• Publications
  • Influence
Gradient descent GAN optimization is locally stable
TLDR
We show that even though GAN optimization does not correspond to a convex-concave game (even for simple parameterizations), under proper conditions, equilibrium points of this optimization procedure are still locally asymptotically stable for the traditional GAN formulation. Expand
  • 217
  • 31
  • PDF
Uniform convergence may be unable to explain generalization in deep learning
TLDR
We present examples of overparameterized linear classifiers and neural networks trained by gradient descent where uniform convergence provably cannot ``explain generalization'' -- even if we take into account the implicit bias of GD. Expand
  • 79
  • 8
  • PDF
Deterministic PAC-Bayesian generalization bounds for deep networks via generalizing noise-resilience
TLDR
The ability of overparameterized deep networks to generalize well has been linked to the fact that stochastic gradient descent (SGD) finds solutions that lie in flat, wide minima in the training loss. Expand
  • 40
  • 6
  • PDF
Geriatrix: Aging what you see and what you don't see. A file system aging approach for modern storage systems
TLDR
We introduce Geriatrix, a simple-to-use profile driven file system aging tool that induces target levels of fragmentation in both allocated files (what you see) and remaining free space, unlike previous approaches that focus on just the former. Expand
  • 14
  • 3
  • PDF
Generalization in Deep Networks: The Role of Distance from Initialization
TLDR
We provide empirical evidences that demonstrate that the model capacity of SGD-trained deep networks is in fact restricted through implicit regularization of {\em the $\ell_2$ distance from the initialization}. Expand
  • 38
  • 2
  • PDF
Every Team Deserves a Second Chance: Identifying When Things Go Wrong (Student Abstract Version)
TLDR
We present a novel benefit of voting, that has not been observed before: we can use the voting patterns to assess the performance of a team and predict their final outcome. Expand
  • 17
  • 2
  • PDF
A Reinforcement Learning Approach to Online Learning of Decision Trees
TLDR
We present RLDT, an RL-based online decision tree algorithm that uses Reinforcement Learning (RL) to actively examine a minimal number of features of a data point to classify it with high accuracy. Expand
  • 4
  • 1
  • PDF
Understanding the Failure Modes of Out-of-Distribution Generalization
TLDR
Empirical studies suggest that machine learning models often rely on features, such as the background, that may be spuriously correlated with the label only during training time, resulting in poor accuracy during test-time. Expand
  • 3
  • 1
  • PDF
Learning-Theoretic Foundations of Algorithm Configuration for Combinatorial Partitioning Problems
TLDR
We address this problem for clustering, max-cut, and other partitioning problems, such as integer quadratic programming, by designing computationally efficient and sample efficient learning algorithms which receive samples from an application-specific distribution and learn a partitioning algorithm with high expected performance. Expand
  • 28
  • PDF
On Adversarial Risk and Training
TLDR
In this work we formally define the notions of adversarial perturbations, adversarial risk and adversarial training and analyze their properties. Expand
  • 10
...
1
2
3
...