From silas at stanford.edu Mon Oct 2 10:53:38 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Mon, 2 Oct 2017 17:53:38 +0000
Subject: [theory-seminar] Theory Seminar on Thursday: Rad Niazadeh
Message-ID:
Hi all,
There will be a theory seminar this Thursday at 4:15pm at Gates 463A. The talk will be given by Rad Niazadeh. Hope you'll make it!
Bernoulli Factories and Blackbox Reductions in Mechanism Design
In this talk, I am going to talk about a recent polynomial-time reduction from Bayesian incentive-compatible mechanism design to Bayesian algorithm design for welfare maximization problems. Unlike prior results, our reduction achieves exact incentive compatibility for problems with multi-dimensional and continuous type spaces. The key technical barrier preventing exact incentive compatibility in prior black-box reductions is that repairing violations of incentive constraints requires understanding the distribution of the mechanism?s output, which is typically #P-hard to compute. Reductions that instead estimate the output distribution by sampling inevitably suffer from sampling error, which typically precludes exact incentive compatibility. We overcome this barrier by employing and generalizing the computational model in the literature on Bernoulli Factories. In a Bernoulli factory problem, one is given a function mapping the bias of an ?input coin? to that of an ?output coin?, and the challenge is to efficiently simulate the output coin given only sample access to the input coin. I will show how to incorporate Bernoulli factories to design a polynomial time algorithm for the following selection problem: a ground set of elements and sampling oracles for each element are given (for unknown input distributions), expected values of the input distributions correspond to the weights of the elements in the set, and we wish to select an element with probability proportional to an exponential function of its weight by efficient sampling. I then show how this algorithm solves a simple single agent truthful mechanism design problem. This truthful mechanism is the key ingredient that can be used to make the approximately incentive compatible reduction of Hartline et al. (2015) exactly incentive compatible.
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang90 at gmail.com Wed Oct 4 12:30:51 2017
From: hongyang90 at gmail.com (Hongyang Zhang)
Date: Wed, 4 Oct 2017 12:30:51 -0700
Subject: [theory-seminar] [theory-lunch] Don Knuth --- A clean analysis of
contention resolution
Message-ID:
Hi Everyone
This Thursday *12:00pm to 1:00pm (Gates 463A)* we are proud to have Don
Knuth as our speaker.
Don will tell us about A clean analysis of contention resolution, which is
based on a nice example in the book by Kleinberg/Tardos.
Best
-hongyang
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Thu Oct 5 10:52:41 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Thu, 5 Oct 2017 17:52:41 +0000
Subject: [theory-seminar] Theory Seminar Today ay 415
Message-ID:
Hi all,
Just a reminder that there will be theory seminar at 4:15pm today at Gates 463A. The talk will be given by Rad Niazadeh and we will have tea and cake. Hope you'll make it!
Bernoulli Factories and Blackbox Reductions in Mechanism Design
In this talk, I am going to talk about a recent polynomial-time reduction from Bayesian incentive-compatible mechanism design to Bayesian algorithm design for welfare maximization problems. Unlike prior results, our reduction achieves exact incentive compatibility for problems with multi-dimensional and continuous type spaces. The key technical barrier preventing exact incentive compatibility in prior black-box reductions is that repairing violations of incentive constraints requires understanding the distribution of the mechanism?s output, which is typically #P-hard to compute. Reductions that instead estimate the output distribution by sampling inevitably suffer from sampling error, which typically precludes exact incentive compatibility. We overcome this barrier by employing and generalizing the computational model in the literature on Bernoulli Factories. In a Bernoulli factory problem, one is given a function mapping the bias of an ?input coin? to that of an ?output coin?, and the challenge is to efficiently simulate the output coin given only sample access to the input coin. I will show how to incorporate Bernoulli factories to design a polynomial time algorithm for the following selection problem: a ground set of elements and sampling oracles for each element are given (for unknown input distributions), expected values of the input distributions correspond to the weights of the elements in the set, and we wish to select an element with probability proportional to an exponential function of its weight by efficient sampling. I then show how this algorithm solves a simple single agent truthful mechanism design problem. This truthful mechanism is the key ingredient that can be used to make the approximately incentive compatible reduction of Hartline et al. (2015) exactly incentive compatible.
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Thu Oct 5 16:08:34 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Thu, 5 Oct 2017 23:08:34 +0000
Subject: [theory-seminar] Theory Seminar is happening in 5 minutes!
Message-ID:
Hi all,
Just a reminder that theory seminar is happening in 5 minutes at Gates 463A.
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Fri Oct 6 13:55:38 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Fri, 6 Oct 2017 13:55:38 -0700
Subject: [theory-seminar] [theory-quals] Oct. 11th 2:30pm on Non-convex
methods
Message-ID:
Hi Everyone
I'm presenting my quals presentation next Wednesday *Oct. 11th at 2:30pm -
3:30pm*, at Gates 498.
It will be a survey talk on Non-convex methods in recovering low-rank
matrices and tensors. The survey is based on several recent papers (details
below). If you are interested in this topic, please feel free to drop by.
?
Quals on Non-Convex Optimization
?
-hongyang
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Mon Oct 9 11:00:33 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Mon, 9 Oct 2017 18:00:33 +0000
Subject: [theory-seminar] Theory Seminar on Thursday
Message-ID:
Hi all,
There will be a theory seminar this Thursday at 4:15pm at Gates 463A. The talk will be given by Thatchapol Saranurak. Hope to see you there!
Thatchaphol Saranurak (KTH)
Dynamic Spanning Forest: Techniques and Connections to Other Fields
I will first give an overview of dynamic algorithms and their connections to other fields. Then, I will present our recent progress on the question "is there a dynamic algorithm with small worst-case update time" for the spanning forest problem, which is among central problems in dynamic algorithms on graphs. Our result guarantees an n^{o(1)} worst-case update time with high probability, where n is the number of nodes. The best worst-case bounds prior to our work are (i) the long-standing O(\sqrt{n}) bound of [Frederickson STOC'83, Eppstein, Galil, Italiano and Nissenzweig FOCS'92] (which is slightly improved by a O(\sqrt{\log(n)}) factor by [Kejlberg-Rasmussen, Kopelowitz, Pettie, Thorup ESA'16]) and (ii) the polylogarithmic bound of [Kapron, King and Mountjoy SODA'13] which works under an oblivious adversary assumption (our result does not make such assumption).
The crucial techniques are about expanders: 1) an algorithm for decomposing a graph into a collection of expanders in near-linear time, and 2) an algorithm for "repairing" the expansion property of an expander after deleting some edges of it. These techniques can be of independent interest.
This talk is based on results by [Nanongkai, Saranurak and Wulff-Nilsen, FOCS'17], [Nanongkai and Saranurak, STOC'17] and [Wulff-Nilsen, STOC'17].
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Tue Oct 10 15:22:40 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Tue, 10 Oct 2017 15:22:40 -0700
Subject: [theory-seminar] [theory-lunch] Jacob Steinhardt on Certifying 2->q
Norm
Message-ID:
Hi Everyone
This *thursday 12:00pm to 1:00pm, at Gates 463A*, Jacob Steinhardt will
tell us about "Certifying 2-q Norms with Sum-of-Squares". See below for the
abstract.
As usual, the talk starts at 12:30pm. See you all!
-hongyang
------------------------------------------------------------
Title: Certifying 2->q Norm with Sum-of-Squares
Abstract: Approximating the 2->q norm is a natural geometric problem which
is closely related to the unique games conjecture. BBHKSZ12 show that the
sum-of-squares (SOS) hierarchy is able to approximate the 2->q norm well on
points sampled from product distributions on the hypercube, but the general
performance of SOS on 2->q norm is not well-understood. I will show that
for distributions satisfying the *Poincare inequality*, m rounds of
sum-of-squares are sufficient to approximate the 2->m norm for all even m
>= 2. This substantially extends the result of BBHKSZ12. As applications we
present improved algorithms for clustering and robust mean estimation. The
proof exploits recent advances in geometric probability by Adamczak and
Wolff.
This is joint work with Pravesh Kothari.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Wed Oct 11 11:53:22 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Wed, 11 Oct 2017 11:53:22 -0700
Subject: [theory-seminar] [theory-quals] Oct. 11th 2:30pm on Non-convex
methods
In-Reply-To:
References:
Message-ID:
Reminder that this is today at 2:30pm - 3:30pm, Gates 498.
-h
On Fri, Oct 6, 2017 at 1:55 PM, Hongyang Zhang
wrote:
> Hi Everyone
>
> I'm presenting my quals presentation next Wednesday *Oct. 11th at 2:30pm
> - 3:30pm*, at Gates 498.
>
> It will be a survey talk on Non-convex methods in recovering low-rank
> matrices and tensors. The survey is based on several recent papers (details
> below). If you are interested in this topic, please feel free to drop by.
> ?
> Quals on Non-Convex Optimization
>
> ?
> -hongyang
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Thu Oct 12 10:16:21 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Thu, 12 Oct 2017 17:16:21 +0000
Subject: [theory-seminar] [theory-lunch] Jacob Steinhardt on Certifying
2->q Norm
In-Reply-To:
References:
Message-ID:
Reminder this is today at noon.
-h
On Tue, 10 Oct 2017 at 3:22 PM Hongyang Zhang
wrote:
> Hi Everyone
>
> This *thursday 12:00pm to 1:00pm, at Gates 463A*, Jacob Steinhardt will
> tell us about "Certifying 2-q Norms with Sum-of-Squares". See below for the
> abstract.
>
> As usual, the talk starts at 12:30pm. See you all!
>
> -hongyang
>
> ------------------------------------------------------------
>
> Title: Certifying 2->q Norm with Sum-of-Squares
>
> Abstract: Approximating the 2->q norm is a natural geometric problem which
> is closely related to the unique games conjecture. BBHKSZ12 show that the
> sum-of-squares (SOS) hierarchy is able to approximate the 2->q norm well on
> points sampled from product distributions on the hypercube, but the general
> performance of SOS on 2->q norm is not well-understood. I will show that
> for distributions satisfying the *Poincare inequality*, m rounds of
> sum-of-squares are sufficient to approximate the 2->m norm for all even m
> >= 2. This substantially extends the result of BBHKSZ12. As applications we
> present improved algorithms for clustering and robust mean estimation. The
> proof exploits recent advances in geometric probability by Adamczak and
> Wolff.
>
> This is joint work with Pravesh Kothari.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Thu Oct 12 11:52:59 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Thu, 12 Oct 2017 18:52:59 +0000
Subject: [theory-seminar] Theory Seminar Today
Message-ID:
Hi all,
Just a reminder that theory seminar will happen today at 415pm at Gates 463A. Hope to see you all there!
Thatchaphol Saranurak (KTH)
Dynamic Spanning Forest: Techniques and Connections to Other Fields
I will first give an overview of dynamic algorithms and their connections to other fields. Then, I will present our recent progress on the question "is there a dynamic algorithm with small worst-case update time" for the spanning forest problem, which is among central problems in dynamic algorithms on graphs. Our result guarantees an n^{o(1)} worst-case update time with high probability, where n is the number of nodes. The best worst-case bounds prior to our work are (i) the long-standing O(\sqrt{n}) bound of [Frederickson STOC'83, Eppstein, Galil, Italiano and Nissenzweig FOCS'92] (which is slightly improved by a O(\sqrt{\log(n)}) factor by [Kejlberg-Rasmussen, Kopelowitz, Pettie, Thorup ESA'16]) and (ii) the polylogarithmic bound of [Kapron, King and Mountjoy SODA'13] which works under an oblivious adversary assumption (our result does not make such assumption).
The crucial techniques are about expanders: 1) an algorithm for decomposing a graph into a collection of expanders in near-linear time, and 2) an algorithm for "repairing" the expansion property of an expander after deleting some edges of it. These techniques can be of independent interest.
This talk is based on results by [Nanongkai, Saranurak and Wulff-Nilsen, FOCS'17], [Nanongkai and Saranurak, STOC'17] and [Wulff-Nilsen, STOC'17].
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Thu Oct 12 13:51:27 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Thu, 12 Oct 2017 13:51:27 -0700
Subject: [theory-seminar] [focs-carpool] Coordinate via spreadsheet
Message-ID:
Hi Everyone
If you are planning to attend FOCS'17 and would like to carpool, please
feel free to put your information in the spreadsheet.
?
FOCS 2017 Carpool
?
-hongyang
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From psimin at stanford.edu Fri Oct 13 12:07:01 2017
From: psimin at stanford.edu (Paris Syminelakis)
Date: Fri, 13 Oct 2017 19:07:01 +0000
Subject: [theory-seminar] FOCS Practice talk Today 1:00pm - 1:30pm @ Gates
463A
Message-ID:
Hi all,
I am going to be giving a FOCS practice talk today between 1:00pm - 1:30pm at Gates 463A and would be great to get some feedback. Everyone is welcome! Abstract follows.
--Paris
Title: Hashing-Based-Estimators for Kernel Density in High Dimensions.
Abstract: Given a set of points $P\subset \R^{d}$ and a kernel $k$, the Kernel Density Estimate at a point $x\in\R^{d}$ is defined as $\mathrm{KDE}_{P}(x)=\frac{1}{|P|}\sum_{y\in P} k(x,y)$. We study the problem of designing a data structure that given a data set $P$ and a kernel function, returns \emph{approximations to the kernel density} of a query point in \emph{sublinear time}. We introduce a class of unbiased estimators for kernel density implemented through locality-sensitive hashing, and give general theorems bounding the variance of such estimators.
These estimators give rise to efficient data structures for estimating the kernel density in high dimensions for a variety of commonly used kernels. Our work is the first to provide data-structures with theoretical guarantees that improve upon simple random sampling in high dimensions.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Tue Oct 17 13:46:42 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Tue, 17 Oct 2017 20:46:42 +0000
Subject: [theory-seminar] Theory Seminar on Thursday
Message-ID:
Hi All,
There will be a theory seminar on Thursday at 415pm in Gates 463A. Hope to see you there!
Rasmus Kyng (Simons)
Approximate Gaussian Elimination for Laplacians
We show how to perform sparse approximate Gaussian elimination for Laplacian matrices. We present a simple, nearly linear time algorithm that approximates a Laplacian matrix by a matrix with a sparse LU factorization. We compute this factorization by subsampling standard Gaussian elimination. This gives the simplest known nearly linear time solver for Laplacian equations. It is a significant step towards practical and provably correct algorithms for this problem. The crux of our proof is the use of matrix martingales to analyze the algorithm.
Joint work with Sushant Sachdeva.
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From ofirgeri at stanford.edu Tue Oct 17 20:00:50 2017
From: ofirgeri at stanford.edu (Ofir Geri)
Date: Wed, 18 Oct 2017 03:00:50 +0000
Subject: [theory-seminar] Fw: [RAIN] RAIN Seminar Wednesday 10/18 12-1 PM at
Y2E2 101: Michal Feldman on Prophet Inequalities
In-Reply-To:
References: ,
Message-ID:
A talk of interest tomorrow at RAIN:
Michal Feldman from Tel-Aviv University is giving a talk on a cool new result regarding prophet inequalities from this week's FOCS.
---------- Forwarded message ----------
From: Anilesh K. Krishnaswamy >
Date: Tue, Oct 17, 2017 at 2:30 AM
Subject: [RAIN] RAIN Seminar Wednesday 10/18 12-1 PM at Y2E2 101
To: internetalgs at lists.stanford.edu
Hi everyone,
Michal Feldman from Tel-Aviv University is speaking at RAIN this Wednesday.
Venue: Y2E2 101
Time: 12 noon - 1 pm.
Details of her talk are below. If you would like to meet with her, please sign up in the following document: https://docs.google.com/spreadsheets/d/16BSjFVlts7JlTvLBwHhbaYyngQ--MqLmrlZyFsf3nl0/edit?usp=sharing
Hope to see you there!
Title: Prophet Inequalities Made Easy: Stochastic Optimization by Pricing Non-Stochastic Inputs
Speaker: Michal Feldman, Tel-Aviv University
Abstract: We present a general framework for stochastic online maximization problems with combinatorial feasibility constraints. The framework establishes prophet inequalities by constructing price-based online approximation algorithms, a natural extension of threshold algorithms for settings beyond binary selection. Our analysis takes the form of an extension theorem: we derive sufficient conditions on prices when all weights are known in advance, then prove that the resulting approximation guarantees extend directly to stochastic settings. Our framework unifies and simplifies much of the existing literature on prophet inequalities and posted price mechanisms, and is used to derive new and improved results for combinatorial markets (with and without complements), multi-dimensional matroids, and sparse packing problems. Finally, we highlight a surprising connection between the smoothness framework for bounding the price of anarchy of mechanisms and our framework, and show that many smooth mechanisms can be recast as posted price mechanisms with comparable performance guarantees.
Bio: Michal Feldman is a Professor of computer science in the Blavatnik School of Computer Science at Tel Aviv University and a researcher at Microsoft Research (MSR) Herzliya. Her research focuses on the intersection of computer science, game theory and microeconomics. She received her Ph.D. from the University of California at Berkeley in 2005, and did her postdoc at the Hebrew University (2005-07). She was a faculty member in the School of Business Administration and the Center for the study of rationality at the Hebrew University (2007-13), and a visiting professor at Harvard University and Microsoft Research New England (2011-13). She serves on the editorial board of various journals, including GEB, MOR, JCSS and ACM TEAC. She is the vice chair of ACM SIGEcom, and served as the PC chair of ACM Conference on Economics and Computation 2015. She is the recipient of various grants and fellowships, including ERC (European Research Council), Marie Curie IOF, Alon, and ISF. She is a member of the Israeli Young Academy and an alumna of the Global Young Academy.
--
----------------------------------------------------------------------------
Anilesh K. Krishnaswamy,
Ph.D. candidate,
Department of Electrical Engineering,
Stanford University.
Cell: 650-387-7272
--++**==--++**==--++**==--++**==--++**==--++**==--++**==
internetalgs mailing list
internetalgs at lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/internetalgs
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Wed Oct 18 10:47:21 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Wed, 18 Oct 2017 10:47:21 -0700
Subject: [theory-seminar] [theory-lunch] Open problem session this week
Message-ID:
Hi Everyone
We will have an open problem session this week. Every speaker will present
an open problem for 5 minutes. If you are keen on presenting some ideas,
please let me know (however there is no guarantee if we run out of slots).
As usual, we meet from 12:00pm to 1:00pm at Gates 463A.
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Wed Oct 18 10:47:21 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Wed, 18 Oct 2017 10:47:21 -0700
Subject: [theory-seminar] [theory-lunch] Open problem session this week
Message-ID:
Hi Everyone
We will have an open problem session this week. Every speaker will present
an open problem for 5 minutes. If you are keen on presenting some ideas,
please let me know (however there is no guarantee if we run out of slots).
As usual, we meet from 12:00pm to 1:00pm at Gates 463A.
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From vaggos at stanford.edu Thu Oct 19 13:25:19 2017
From: vaggos at stanford.edu (Vaggos Chatziafratis)
Date: Thu, 19 Oct 2017 20:25:19 +0000
Subject: [theory-seminar] [theory-lunch] Open problem session this week
In-Reply-To:
References:
Message-ID:
I found a cell phone, just after Theory Lunch.
Whomever lost it, just send me an email, or come pick it up from 463A.
best,
vaggos
________________________________
From: theory-seminar on behalf of Hongyang Zhang
Sent: Wednesday, October 18, 2017 10:47:21 AM
To: theory-seminar at lists.stanford.edu; thseminar at cs.stanford.edu
Subject: [theory-seminar] [theory-lunch] Open problem session this week
Hi Everyone
We will have an open problem session this week. Every speaker will present an open problem for 5 minutes. If you are keen on presenting some ideas, please let me know (however there is no guarantee if we run out of slots).
As usual, we meet from 12:00pm to 1:00pm at Gates 463A.
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From vaggos at stanford.edu Thu Oct 19 13:25:19 2017
From: vaggos at stanford.edu (Vaggos Chatziafratis)
Date: Thu, 19 Oct 2017 20:25:19 +0000
Subject: [theory-seminar] [theory-lunch] Open problem session this week
In-Reply-To:
References:
Message-ID:
I found a cell phone, just after Theory Lunch.
Whomever lost it, just send me an email, or come pick it up from 463A.
best,
vaggos
________________________________
From: theory-seminar on behalf of Hongyang Zhang
Sent: Wednesday, October 18, 2017 10:47:21 AM
To: theory-seminar at lists.stanford.edu; thseminar at cs.stanford.edu
Subject: [theory-seminar] [theory-lunch] Open problem session this week
Hi Everyone
We will have an open problem session this week. Every speaker will present an open problem for 5 minutes. If you are keen on presenting some ideas, please let me know (however there is no guarantee if we run out of slots).
As usual, we meet from 12:00pm to 1:00pm at Gates 463A.
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Thu Oct 19 15:59:10 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Thu, 19 Oct 2017 22:59:10 +0000
Subject: [theory-seminar] Theory Seminar in 15 minutes
Message-ID:
Hi All,
Just a reminder that theory seminar will be at Gates 463A in 15 minutes.
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Mon Oct 23 15:30:18 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Mon, 23 Oct 2017 22:30:18 +0000
Subject: [theory-seminar] Theory Seminar on Thursday
Message-ID:
Hi All,
There will be a theory seminar on Thursday at 415pm in Gates 463A. Hope you can make it!
Raghu Meka (UCLA)
Learning discrete Markov Random Fields with nearly optimal runtime and sample complexity.
We give an algorithm for learning the structure of an undirected graphical model that has essentially optimal sample complexity and running time. We make no assumptions on the structure of the graphical model. For Ising models, this subsumes and improves on all prior work. For general t-wise MRFs, these are the first results of their kind.
Our approach is new and uses a multiplicative-weight update algorithm. Our algorithm-- Sparsitron-- is easy to implement (has only one parameter) and holds in the online setting. It also gives the first provably efficient solution to the problem of learning sparse Generalized Linear Models (GLMs).
Joint work with Adam Klivans.
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Wed Oct 25 10:09:53 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Wed, 25 Oct 2017 10:09:53 -0700
Subject: [theory-seminar] [theory-lunch] Anna Ma -- Iterative methods
Message-ID:
Hi Everyone
Anna Ma, who is visiting Berkeley this semester, will give a talk on
"Iterative methods for solving factorized linear systems" (see abstract
below).
As usual, we meet from 12:00pm to 1:00pm, at Gates 463A. See you all.
============================
Iterative methods for solving factorized linear systems
Abstract. Stochastic iterative algorithms such as the Kacmarz and
Gauss-Seidel methods have gained recent attention because of their speed,
simplicity, and the ability to approximately solve large-scale linear
systems of equations without needing to access the entire matrix. In this
work, we consider the setting where we wish to solve a linear system in a
large matrix X that is stored in a factorized form, X = U V ; this setting
either arises naturally in many applications or may be imposed when working
with large low- rank datasets for reasons of space required for storage. We
propose a variant of the randomized Kaczmarz method for such systems that
takes advantage of the factored form, and avoids computing X . We prove an
exponential convergence rate and supplement our theoretical guarantees with
experimental evidence demonstrating that the factored variant yields
significant acceleration in convergence.
============================
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Wed Oct 25 10:09:53 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Wed, 25 Oct 2017 10:09:53 -0700
Subject: [theory-seminar] [theory-lunch] Anna Ma -- Iterative methods
Message-ID:
Hi Everyone
Anna Ma, who is visiting Berkeley this semester, will give a talk on
"Iterative methods for solving factorized linear systems" (see abstract
below).
As usual, we meet from 12:00pm to 1:00pm, at Gates 463A. See you all.
============================
Iterative methods for solving factorized linear systems
Abstract. Stochastic iterative algorithms such as the Kacmarz and
Gauss-Seidel methods have gained recent attention because of their speed,
simplicity, and the ability to approximately solve large-scale linear
systems of equations without needing to access the entire matrix. In this
work, we consider the setting where we wish to solve a linear system in a
large matrix X that is stored in a factorized form, X = U V ; this setting
either arises naturally in many applications or may be imposed when working
with large low- rank datasets for reasons of space required for storage. We
propose a variant of the randomized Kaczmarz method for such systems that
takes advantage of the factored form, and avoids computing X . We prove an
exponential convergence rate and supplement our theoretical guarantees with
experimental evidence demonstrating that the factored variant yields
significant acceleration in convergence.
============================
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From silas at stanford.edu Thu Oct 26 16:03:02 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Thu, 26 Oct 2017 23:03:02 +0000
Subject: [theory-seminar] Theory Seminar in 10 minutes in 463A
Message-ID:
Hope to see you there!
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From jacob.steinhardt at gmail.com Thu Oct 26 21:54:58 2017
From: jacob.steinhardt at gmail.com (Jacob Steinhardt)
Date: Thu, 26 Oct 2017 21:54:58 -0700
Subject: [theory-seminar] STOC paper swap
Message-ID:
Hi all,
As discussed in theory lunch today, we will have a paper swap for STOC
Monday at 6pm, in the usual theory seminar room. Please sign up here
if you have a paper(s) that you would like reviewed at the paper swap.
Later we will have people sign up as reviewers.
Greg has graciously agreed to pay for food, in case that affects anyone's
interest in coming.
-Jacob
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From jacob.steinhardt at gmail.com Sat Oct 28 10:36:32 2017
From: jacob.steinhardt at gmail.com (Jacob Steinhardt)
Date: Sat, 28 Oct 2017 10:36:32 -0700
Subject: [theory-seminar] STOC paper swap
In-Reply-To:
References:
Message-ID:
Okay, looks like we have a fair number of papers signed up, so please start
signing up to review papers. If you signed up a paper to be reviewed, then
you should definitely sign up as a reviewer, but you can also sign up even
if you are not submitting something if you want to help our community have
stronger submissions (plus there will be free food :)).
-Jacob
On Thu, Oct 26, 2017 at 9:54 PM, Jacob Steinhardt <
jacob.steinhardt at gmail.com> wrote:
> Hi all,
>
> As discussed in theory lunch today, we will have a paper swap for STOC
> Monday at 6pm, in the usual theory seminar room. Please sign up here
>
> if you have a paper(s) that you would like reviewed at the paper swap.
> Later we will have people sign up as reviewers.
>
> Greg has graciously agreed to pay for food, in case that affects anyone's
> interest in coming.
>
> -Jacob
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From jacob.steinhardt at gmail.com Sat Oct 28 10:41:09 2017
From: jacob.steinhardt at gmail.com (Jacob Steinhardt)
Date: Sat, 28 Oct 2017 10:41:09 -0700
Subject: [theory-seminar] STOC paper swap
In-Reply-To:
References:
Message-ID:
P.S. I would recommend signing up for either 1 or 2 papers to read.
(Signing up for just 1 is completely fine.)
-Jacob
On Sat, Oct 28, 2017 at 10:36 AM, Jacob Steinhardt <
jacob.steinhardt at gmail.com> wrote:
> Okay, looks like we have a fair number of papers signed up, so please
> start signing up to review papers. If you signed up a paper to be reviewed,
> then you should definitely sign up as a reviewer, but you can also sign up
> even if you are not submitting something if you want to help our community
> have stronger submissions (plus there will be free food :)).
>
> -Jacob
>
> On Thu, Oct 26, 2017 at 9:54 PM, Jacob Steinhardt <
> jacob.steinhardt at gmail.com> wrote:
>
>> Hi all,
>>
>> As discussed in theory lunch today, we will have a paper swap for STOC
>> Monday at 6pm, in the usual theory seminar room. Please sign up here
>>
>> if you have a paper(s) that you would like reviewed at the paper swap.
>> Later we will have people sign up as reviewers.
>>
>> Greg has graciously agreed to pay for food, in case that affects anyone's
>> interest in coming.
>>
>> -Jacob
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From jacob.steinhardt at gmail.com Mon Oct 30 09:31:07 2017
From: jacob.steinhardt at gmail.com (Jacob Steinhardt)
Date: Mon, 30 Oct 2017 09:31:07 -0700
Subject: [theory-seminar] STOC paper swap
In-Reply-To:
References:
Message-ID:
Okay, I currently have the following people signed up:
-Aditi
-Jacob
-Brian Axelrod
-Ray
-Michael Kim
-Joshua Wang
-Ben Plautt
-Weihao
We could use 2-3 more reviewers, so if anyone else wants to come, please
let me know (also so I know how much food to order).
If you have a paper, I recommend making a dropbox link to it, as that
allows people to make inline comments. (Or you can print it out the
old-fashioned way.)
-Jacob
On Sat, Oct 28, 2017 at 10:41 AM, Jacob Steinhardt <
jacob.steinhardt at gmail.com> wrote:
> P.S. I would recommend signing up for either 1 or 2 papers to read.
> (Signing up for just 1 is completely fine.)
>
> -Jacob
>
> On Sat, Oct 28, 2017 at 10:36 AM, Jacob Steinhardt <
> jacob.steinhardt at gmail.com> wrote:
>
>> Okay, looks like we have a fair number of papers signed up, so please
>> start signing up to review papers. If you signed up a paper to be reviewed,
>> then you should definitely sign up as a reviewer, but you can also sign up
>> even if you are not submitting something if you want to help our community
>> have stronger submissions (plus there will be free food :)).
>>
>> -Jacob
>>
>> On Thu, Oct 26, 2017 at 9:54 PM, Jacob Steinhardt <
>> jacob.steinhardt at gmail.com> wrote:
>>
>>> Hi all,
>>>
>>> As discussed in theory lunch today, we will have a paper swap for STOC
>>> Monday at 6pm, in the usual theory seminar room. Please sign up here
>>>
>>> if you have a paper(s) that you would like reviewed at the paper swap.
>>> Later we will have people sign up as reviewers.
>>>
>>> Greg has graciously agreed to pay for food, in case that affects
>>> anyone's interest in coming.
>>>
>>> -Jacob
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From reingold at stanford.edu Mon Oct 30 09:33:44 2017
From: reingold at stanford.edu (Omer Reingold)
Date: Mon, 30 Oct 2017 09:33:44 -0700
Subject: [theory-seminar] STOC paper swap
In-Reply-To:
References:
Message-ID:
Thanks for organizing it Jacob!!
Omer
On Mon, Oct 30, 2017 at 9:31 AM, Jacob Steinhardt <
jacob.steinhardt at gmail.com> wrote:
> Okay, I currently have the following people signed up:
>
> -Aditi
> -Jacob
> -Brian Axelrod
> -Ray
> -Michael Kim
> -Joshua Wang
> -Ben Plautt
> -Weihao
>
> We could use 2-3 more reviewers, so if anyone else wants to come, please
> let me know (also so I know how much food to order).
>
> If you have a paper, I recommend making a dropbox link to it, as that
> allows people to make inline comments. (Or you can print it out the
> old-fashioned way.)
>
> -Jacob
>
> On Sat, Oct 28, 2017 at 10:41 AM, Jacob Steinhardt <
> jacob.steinhardt at gmail.com> wrote:
>
>> P.S. I would recommend signing up for either 1 or 2 papers to read.
>> (Signing up for just 1 is completely fine.)
>>
>> -Jacob
>>
>> On Sat, Oct 28, 2017 at 10:36 AM, Jacob Steinhardt <
>> jacob.steinhardt at gmail.com> wrote:
>>
>>> Okay, looks like we have a fair number of papers signed up, so please
>>> start signing up to review papers. If you signed up a paper to be reviewed,
>>> then you should definitely sign up as a reviewer, but you can also sign up
>>> even if you are not submitting something if you want to help our community
>>> have stronger submissions (plus there will be free food :)).
>>>
>>> -Jacob
>>>
>>> On Thu, Oct 26, 2017 at 9:54 PM, Jacob Steinhardt <
>>> jacob.steinhardt at gmail.com> wrote:
>>>
>>>> Hi all,
>>>>
>>>> As discussed in theory lunch today, we will have a paper swap for STOC
>>>> Monday at 6pm, in the usual theory seminar room. Please sign up here
>>>>
>>>> if you have a paper(s) that you would like reviewed at the paper swap.
>>>> Later we will have people sign up as reviewers.
>>>>
>>>> Greg has graciously agreed to pay for food, in case that affects
>>>> anyone's interest in coming.
>>>>
>>>> -Jacob
>>>>
>>>
>>>
>>
>
> _______________________________________________
> 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 silas at stanford.edu Mon Oct 30 11:01:49 2017
From: silas at stanford.edu (Shashwat Silas)
Date: Mon, 30 Oct 2017 18:01:49 +0000
Subject: [theory-seminar] Theory Seminar on Thursday
Message-ID:
Hi All,
There will be a theory seminar on Thursday at 415pm at Gates 463A. Hope you can make it!
Avishay Tal (Stanford)
Computing Requires Larger Formulas than Approximating
A de-Morgan formula over Boolean variables x_1, ..., x_n is a binary tree whose internal nodes are marked with AND or OR gates and whose leaves are marked with variables or their negation. We define the size of the formula as the number of leaves in it. Proving that some explicit function (in P or NP) requires large formulas is a central open question in computational complexity.
In this work, we introduce a size-amplification hardness reduction for de-Morgan formulas. We show that average-case hardness implies worst-case hardness for a larger size. More precisely, if a function f cannot be computed correctly on more than 1/2 + eps of the inputs by any formula of size s, then computing f correctly on all inputs requires size ~s*log(1/eps). The tradeoff is essentially tight. Quite surprisingly, the proof relies on a result from quantum query complexity by Reichardt [SODA, 2011].
As an application, we improve the best known formula size lower bounds for explicit functions by logarithmic factors to ~n^3/log(n). In addition, we propose candidates for explicit functions that we believe have formula size ~n^4, and prove non trivial super-quadratic formula size lower bounds for them using our reduction. Learning discrete Markov Random Fields with nearly optimal runtime and sample complexity.
Thanks,
Shashwat
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Tue Oct 31 10:58:28 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Tue, 31 Oct 2017 10:58:28 -0700
Subject: [theory-seminar] =?utf-8?q?=5Btheory-lunch=5D_Isma=C3=ABl_Lemhadr?=
=?utf-8?q?i_on_Community_detection?=
Message-ID:
Hi Everyone
This Thursday, Isma?l Lemhadri will tell us about "Community detection in
the stochastic block model" (see abstract below). Isma?l is currently doing
his PhD in stats at Stanford.
As usual, we meet from 12:00pm to 1:00pm at Gates 463A.
=========================
Community detection in the stochastic block model
I analyze the problem of community recovery in the context of the
stochastic block model. To do so I introduce a semi-definite program that
interprets as a relaxed version of K-means. I prove that it achieves exact
recovery with high probability as soon as a simple condition on the
'within-between' covariance gap is satisfied. I conclude by comparing our
condition to classical SDP formulations, and I discuss a few
generalizations.
============================
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
From hongyang at cs.stanford.edu Tue Oct 31 10:58:28 2017
From: hongyang at cs.stanford.edu (Hongyang Zhang)
Date: Tue, 31 Oct 2017 10:58:28 -0700
Subject: [theory-seminar] =?utf-8?q?=5Btheory-lunch=5D_Isma=C3=ABl_Lemhadr?=
=?utf-8?q?i_on_Community_detection?=
Message-ID:
Hi Everyone
This Thursday, Isma?l Lemhadri will tell us about "Community detection in
the stochastic block model" (see abstract below). Isma?l is currently doing
his PhD in stats at Stanford.
As usual, we meet from 12:00pm to 1:00pm at Gates 463A.
=========================
Community detection in the stochastic block model
I analyze the problem of community recovery in the context of the
stochastic block model. To do so I introduce a semi-definite program that
interprets as a relaxed version of K-means. I prove that it achieves exact
recovery with high probability as soon as a simple condition on the
'within-between' covariance gap is satisfied. I conclude by comparing our
condition to classical SDP formulations, and I discuss a few
generalizations.
============================
Best,
Hongyang
--
Hongyang Zhang ?????
cs.stanford.edu/people/hongyang/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: