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
In-Reply-To:
References:
Message-ID:
?Hello everyone,
Because of predicted rain, we'll have theory lunch on gather.town (https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh) tomorrow at noon. Our speaker Sidhanth will also join us and then give the talk at 12:30pm.
Bring your food and come to the lunch!
Cheers,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 18, 2021 11:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Wed Oct 20 14:56:12 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Wed, 20 Oct 2021 21:56:12 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
In-Reply-To:
References:
Message-ID:
?Hello everyone,
Because of predicted rain, we'll have theory lunch on gather.town (https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh) tomorrow at noon. Our speaker Sidhanth will also join us and then give the talk at 12:30pm.
Bring your food and come to the lunch!
Cheers,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 18, 2021 11:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 21 11:52:07 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 21 Oct 2021 18:52:07 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
In-Reply-To:
References:
Message-ID:
A gentle reminder: This is happening in 10 minutes on gather: https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh
Sidhanth is happy to meet with our students/faculty on gather/zoom. Ping him on gather if you're interested.
Best,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Wednesday, October 20, 2021 2:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: Re: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
?Hello everyone,
Because of predicted rain, we'll have theory lunch on gather.town (https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh) tomorrow at noon. Our speaker Sidhanth will also join us and then give the talk at 12:30pm.
Bring your food and come to the lunch!
Cheers,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 18, 2021 11:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 21 11:52:07 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 21 Oct 2021 18:52:07 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
In-Reply-To:
References:
Message-ID:
A gentle reminder: This is happening in 10 minutes on gather: https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh
Sidhanth is happy to meet with our students/faculty on gather/zoom. Ping him on gather if you're interested.
Best,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Wednesday, October 20, 2021 2:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: Re: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
?Hello everyone,
Because of predicted rain, we'll have theory lunch on gather.town (https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh) tomorrow at noon. Our speaker Sidhanth will also join us and then give the talk at 12:30pm.
Bring your food and come to the lunch!
Cheers,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 18, 2021 11:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From tavorb at stanford.edu Thu Oct 21 12:04:32 2021
From: tavorb at stanford.edu (Tavor Baharav)
Date: Thu, 21 Oct 2021 12:04:32 -0700
Subject: [theory-seminar]
=?utf-8?q?=22AI_for_clinical_trials_and_clinical?=
=?utf-8?q?_trials_for_AI=22_=E2=80=93_James_Zou_=28Thu=2C_21-Oct_?=
=?utf-8?b?QCA0OjAwcG0p?=
In-Reply-To:
References:
Message-ID:
A gentle reminder that this talk is today at 4:00pm in Packard 101. Please
join us at 3:30pm at the Grove outside Packard for coffee and snacks.
On Mon, Oct 18, 2021 at 10:15 AM Tavor Baharav wrote:
> AI for clinical trials and clinical trials for AIJames Zou ? Professor,
> Stanford
>
> Thu, 21-Oct / 4:00pm / Packard 101
> Note: this talk will be held in person in Packard 101, and will start at
> 4pm. The talk will be streamed on Zoom for those who cannot attend:
> https://stanford.zoom.us/meeting/register/tJckfuCurzkvEtKKOBvDCrPv3McapgP6HygJ.
> Please join us for coffee starting at 3:30pm at the Grove outside Packard.
> Abstract
>
> Clinical trials are both the gate-keeper and bottleneck of medicine. They
> can be very costly and challenging to conduct. This talk explores how AI
> can make trials more efficient and, on the flip side, how to use trials to
> evaluate AI rigorously. I will first discuss Trial Pathfinder, a
> computational framework that generates synthetic patient cohorts from
> medical records to optimize cancer trial designs (Liu et al. Nature 2021).
> Trial Pathfinder enables clinical trials to be more inclusive, benefiting
> diverse patients and trial sponsors. In the 2nd part, I will discuss
> insights that we learned from conducting some of the first trials testing
> real-time AIs at Stanford and analyzing data from >100 FDA-approved medical
> AIs (Wu et al. Nature Medicine 2021). These analyses raise new technical
> questions and approaches to audit ML models and understand why it makes
> certain mistakes, which are critical to making ML more trustworthy.
> Bio
>
> James Zou is an assistant professor of biomedical data science and, by
> courtesy, of CS and EE at Stanford University. He works on making AI more
> trustworthy, reliable and fair. He is particularly interested in developing
> AI to improve health outcomes, enable biotech discoveries, and make medical
> care more accessible. His group develops theoretical foundations and new
> algorithms and also deploys these new methods in hospitals and the biotech
> industry. James has received a Sloan Fellowship, NSF CAREER Award,
> Chan-Zuckerberg Investigator award, faculty awards from Google, Tencent and
> Amazon, and multiple best paper awards.
>
> *This talk is hosted by the ISL Colloquium
> . To receive talk announcements, subscribe
> to the mailing list isl-colloq at lists.stanford.edu
> .*
> ------------------------------
>
> Mailing list: https://mailman.stanford.edu/mailman/listinfo/isl-colloq
> This talk: http://isl.stanford.edu/talks/talks/2021q4/james-zou/
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 21 12:14:04 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 21 Oct 2021 19:14:04 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
In-Reply-To:
References:
Message-ID:
The password is SongComplexity
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Thursday, October 21, 2021 11:52 AM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: Re: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
A gentle reminder: This is happening in 10 minutes on gather: https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh
Sidhanth is happy to meet with our students/faculty on gather/zoom. Ping him on gather if you're interested.
Best,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Wednesday, October 20, 2021 2:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: Re: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
?Hello everyone,
Because of predicted rain, we'll have theory lunch on gather.town (https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh) tomorrow at noon. Our speaker Sidhanth will also join us and then give the talk at 12:30pm.
Bring your food and come to the lunch!
Cheers,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 18, 2021 11:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 21 12:14:04 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 21 Oct 2021 19:14:04 +0000
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
In-Reply-To:
References:
Message-ID:
The password is SongComplexity
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Thursday, October 21, 2021 11:52 AM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: Re: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
A gentle reminder: This is happening in 10 minutes on gather: https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh
Sidhanth is happy to meet with our students/faculty on gather/zoom. Ping him on gather if you're interested.
Best,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Wednesday, October 20, 2021 2:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: Re: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
?Hello everyone,
Because of predicted rain, we'll have theory lunch on gather.town (https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh) tomorrow at noon. Our speaker Sidhanth will also join us and then give the talk at 12:30pm.
Bring your food and come to the lunch!
Cheers,
Junyao
________________________________
From: theory-seminar on behalf of Junyao Zhao
Sent: Monday, October 18, 2021 11:56 PM
To: thseminar at cs.stanford.edu ; theory-seminar at lists.stanford.edu
Subject: [theory-seminar] Theory Lunch 10/21: Sidhanth Mohanty (Berkeley)
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad.
(However, there's a small probability that Thursday will be rainy. If it rains on Thursday at noon, we'll have theory lunch on gather.town:
https://gather.town/invite?token=CEbVffjYIii5gdTpKGSk5p3M52dLigMh.
I'll send you an announcement if we decide to switch to gather.town.)
As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Sidhanth Mohanty from Berkeley.
Sidhanth will tell us about: Certified counting and random constraint satisfaction problems
Abstract: A random 3SAT instance on n variables and Cn clauses is unsatisfiable with high probability once C is a large enough constant. However, efficient algorithms to certify unsatisfiability of a random 3SAT instance are only known C >> sqrt{n}. Moreover, there are strong lower bounds against algorithms based on spectral and convex relaxation techniques when C << sqrt{n}, suggesting the possibility of an information-computation gap. We investigate whether this information-computation gap prevails if we relax the algorithmic task a bit -- in particular instead of certifying that there are no satisfying assignments, we study the computational complexity of certifying an upper bound on the number of satisfying assignments for values of C << sqrt{n}. In this talk, we will discuss the landscape of when this problem is algorithmically tractable and when there is evidence of intractability, along with some of the algorithmic ideas.
Based on joint work with Jun-Ting Hsieh and Jeff Xu. https://arxiv.org/abs/2106.12710
Lunch menu (if it doesn't rain):
(i) Turkey sandwich,
(ii) Vegetarian sandwich (mozzarella, basil with balsamic vinegar).
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From kabirc at stanford.edu Mon Oct 25 09:26:47 2021
From: kabirc at stanford.edu (Kabir Chandrasekher)
Date: Mon, 25 Oct 2021 09:26:47 -0700
Subject: [theory-seminar] =?utf-8?q?=22On_the_Convergence_of_Langevin_Mont?=
=?utf-8?q?e_Carlo=3A_The_Interplay_between_Tail_Growth_and_Smoothn?=
=?utf-8?b?ZXNzIiDigJMgTXVyYXQgRXJkb2dkdSAoVGh1LCAyOC1PY3QgQCA0OjAw?=
=?utf-8?b?cG0p?=
Message-ID:
On the Convergence of Langevin Monte Carlo: The Interplay between Tail
Growth and Smoothness Murat Erdogdu ? Professor, University of Toronto
Thu, 28-Oct / 4:00pm / Packard 101 (in person)
* Please join us for coffee and snacks at 3:30pm in the Grove outside
Packard (near Bytes' outdoor seating). The talk will be streamed on Zoom
for those unable to attend in person:
https://stanford.zoom.us/meeting/register/tJckfuCurzkvEtKKOBvDCrPv3McapgP6HygJ
*
Abstract
We study sampling from a target distribution $e^{-f}$ using the unadjusted
Langevin Monte Carlo (LMC) algorithm. For any potential function $f$ whose
tails behave like $|x|^\alpha$ for $\alpha \in [1,2]$, and has
$\beta$-H\?older continuous gradient, we derive the sufficient number of
steps to reach the $\eps$-neighborhood of a $d$-dimensional target
distribution as a function of $\alpha$ and $\beta$. Our rate estimate, in
terms of $\eps$ dependency, is not directly influenced by the tail growth
rate $\alpha$ of the potential function as long as its growth is at least
linear, and it only relies on the order of smoothness $\beta$.
Our rate recovers the best known rate which was established for strongly
convex potentials with Lipschitz gradient in terms of $\eps$ dependency,
but we show that the same rate is achievable for a wider class of
potentials that are degenerately convex at infinity.
Bio
Murat is currently an assistant professor at the University of Toronto in
departments of Computer Science and Statistical Sciences. He is also a
faculty member of the Vector Institute, and a CIFAR Chair in AI. Before, he
was a postdoctoral researcher at Microsoft Research - New England lab. His
research interests include optimization, machine learning, statistics,
applied probability, and connections among these fields. He obtained his
Ph.D. from the Department of Statistics at Stanford University. He has an
M.S. degree in Computer Science from Stanford, and B.S. degrees in
Electrical Engineering and Mathematics, both from Bogazici University.
*This talk is hosted by the ISL Colloquium
. To receive talk announcements, subscribe
to the mailing list isl-colloq at lists.stanford.edu
.*
------------------------------
Mailing list: https://mailman.stanford.edu/mailman/listinfo/isl-colloq
This talk: http://isl.stanford.edu/talks/talks/2021q4/murat-erdogdu/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Mon Oct 25 09:51:34 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Mon, 25 Oct 2021 16:51:34 +0000
Subject: [theory-seminar] Theory Lunch 10/28: Guy Bresler (MIT) + Theory
Seminar 11/5: Vijay Vazirani (UC Irvine)
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Guy Bresler. Guy will tell us about: The Algorithmic Phase Transition of Random k-SAT for Low Degree Polynomials
Abstract: We study the algorithmic task of finding a satisfying assignment of a uniformly random k-SAT formula F with n variables and m clauses. It is known that a satisfying assignment exists with high probability at clause density m/n < 2^k log 2 - (1/2) (log 2 + 1) + o_k(1), while the best polynomial-time algorithm known, the Fix algorithm of Coja-Oghlan, finds a satisfying assignment at the much lower clause density (1 - o_k(1)) 2^k log k / k. This prompts the question: is it possible to efficiently find a satisfying assignment at higher clause densities?
To understand the algorithmic threshold of random k-SAT, we study the limits of low degree polynomial algorithms, which are a powerful class of algorithms including Fix, Survey Propagation guided decimation (with bounded or mildly growing number of message passing rounds), and paradigms such as message passing and local graph algorithms. We show that low degree polynomial algorithms can find a satisfying assignment at clause density (1 - o_k(1)) 2^k log k / k, matching Fix, and not at clause density (1 + o_k(1)) c* 2^k log k / k, where c* ~ 4.911. This shows the first sharp (up to constant factor) computational phase transition of random k-SAT for a class of algorithms. Our proof establishes and leverages a new many-way overlap gap property tailored to random k-SAT, which rigorously rules out efficient algorithms via clustering of the solution space. Joint work with Brice Huang.
Looking ahead to next week, because next Thursday is STOC deadline (good luck to those who are submitting), we won't have talk for theory lunch. However, we'll have a special theory seminar at 3pm next Friday (Nov 5) in Y2E2 300.
Vijay Vazirani will give a talk about: Online Bipartite Matching and Adwords
Abstract: Over the last three decades, the online bipartite matching (OBM) problem has emerged as a central problem in the area of Online Algorithms. Perhaps even more important is its role in the area of Matching-Based Market Design. The recent resurgence of this area, with the revolutions of the Internet and mobile computing, has opened up novel, path-breaking applications, and OBM has emerged as its paradigmatic algorithmic problem.
In a 1990 joint paper with Richard Karp and Umesh Vazirani, we gave an optimal algorithm, called RANKING, for OBM, achieving a competitive ratio of (1 ? 1/e); however, its analysis was difficult to comprehend. Over the years, several researchers simplified the analysis. We will start by presenting a ?textbook quality? proof of RANKING. We will then provide a broad overview of the area of Matching-Based Market Design and pinpoint the role of OBM.
The simplicity of the new proof of RANKING raises the possibility of extending it all the way to a generalization of OBM called the adwords problem. This problem is both notoriously difficult and very significant, the latter because it captures a key computational issue arising in Google?s AdWords market. If time remains, we will show how far this endeavor has gone and what remains.
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Mon Oct 25 09:51:34 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Mon, 25 Oct 2021 16:51:34 +0000
Subject: [theory-seminar] Theory Lunch 10/28: Guy Bresler (MIT) + Theory
Seminar 11/5: Vijay Vazirani (UC Irvine)
Message-ID:
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Guy Bresler. Guy will tell us about: The Algorithmic Phase Transition of Random k-SAT for Low Degree Polynomials
Abstract: We study the algorithmic task of finding a satisfying assignment of a uniformly random k-SAT formula F with n variables and m clauses. It is known that a satisfying assignment exists with high probability at clause density m/n < 2^k log 2 - (1/2) (log 2 + 1) + o_k(1), while the best polynomial-time algorithm known, the Fix algorithm of Coja-Oghlan, finds a satisfying assignment at the much lower clause density (1 - o_k(1)) 2^k log k / k. This prompts the question: is it possible to efficiently find a satisfying assignment at higher clause densities?
To understand the algorithmic threshold of random k-SAT, we study the limits of low degree polynomial algorithms, which are a powerful class of algorithms including Fix, Survey Propagation guided decimation (with bounded or mildly growing number of message passing rounds), and paradigms such as message passing and local graph algorithms. We show that low degree polynomial algorithms can find a satisfying assignment at clause density (1 - o_k(1)) 2^k log k / k, matching Fix, and not at clause density (1 + o_k(1)) c* 2^k log k / k, where c* ~ 4.911. This shows the first sharp (up to constant factor) computational phase transition of random k-SAT for a class of algorithms. Our proof establishes and leverages a new many-way overlap gap property tailored to random k-SAT, which rigorously rules out efficient algorithms via clustering of the solution space. Joint work with Brice Huang.
Looking ahead to next week, because next Thursday is STOC deadline (good luck to those who are submitting), we won't have talk for theory lunch. However, we'll have a special theory seminar at 3pm next Friday (Nov 5) in Y2E2 300.
Vijay Vazirani will give a talk about: Online Bipartite Matching and Adwords
Abstract: Over the last three decades, the online bipartite matching (OBM) problem has emerged as a central problem in the area of Online Algorithms. Perhaps even more important is its role in the area of Matching-Based Market Design. The recent resurgence of this area, with the revolutions of the Internet and mobile computing, has opened up novel, path-breaking applications, and OBM has emerged as its paradigmatic algorithmic problem.
In a 1990 joint paper with Richard Karp and Umesh Vazirani, we gave an optimal algorithm, called RANKING, for OBM, achieving a competitive ratio of (1 ? 1/e); however, its analysis was difficult to comprehend. Over the years, several researchers simplified the analysis. We will start by presenting a ?textbook quality? proof of RANKING. We will then provide a broad overview of the area of Matching-Based Market Design and pinpoint the role of OBM.
The simplicity of the new proof of RANKING raises the possibility of extending it all the way to a generalization of OBM called the adwords problem. This problem is both notoriously difficult and very significant, the latter because it captures a key computational issue arising in Google?s AdWords market. If time remains, we will show how far this endeavor has gone and what remains.
Cheers,
Junyao
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From aviad at cs.stanford.edu Mon Oct 25 10:54:34 2021
From: aviad at cs.stanford.edu (Aviad Rubinstein)
Date: Mon, 25 Oct 2021 10:54:34 -0700
Subject: [theory-seminar] Theory Lunch 10/28: Guy Bresler (MIT) + Theory
Seminar 11/5: Vijay Vazirani (UC Irvine)
In-Reply-To:
References:
Message-ID:
Hi theorists,
Quick update about next week- Vijay's talk will be postponed to a later
date (TBD).
See you at Guy's talk this Thursday and good luck with the STOC deadline ;)
Aviad
On Mon, Oct 25, 2021 at 10:01 AM Junyao Zhao wrote:
> Hi everyone,
>
> This week's theory lunch will take place Thursday at noon in the Engineering
> Quad
> .
> As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our
> speaker this week is Guy Bresler. Guy will tell us about: *The
> Algorithmic Phase Transition of Random k-SAT for Low Degree Polynomials*
>
> *Abstract:* We study the algorithmic task of finding a satisfying
> assignment of a uniformly random k-SAT formula F with n variables and m
> clauses. It is known that a satisfying assignment exists with high
> probability at clause density m/n < 2^k log 2 - (1/2) (log 2 + 1) + o_k(1),
> while the best polynomial-time algorithm known, the Fix algorithm of
> Coja-Oghlan, finds a satisfying assignment at the much lower clause density
> (1 - o_k(1)) 2^k log k / k. This prompts the question: is it possible to
> efficiently find a satisfying assignment at higher clause densities?
>
> To understand the algorithmic threshold of random k-SAT, we study the
> limits of low degree polynomial algorithms, which are a powerful class of
> algorithms including Fix, Survey Propagation guided decimation (with
> bounded or mildly growing number of message passing rounds), and paradigms
> such as message passing and local graph algorithms. We show that low degree
> polynomial algorithms can find a satisfying assignment at clause density (1
> - o_k(1)) 2^k log k / k, matching Fix, and not at clause density (1 +
> o_k(1)) c* 2^k log k / k, where c* ~ 4.911. This shows the first sharp (up
> to constant factor) computational phase transition of random k-SAT for a
> class of algorithms. Our proof establishes and leverages a new many-way
> overlap gap property tailored to random k-SAT, which rigorously rules out
> efficient algorithms via clustering of the solution space. Joint work with
> Brice Huang.
>
> Looking ahead to next week, because next Thursday is STOC deadline (good
> luck to those who are submitting), we won't have talk for theory lunch.
> However, we'll have a special theory seminar at 3pm next Friday (Nov 5) in Y2E2
> 300
>
> .
> Vijay Vazirani will give a talk about: *Online Bipartite Matching and
> Adwords*
>
> *Abstract:* Over the last three decades, the online bipartite matching
> (OBM) problem has emerged as a central problem in the area of Online
> Algorithms. Perhaps even more important is its role in the area of
> Matching-Based Market Design. The recent resurgence of this area, with the
> revolutions of the Internet and mobile computing, has opened up novel,
> path-breaking applications, and OBM has emerged as its paradigmatic
> algorithmic problem.
>
> In a 1990 joint paper with Richard Karp and Umesh Vazirani, we gave an
> optimal algorithm, called RANKING, for OBM, achieving a competitive ratio
> of (1 ? 1/e); however, its analysis was difficult to comprehend. Over the
> years, several researchers simplified the analysis. We will start by
> presenting a ?textbook quality? proof of RANKING. We will then provide a
> broad overview of the area of Matching-Based Market Design and pinpoint the
> role of OBM.
>
> The simplicity of the new proof of RANKING raises the possibility of
> extending it all the way to a generalization of OBM called the adwords
> problem. This problem is both notoriously difficult and very significant,
> the latter because it captures a key computational issue arising in
> Google?s AdWords market. If time remains, we will show how far this
> endeavor has gone and what remains.
>
> Cheers,
> Junyao
>
> _______________________________________________
> theory-seminar mailing list
> theory-seminar at lists.stanford.edu
> https://mailman.stanford.edu/mailman/listinfo/theory-seminar
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From aviad at cs.stanford.edu Mon Oct 25 10:54:34 2021
From: aviad at cs.stanford.edu (Aviad Rubinstein)
Date: Mon, 25 Oct 2021 10:54:34 -0700
Subject: [theory-seminar] Theory Lunch 10/28: Guy Bresler (MIT) + Theory
Seminar 11/5: Vijay Vazirani (UC Irvine)
In-Reply-To:
References:
Message-ID:
Hi theorists,
Quick update about next week- Vijay's talk will be postponed to a later
date (TBD).
See you at Guy's talk this Thursday and good luck with the STOC deadline ;)
Aviad
On Mon, Oct 25, 2021 at 10:01 AM Junyao Zhao wrote:
> Hi everyone,
>
> This week's theory lunch will take place Thursday at noon in the Engineering
> Quad
> .
> As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our
> speaker this week is Guy Bresler. Guy will tell us about: *The
> Algorithmic Phase Transition of Random k-SAT for Low Degree Polynomials*
>
> *Abstract:* We study the algorithmic task of finding a satisfying
> assignment of a uniformly random k-SAT formula F with n variables and m
> clauses. It is known that a satisfying assignment exists with high
> probability at clause density m/n < 2^k log 2 - (1/2) (log 2 + 1) + o_k(1),
> while the best polynomial-time algorithm known, the Fix algorithm of
> Coja-Oghlan, finds a satisfying assignment at the much lower clause density
> (1 - o_k(1)) 2^k log k / k. This prompts the question: is it possible to
> efficiently find a satisfying assignment at higher clause densities?
>
> To understand the algorithmic threshold of random k-SAT, we study the
> limits of low degree polynomial algorithms, which are a powerful class of
> algorithms including Fix, Survey Propagation guided decimation (with
> bounded or mildly growing number of message passing rounds), and paradigms
> such as message passing and local graph algorithms. We show that low degree
> polynomial algorithms can find a satisfying assignment at clause density (1
> - o_k(1)) 2^k log k / k, matching Fix, and not at clause density (1 +
> o_k(1)) c* 2^k log k / k, where c* ~ 4.911. This shows the first sharp (up
> to constant factor) computational phase transition of random k-SAT for a
> class of algorithms. Our proof establishes and leverages a new many-way
> overlap gap property tailored to random k-SAT, which rigorously rules out
> efficient algorithms via clustering of the solution space. Joint work with
> Brice Huang.
>
> Looking ahead to next week, because next Thursday is STOC deadline (good
> luck to those who are submitting), we won't have talk for theory lunch.
> However, we'll have a special theory seminar at 3pm next Friday (Nov 5) in Y2E2
> 300
>
> .
> Vijay Vazirani will give a talk about: *Online Bipartite Matching and
> Adwords*
>
> *Abstract:* Over the last three decades, the online bipartite matching
> (OBM) problem has emerged as a central problem in the area of Online
> Algorithms. Perhaps even more important is its role in the area of
> Matching-Based Market Design. The recent resurgence of this area, with the
> revolutions of the Internet and mobile computing, has opened up novel,
> path-breaking applications, and OBM has emerged as its paradigmatic
> algorithmic problem.
>
> In a 1990 joint paper with Richard Karp and Umesh Vazirani, we gave an
> optimal algorithm, called RANKING, for OBM, achieving a competitive ratio
> of (1 ? 1/e); however, its analysis was difficult to comprehend. Over the
> years, several researchers simplified the analysis. We will start by
> presenting a ?textbook quality? proof of RANKING. We will then provide a
> broad overview of the area of Matching-Based Market Design and pinpoint the
> role of OBM.
>
> The simplicity of the new proof of RANKING raises the possibility of
> extending it all the way to a generalization of OBM called the adwords
> problem. This problem is both notoriously difficult and very significant,
> the latter because it captures a key computational issue arising in
> Google?s AdWords market. If time remains, we will show how far this
> endeavor has gone and what remains.
>
> Cheers,
> Junyao
>
> _______________________________________________
> theory-seminar mailing list
> theory-seminar at lists.stanford.edu
> https://mailman.stanford.edu/mailman/listinfo/theory-seminar
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From tpulkit at stanford.edu Tue Oct 26 12:21:56 2021
From: tpulkit at stanford.edu (Pulkit Tandon)
Date: Tue, 26 Oct 2021 19:21:56 +0000
Subject: [theory-seminar] "Universal Probability and Its Applications" -
Alankrita Bhatt (Friday, October 29th, 1pm)
Message-ID:
Hi everyone,
We will be restarting the Information Theory Forum (IT Forum) talks. The first talk of the quarter will be organized this Friday (29th October), 1pm, accessible via Zoom. Details below:
Universal Probability and Its Applications
Alankrita Bhatt, UCSD
Fri, 29th October, 1pm
Zoom Link: https://stanford.zoom.us/j/98545179136?pwd=dUhOQkN1YnpJWEtGRFlUS3hFRnNZdz09
pwd: 257808
Abstract:
In this talk, I will present the universal probability approach to a statistical task. This approach gives us general principles and guidelines for assigning sequential probabilities to data (based on which a statistical decision can then be made), and has been used successfully over the years to problems in compression and estimation among others. I will illustrate this approach via two example applications?sequential prediction and portfolio selection with side information.
Bio:
Alankrita Bhatt is a PhD student in the department of electrical and computer engineering, at the University of California San Diego. Her research interests lie broadly in the field of information theory. Prior to joining UCSD, she received a bachelor's degree in electrical engineering from the Indian Institute of Technology Kanpur.
Best
Pulkit Tandon
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From marykw at stanford.edu Wed Oct 27 11:27:23 2021
From: marykw at stanford.edu (Mary Wootters)
Date: Wed, 27 Oct 2021 11:27:23 -0700
Subject: [theory-seminar] Faculty openings at Buffalo
Message-ID:
Hi graduating Ph.D. students / post-docs,
Atri Rudra asked me to advertise that Buffalo is hiring in theory this
year, looking for candidates in both algorithms and complexity. See the
job ad here:
https://engineering.buffalo.edu/computer-science-engineering/news-and-events/employment-opportunities.html#assistant-associate-or-full-professor
Best,
Mary
--
Mary Wootters (she/her)
Assistant Professor of Computer Science and Electrical Engineering
Stanford University
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From kabirc at stanford.edu Thu Oct 28 08:36:12 2021
From: kabirc at stanford.edu (Kabir Chandrasekher)
Date: Thu, 28 Oct 2021 08:36:12 -0700
Subject: [theory-seminar]
=?utf-8?q?=22On_the_Convergence_of_Langevin_Mont?=
=?utf-8?q?e_Carlo=3A_The_Interplay_between_Tail_Growth_and_Smoothn?=
=?utf-8?b?ZXNzIiDigJMgTXVyYXQgRXJkb2dkdSAoVGh1LCAyOC1PY3QgQCA0OjAw?=
=?utf-8?b?cG0p?=
In-Reply-To:
References:
Message-ID:
Reminder that this talk will be today at 4pm in Packard 101 (and streamed
via Zoom:
https://stanford.zoom.us/meeting/register/tJckfuCurzkvEtKKOBvDCrPv3McapgP6HygJ).
Please join us for coffee and snacks at 3:30pm in the Grove outside Packard
(near Bytes' outdoor seating).
On Mon, Oct 25, 2021 at 9:26 AM Kabir Chandrasekher
wrote:
> On the Convergence of Langevin Monte Carlo: The Interplay between Tail
> Growth and Smoothness Murat Erdogdu ? Professor, University of Toronto
>
> Thu, 28-Oct / 4:00pm / Packard 101 (in person)
>
> * Please join us for coffee and snacks at 3:30pm in the Grove outside
> Packard (near Bytes' outdoor seating). The talk will be streamed on Zoom
> for those unable to attend in person:
> https://stanford.zoom.us/meeting/register/tJckfuCurzkvEtKKOBvDCrPv3McapgP6HygJ
>
> *
> Abstract
>
> We study sampling from a target distribution $e^{-f}$ using the unadjusted
> Langevin Monte Carlo (LMC) algorithm. For any potential function $f$ whose
> tails behave like $|x|^\alpha$ for $\alpha \in [1,2]$, and has
> $\beta$-H\?older continuous gradient, we derive the sufficient number of
> steps to reach the $\eps$-neighborhood of a $d$-dimensional target
> distribution as a function of $\alpha$ and $\beta$. Our rate estimate, in
> terms of $\eps$ dependency, is not directly influenced by the tail growth
> rate $\alpha$ of the potential function as long as its growth is at least
> linear, and it only relies on the order of smoothness $\beta$.
>
> Our rate recovers the best known rate which was established for strongly
> convex potentials with Lipschitz gradient in terms of $\eps$ dependency,
> but we show that the same rate is achievable for a wider class of
> potentials that are degenerately convex at infinity.
> Bio
>
> Murat is currently an assistant professor at the University of Toronto in
> departments of Computer Science and Statistical Sciences. He is also a
> faculty member of the Vector Institute, and a CIFAR Chair in AI. Before, he
> was a postdoctoral researcher at Microsoft Research - New England lab. His
> research interests include optimization, machine learning, statistics,
> applied probability, and connections among these fields. He obtained his
> Ph.D. from the Department of Statistics at Stanford University. He has an
> M.S. degree in Computer Science from Stanford, and B.S. degrees in
> Electrical Engineering and Mathematics, both from Bogazici University.
>
> *This talk is hosted by the ISL Colloquium
> . To receive talk announcements, subscribe
> to the mailing list isl-colloq at lists.stanford.edu
> .*
> ------------------------------
>
> Mailing list: https://mailman.stanford.edu/mailman/listinfo/isl-colloq
> This talk: http://isl.stanford.edu/talks/talks/2021q4/murat-erdogdu/
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 28 11:50:31 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 28 Oct 2021 18:50:31 +0000
Subject: [theory-seminar] Theory Lunch 10/28: Guy Bresler (MIT)
Message-ID:
?A gentle reminder: This is happening in 10 minutes.
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Guy Bresler. Guy will tell us about: The Algorithmic Phase Transition of Random k-SAT for Low Degree Polynomials
Abstract: We study the algorithmic task of finding a satisfying assignment of a uniformly random k-SAT formula F with n variables and m clauses. It is known that a satisfying assignment exists with high probability at clause density m/n < 2^k log 2 - (1/2) (log 2 + 1) + o_k(1), while the best polynomial-time algorithm known, the Fix algorithm of Coja-Oghlan, finds a satisfying assignment at the much lower clause density (1 - o_k(1)) 2^k log k / k. This prompts the question: is it possible to efficiently find a satisfying assignment at higher clause densities?
To understand the algorithmic threshold of random k-SAT, we study the limits of low degree polynomial algorithms, which are a powerful class of algorithms including Fix, Survey Propagation guided decimation (with bounded or mildly growing number of message passing rounds), and paradigms such as message passing and local graph algorithms. We show that low degree polynomial algorithms can find a satisfying assignment at clause density (1 - o_k(1)) 2^k log k / k, matching Fix, and not at clause density (1 + o_k(1)) c* 2^k log k / k, where c* ~ 4.911. This shows the first sharp (up to constant factor) computational phase transition of random k-SAT for a class of algorithms. Our proof establishes and leverages a new many-way overlap gap property tailored to random k-SAT, which rigorously rules out efficient algorithms via clustering of the solution space. Joint work with Brice Huang.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From junyaoz at stanford.edu Thu Oct 28 11:50:31 2021
From: junyaoz at stanford.edu (Junyao Zhao)
Date: Thu, 28 Oct 2021 18:50:31 +0000
Subject: [theory-seminar] Theory Lunch 10/28: Guy Bresler (MIT)
Message-ID:
?A gentle reminder: This is happening in 10 minutes.
Hi everyone,
This week's theory lunch will take place Thursday at noon in the Engineering Quad. As usual, we'll start with some socializing, followed by a talk at 12:30pm. Our speaker this week is Guy Bresler. Guy will tell us about: The Algorithmic Phase Transition of Random k-SAT for Low Degree Polynomials
Abstract: We study the algorithmic task of finding a satisfying assignment of a uniformly random k-SAT formula F with n variables and m clauses. It is known that a satisfying assignment exists with high probability at clause density m/n < 2^k log 2 - (1/2) (log 2 + 1) + o_k(1), while the best polynomial-time algorithm known, the Fix algorithm of Coja-Oghlan, finds a satisfying assignment at the much lower clause density (1 - o_k(1)) 2^k log k / k. This prompts the question: is it possible to efficiently find a satisfying assignment at higher clause densities?
To understand the algorithmic threshold of random k-SAT, we study the limits of low degree polynomial algorithms, which are a powerful class of algorithms including Fix, Survey Propagation guided decimation (with bounded or mildly growing number of message passing rounds), and paradigms such as message passing and local graph algorithms. We show that low degree polynomial algorithms can find a satisfying assignment at clause density (1 - o_k(1)) 2^k log k / k, matching Fix, and not at clause density (1 + o_k(1)) c* 2^k log k / k, where c* ~ 4.911. This shows the first sharp (up to constant factor) computational phase transition of random k-SAT for a class of algorithms. Our proof establishes and leverages a new many-way overlap gap property tailored to random k-SAT, which rigorously rules out efficient algorithms via clustering of the solution space. Joint work with Brice Huang.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From marykw at stanford.edu Sun Oct 31 07:35:06 2021
From: marykw at stanford.edu (Mary Wootters)
Date: Sun, 31 Oct 2021 09:35:06 -0500
Subject: [theory-seminar] CA Position: CS250/EE387, apply through EE,
deadline Nov. 7.
In-Reply-To:
References:
Message-ID:
Hi all,
I'm teaching CS250/EE387 (Algebraic Error Correcting Codes) in the Winter.
If you'd be interested in CA'ing for the course, please apply! Please
apply here by Nov. 7:
https://ee.stanford.edu/student-resources/CA-appointments/how-to-apply.
(More information about applying is below).
Even if you haven't taken the course before, if you have some experience
with error correcting codes -- or a lot of theory/math maturity and are
willing to learn a bunch of stuff quickly -- then consider applying.
*The deadline to apply is this Friday, November 7. * (Note that since this
course is managed by the EE department, the timeline is a bit different
than CS CA-ships.)
Please let me know if you have any questions about the course or about the
CA position!
Best,
Mary
---------- Forwarded message ---------
From: TA Administrator
Date: Mon, Oct 25, 2021 at 2:01 PM
Subject: Winter Quarter CA and Grader Application 2021-2022_NOW OPEN
To: , ,
, ,
The EE department is now accepting applications for *Winter Quarter
Course Assistant and Grader* positions.
The application deadline is 11:59pm on *Sunday, November 7*.
The *Winter* CA and Grader application can be found here:
https://ee.stanford.edu/student-resources/CA-appointments/how-to-apply
In order to be considered for an appointment, please complete the
application by answering all of the questions to the best of your knowledge
and ability. Additionally, please select all of the courses for which you
would feel comfortable and confident in CAing or grading. Once you submit
your application, you will be able to update or edit the information provide up
until the application deadline.
Students will be notified via email in *early-mid December* if they
receive an appointment.
*Note*: Anyone wishing to be considered for a CA or grader position
MUST submit an application (this includes current CAs). Please do not
assume that you will have an appointment just because an instructor
has told you that you are his/her choice.
Only students who have submitted an application will be considered.
Students who expect to hold teaching assistantships must be screened
for English Proficiency if they are from countries where English is
not the first language. Some students may be eligible for a waiver.
To be screened, please visit TA Screening
.
The screening consists of a simulated office-hour conversation (about
30 minutes) between a CA/section leader and a undergraduate student.
The CA/section leader being tested does not need to prepare for the
screening. No exceptions are allowed for this requirement. Failure
to comply may result in the delay or loss of an appointment.
If you have any questions or concerns, please contact us at
ta at ee.stanford.edu.
To view more information about Course Assistantships, please visit
EE's main website at:
http://ee.stanford.edu/ ? Student Resources tab ? Student Employment.
Good luck and thank you again for applying!
Best wishes,
EE Student Services
--
Mary Wootters (she/her)
Assistant Professor of Computer Science and Electrical Engineering
Stanford University
-------------- next part --------------
An HTML attachment was scrubbed...
URL: