Category: Uncategorized

Welcome back, data science enthusiasts! We had an eventful seminar series last semester and have some new plans springing up for the new semester.

We’re kicking off our new batch of sessions with a talk by Binghui Wang from Dr. Neil Gong’s group, who will be presenting an introduction of Adversarial Machine Learning, a research filed that lies at the intersection of Machine Learning and Computer Security.

Date: Friday, 13th January, 2017
Time: 3pm to 4pm
Venue: 2222, Coover Hall

You can find the slides for the talk here: Slides

Abstract: I will motivate the problem of robust PCA through some examples, followed by a brief introduction to the existing approaches to solve this problem. I will describe the recently proposed Non-Convex algorithm in some detail. I also intend to go over the proof outline for demonstrating the performance guarantees by discussing a few key points in detail. If time permits I will show some results of the proposed algorithm obtained on simulated data.

Gauri Jagatap, from Dr. Chinmay Hegde’s group will be presenting an overview of phase retrieval problems in signal processing, this Friday. She will primarily speak on a popular phase recovery strategy called AltMinPhase, based on the paper “Phase Retrieval Using Alternating Minimization” by Praneeth Netrapalli, Prateek Jain, and Sujay Sanghavi.

She will later also introduce a newer approach for phase retrieval of sparse signals, called “Efficient Compressive Phase Retrieval with Constrained Sensing Vectors“, by Sohail Bahmani, Justin Romberg.

Phase retrieval is essentially the problem of recovering the phase of a signal from magnitude measurements. In several applications in crystallography, optics, spectroscopy and tomography, it is harder or infeasible to record the phase of measurements, while recording the magnitudes is significantly easier.

Date: 11/18/2016

Time: 3:00pm – 4:00pm

Venue: 2222, Coover

Slides:Phase Retrieval (updated)

Davood Hajinezhad from Dr. Mingyi Hong’s group will be presenting this week on “A Nonconvex Primal-Dual Splitting Method for Distributed and Stochastic Optimization”. Details of the talk are as given below:

Abstract: We study a stochastic and distributed algorithm for nonconvex  problems whose objective consists of a sum of $latex N$ nonconvex $latex L_i/N$-smooth functions, plus a  nonsmooth regularizer. The proposed NonconvEx primal-dual SpliTTing (NESTT) algorithm splits the problem into $latex N$ subproblems, and utilizes an augmented Lagrangian based primal-dual scheme to solve it in a distributed and stochastic manner. With a special non-uniform sampling, a version of NESTT achieves $latex \epsilon$-stationary solution  using $latex O((\sum_{i=1}^N\sqrt{L_i/N})^2/\epsilon)$ gradient evaluations, which can be up to $latex O(N)$ times better than the (proximal) gradient descent methods. It also achieves Q-linear convergence rate for nonconvex $latex \ell_1$ penalized quadratic problems with polyhedral constraints. Further, we reveal  a fundamental connection between primal-dual based methods and a few primal only methods such as IAG/SAG/SAGA.

Han Guo from Dr. Namrata Vaswani’s group will be presenting this week, on the topic of Video Denoising and Enhancement via Dynamic Sparse and Low-rank Matrix Decomposition. Details of the talk are as follows:

Pan Zhong from Dr. Zhengdao Wang’s group will be presenting this week. She will be speaking on an interesting topic in statistical learning theory called VC dimension. Details of the talk are as listed below:

Venue: Coover, 2222
Time: 3:00pm to 4:00pm
Date: 10/28/2016

References:
i. Chapter 4 of Statistical Learning Theory by Vapnik, Vladimir Naumovich.
ii. Carlos Guestrin’s slides from the Fall 2007 course on Machine Learning at CMU.

You can also find the slides for the talk here.

Some exciting research going on at the Big Picture research group at Google:

As machine learning is increasingly used to make important decisions across core social domains, the work of ensuring that these decisions aren’t discriminatory becomes crucial.

Here we discuss “threshold classifiers,” a part of some machine learning systems that is critical to issues of discrimination. A threshold classifier essentially makes a yes/no decision, putting things in one category or another. We look at how these classifiers work, ways they can potentially be unfair, and how you might turn an unfair classifier into a fairer one. As an illustrative example, we focus on loan granting scenarios where a bank may grant or deny a loan based on a single, automatically computed number such as a credit score…

Read more about it here.

Source: research.google.com