works on

From the 1 of 15 linked papers with an AI index.

activity
20192022
most citedA Theoretical Analysis of Contrastive Unsupervised Representation Learning

204 citations · 389 across the 11 of their papers we have counts for

collaborators

15 papers

cs.LG202215 cited

Understanding Contrastive Learning Requires Incorporating Inductive Biases

Nikunj Saunshi, Jordan Ash, Surbhi Goel +5

Contrastive learning is a popular form of self-supervised learning that encourages augmentations (views) of the same input to have more similar representations compared to augmenta…

cs.LG20212 cited

Rip van Winkle's Razor: A Simple Estimate of Overfit to Test Data

Sanjeev Arora, Yi Zhang

Traditional statistics forbids use of test data (a.k.a. holdout data) during training. Dwork et al. 2015 pointed out that current practices in machine learning, whereby researchers…

cs.LG2021

On the Validity of Modeling SGD with Stochastic Differential Equations (SDEs)

Zhiyuan Li, Sadhika Malladi, Sanjeev Arora

It is generally recognized that finite learning rate (LR), in contrast to infinitesimal LR, is important for good generalization in real-life deep nets. Most attempted explanations…

cs.CL2020

TextHide: Tackling Data Privacy in Language Understanding Tasks

Yangsibo Huang, Zhao Song, Danqi Chen +2

An unsolved challenge in distributed or federated learning is to effectively mitigate privacy risks without slowing down training or reducing accuracy. In this paper, we propose Te…

cs.LG202010 cited

Reconciling Modern Deep Learning with Traditional Optimization Analyses: The Intrinsic Learning Rate

Zhiyuan Li, Kaifeng Lyu, Sanjeev Arora

The paper proposes an 'intrinsic learning rate' that combines the usual learning rate and weight decay to model the training dynamics of normalized deep networks via stochastic dif…

cs.CL2020

A Mathematical Exploration of Why Language Models Help Solve Downstream Tasks

Nikunj Saunshi, Sadhika Malladi, Sanjeev Arora

Autoregressive language models, pretrained using large text corpora to do well on next word prediction, have been successful at solving many downstream tasks, even with zero-shot u…