Machine Learning

Machine Learning Seminars


Last changed Fri, 2012-05-11 22:58

Tuesday, May 20, 2014 - 12:30

A General Framework for Fast Stagewise Algorithms

Speaker: Ryan Tibshirani, Carnegie Mellon University
Location: CSE 305

Forward stagewise regression follows a very simple strategy for constructing a sequence of sparse regression estimates: starting with all coefficients equal to zero, it iteratively updates the coefficient (by a small amount ε) corresponding to the variable that has maximal absolute inner product with the current residual. This procedure has an interesting connection to the lasso: under some conditions, it can be shown that the sequence of forward stagewise estimates exactly coincides with the lasso path, as the step size ε goes to zero. Further, essentially the same equivalence holds outside of the regression setting, for minimizing a differentiable convex loss function subject to an l1 norm constraint (and the stagewise algorithm now updating the coefficient corresponding to the maximal absolute component of the gradient).

Even when they do not match their l1-constrained analogues, stagewise estimates provide a useful approximation, and are computationally appealing. Their success in sparse modeling motivates the question: can a simple, effective strategy like forward stagewise be applied more broadly in other regularization settings, beyond the l1 norm and sparsity? This is the focus of the talk; we present a general framework for stagewise estimation, which yields fast algorithms for problems such as group-structured learning, matrix completion, image denoising, and more.

Tuesday, May 6, 2014 - 12:30

Bayesian Optimization of Machine Learning Algorithms

Speaker: Ryan Adams, Harvard University
Location: CSE 403

Machine learning algorithms frequently involve careful tuning of learning parameters and model hyperparameters. Unfortunately, this tuning is often a "black art" requiring expert experience, rules of thumb, or sometimes brute-force search. There is therefore great appeal for automatic approaches that can optimize the performance of any given learning algorithm to the problem at hand. I will describe my recent work on solving this problem with Bayesian nonparametrics, using Gaussian processes. This approach of "Bayesian optimization" models the generalization performance as an unknown objective function with a GP prior. I will discuss new algorithms that account for variable cost in function evaluation and take advantage of parallelism in evaluation. These new algorithms improve on previous automatic procedures and can reach or surpass human expert-level optimization for many algorithms including latent Dirichlet allocation for text analysis, structured SVMs for protein motif finding, and convolutional neural networks for visual object recognition.

Tuesday, April 29, 2014 - 12:30

Sparse latent factor models to uncover structure in genomic data

Speaker: Barbara Engelhardt, Duke University
Location: CSE 305

Within high-dimensional genomic data, including genetic markers, complex phenotypes, and gene expression measurements, hides a substantial amount of low dimensional structure. In order to extract this low dimensional structure from high dimensional data, latent factor models with appropriate priors and modeling assumptions can be applied. A key advantage of sparse latent factor models in genomic applications is interpretability: the low dimensional structure can often be understood in a biological context. Model specification here is important, however, as genomics data generally includes small numbers of samples n with respect to enormous numbers of features p. Sparse linear latent factor models have been used to identify population substructure. Applying these models to high dimensional noisy phenotypes produces interpretable phenotypes that can subsequently tested for association with genetic variants. This same idea can be considered for gene expression traits, where the low dimensional structure captures groups of interacting genes, and the associated genetic variants represent master regulators. In this talk, I will discuss recent work on sparse latent factor models, and results on genomic data.

Tuesday, April 22, 2014 - 12:30

Submodular combinatorial problems in Machine Learning: Algorithms and Applications

Speaker: Rishabh Iyer, UW Electrical Engineering
Location: CSE 305

Having long been recognized in combinatorics, the concept of submodularity is becoming increasingly important in machine learning, since it can capture intuitive and yet non-trivial interactions between variables. This expressiveness has found many applications in machine learning, particularly in understanding structured data like text, vision and speech. Submodular functions naturally capture aspects of information, coverage and diversity in maximization problems, while also modelling notions of cooperation, sparsity, complexity and economies of scale in minimization problems. In this talk, we shall consider a large class of submodular optimization problems and motivate them with several real world applications, particularly in machine learning. We shall then provide a unifying algorithmic framework for solving several of these optimization problems, and highlight the scalability and practicality of this approach. We willl also highlight novel theoretical characterizations, which provide better connections between theory and practice. We shall ground this entire talk with a large number of applications in machine learning, including image segmentation, image correspondence, image collection summarization, feature selection, and data subset selection. This talk should be self contained and shall not require prior knowledge of submodular functions and optimization.

(this is based on joint work with Jeff Bilmes, Stefanie Jegelka, Sebastian Tschiatschek and Kai Wei)

Tuesday, April 8, 2014 - 12:30

Theoretical and Algorithmic Foundations of Online Learning

Speaker: Sasha Rakhlin, Wharton School, University of Pennsylvania
Location: CSE 305

A major challenge for machine learning in the next decade is the development of methods that continuously predict and learn from a stream of data. The scale of data and the non-stationary nature of the environment make the prevalent ``learn from a batch of i.i.d. data'' paradigm of Statistical Learning inadequate. Despite the extensive literature on sequential prediction (online learning) methods, the theoretical understanding of the subject has been lacking, and the results have been obtained on a case-by-case basis. This stands in contrast to Statistical Learning, where the inherent complexities have been well-understood in the last forty years. In this talk, we focus on no-regret online learning and develop the relevant notions of complexity in a surprising parallel to Statistical Learning Theory. This non-constructive study of inherent complexity is then augmented with a recipe for developing efficient online learning algorithms via a notion of a relaxation. To demonstrate the utility of our approach, we develop a new family of randomized methods, as well as new algorithms for collaborative filtering and node prediction.

Monday, March 31, 2014 - 12:00

Scaling and Generalizing Variational Inference

Speaker: David Blei, Princeton University
Location: CSE 305

Latent variable models have become a key tool for the modern statistician, letting us express complex assumptions about the hidden structures that underlie our data. Latent variable models have been successfully applied in numerous fields including natural language processing, computer vision, electronic medical records, genetics, neuroscience, astronomy, political science, sociology, the digital humanities, and many others.

The central computational problem in latent variable modeling is posterior inference, the problem of approximating the conditional distribution of the latent variables given the observations. Posterior inference is central to both exploratory tasks, where we investigate hidden structures that underlie our data, and predictive tasks, where we use the inferred structures to generalize about future data. Approximate posterior inference algorithms have revolutionized Bayesian statistics, revealing its potential as a usable and general-purpose language for data analysis.

Bayesian statistics, however, has not yet reached this potential. First, statisticians and scientists regularly encounter massive data sets, but existing approximate inference algorithms do not scale well. Second, most approximate inference algorithms are not generic; each must be adapted to the specific model at hand. This often requires significant model-specific analysis, which precludes us from easily exploring a variety of models.

In this talk I will discuss our recent research on addressing these two limitations. First I will describe stochastic variational inference, an approximate inference algorithm for handling massive data sets. Stochastic inference is easily applied to a large class of Bayesian models, including time-series models, factor models, and Bayesian nonparametric models. I will demonstrate its application to probabilistic topic models of text conditioned on millions of articles. Stochastic inference opens the door to scalable Bayesian computation for modern data analysis.

Then I will discuss black box variational inference. Black box inference is a generic algorithm for approximating the posterior. We can easily apply it to many models with little model-specific derivation and few restrictions on their properties. Black box inference performs better than similarly generic sampling algorithms, such as Metropolis-Hastings inside Gibbs, and can be composed with stochastic inference to handle massive data. I will demonstrate its use on a suite of nonconjugate models of longitudinal healthcare data.

This is joint work based on these two papers:

M. Hoffman, D. Blei, J. Paisley, and C. Wang. Stochastic variational inference. Journal of Machine Learning Research, 14:1303-1347.

R. Ranganath and D. Blei. Black box variational inference. Artificial Intelligence and Statistics, 2014.

Tuesday, February 4, 2014 - 12:30

On Estimating Many Effect-sizes, Selection Bias, and Local Adaptive Shrinkage

Speaker: Noah Simon, UW Biostatistics
Location: CSE 305

There is a growing scientific interest in simultaneously estimating the means (or effect-sizes) of many different features. Classical (but very unintuitive) results in Stein-estimation show that even if these features are independent, cleverly coupling their mean estimates can greatly improve estimation accuracy. These ideas have been extended to build locally adaptive estimators based on perspectives from non-parametric empirical Bayesian estimation (and compound decision theory). Unfortunately, these estimators are not simple, and are really only tractable in highly idealized scenarios.

In recent work, we introduce a different framework for this estimation problem. Our framework is intuitive and shows how estimates from "independent features" become coupled through selection bias. We also give a simple general estimator based on resampling which is applicable and performs well in a wide variety of scenarios.

In this talk I will discuss the intuition behind "coupling estimates from independent features." I will review empirical Bayes/Stein Estimation, and I will introduce our framework. I will explain the connections between these estimators, show how they compare in practice.

This presentation is intended for a general machine learning audience.

This is joint work with Richard Simon (NIH) and recently Kean Ming Tan and Daniela Witten (UW) with occasional skepticism from Brad Efron (Stanford).

Tuesday, January 14, 2014 - 12:30

Recovering Block-structured Activations Using Compressive Measurements

Speaker: Mladen Kolar, University of Chicago
Location: CSE 305

In recent years, with the advancement of large-scale data acquisition technology in various engineering, scientific and socio-economical domains, traditional machine learning and statistical methods have started to struggle with massive amounts of increasingly high-dimensional data. Luckily, in many problems there is a simple structure underlying high-dimensional data that can be exploited to make learning feasible.

In this talk, I will focus on the problem of detection and localization of a contiguous block of weak activation in a large matrix, from a small number of noisy, possibly adaptive, compressive measurements. This is closely related to the problem of compressed sensing, where the task is to estimate a sparse vector using a small number of linear measurements. Contrary to results in compressed sensing, where it has been shown that neither adaptivity nor contiguous structure help much, we show that for reliable localization the magnitude of the weakest signals is strongly influenced by both structure and the ability to choose measurements adaptively while for detection neither adaptivity nor structure reduce the requirement on the magnitude of the signal. We characterize the precise tradeoffs between the various problem parameters, the signal strength and the number of measurements required to reliably detect and localize the block of activation. The sufficient conditions are complemented with information theoretic lower bounds.

Joint work with Sivaraman Balakrishnan, Alessandro Rinaldo and Aarti Singh.

Tuesday, November 12, 2013 - 12:15

Using machine learning to achieve MOOC-scale grading and feedback for open-ended assignments

Speaker: Jonathan Huang, Stanford
Location: CSE Gates Commons
Massive open online courses (MOOCs) are one of the latest internet revolutions and may cure the “cost disease” of higher education. However, dealing with tens of thousands of homework submissions (particularly open-ended ones) remains a challenge in the online classroom. In courses where the student-teacher ratio can be ten thousand to one or worse, it is impossible for instructors to personally give feedback to students or to understand the multitude of approaches taken, and pitfalls encountered by their students.

Peer grading serves as a critical tool for scaling the grading of complex, open-ended assignments to courses with tens or hundreds of thousands of students. But despite promising initial trials, it does not always deliver accurate results compared to human experts. I will talk about methods for estimating and correcting for grader biases and reliabilities, showing significant improvement in peer grading accuracy on real data with 63,199 peer grades from Coursera's HCI course offerings --- the largest peer grading networks analyzed to date. I will also discuss the relationship between grader biases and reliabilities to other student factors such as student engagement, performance as well as commenting style.

I will also discuss scalable feedback in the context of coding assignments, which have more structure than arbitrary open ended assignments. I outline a novel way to decompose online coding submissions into a vocabulary of “code phrases”. Based on this vocabulary, we have architected a queryable index that allows fast searches into the massive dataset of student homework submissions. To demonstrate the utility of our homework search engine we index over a million code submissions from users worldwide in Stanford’s Machine Learning massive open online course (MOOC) and then (a) semi-automatically learn shared structure amongst homework submissions and (b) generate specific feedback for student mistakes.

This is joint work with Chris Piech, Andy Nguyen, Zhenghao Chen, Chuong Do, Andrew Ng, Daphne Koller and Leonidas Guibas

Tuesday, October 29, 2013 - 12:30

A Characterization of Online Learning Problems

Speaker: Ofer Dekel, MSR
Location: CSE Gates Commons
We study the learning rate of the adversarial online learning problem in a finite action space. The “adversarial multi-armed bandit” problem and “predicting with expert advice” are two well-known special cases of this fundamental online learning setting. We present a rich and complete characterization of the inherent difficulty of this problem as we vary the power of the adversary and the amount of information provided to the learning algorithm. Specifically, we classify the different variants of this problem into three distinct difficulty classes: easy problems (those with sqrt(T) regret), such as learning against oblivious adversaries; hard problems (with T^(2/3) regret), such as the multi-armed bandit with switching costs; and unlearnable problems (with linear regret), such as learning against adversaries that have an unbounded memory. Our new characterization has two important implications. First, we formally confirm the common knowledge that learning with bandit feedback can be strictly harder than learning with full information. Second, we prove that learning in the bandit setting requires extensive switching between actions.

Joint work with Raman Arora (TTI-C), Nicolo Cesa-Bianchi (Milano), Jian Ding (UChicago), Tomer Koren (Technion), Yuval Peres (MSR), Ohad Shamir (Weizmann), and Ambuj Tewari (UMICH)

The talk is intended for a general machine learning audience.

Pages