no code implementations • 6 Feb 2024 • Michael Zhang, Kush Bhatia, Hermann Kumbong, Christopher Ré
Experiments show Hedgehog recovers over 99% of standard Transformer quality in train-from-scratch and finetuned-conversion settings, outperforming prior linear attentions up to 6 perplexity points on WikiText-103 with causal GPTs, and up to 8. 7 GLUE score points on finetuned bidirectional BERTs.
1 code implementation • 25 Oct 2023 • Gabriel Mukobi, Peter Chatain, Su Fong, Robert Windesheim, Gitta Kutyniok, Kush Bhatia, Silas Alberti
Here, we focus on two prevalent methods used to align these models, Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF).
no code implementations • 23 Feb 2023 • Kush Bhatia, Wenshuo Guo, Jacob Steinhardt
We specifically show that the well-studied problem of Gaussian process (GP) bandit optimization is a special case of our framework, and that our bounds either improve or are competitive with known regret guarantees for the Mat\'ern kernel.
no code implementations • 23 Jan 2023 • Pranjal Awasthi, Kush Bhatia, Sreenivas Gollapudi, Kostas Kollias
For the linear contextual bandit setup, our algorithm, based on an iterative least squares planner, achieves policy regret $\tilde{O}(\sqrt{dT} + \Delta)$.
no code implementations • 9 Dec 2022 • Joey Hong, Kush Bhatia, Anca Dragan
This begs the question: how accurate do these models need to be in order for the reward inference to be accurate?
3 code implementations • 5 Oct 2022 • Simran Arora, Avanika Narayan, Mayee F. Chen, Laurel Orr, Neel Guha, Kush Bhatia, Ines Chami, Frederic Sala, Christopher Ré
Prompting is a brittle process wherein small modifications to the prompt can cause large variations in the model predictions, and therefore significant effort is dedicated towards designing a painstakingly "perfect prompt" for a task.
Ranked #1 on Question Answering on Story Cloze
no code implementations • 22 Jul 2022 • Kush Bhatia, Nikki Lijing Kuang, Yi-An Ma, Yixin Wang
Focusing on Gaussian inferential models (or variational approximating families) with diagonal plus low-rank precision matrices, we initiate a theoretical study of the trade-offs in two aspects, Bayesian posterior inference error and frequentist uncertainty quantification error.
1 code implementation • ICLR 2022 • Alexander Pan, Kush Bhatia, Jacob Steinhardt
Reward hacking -- where RL agents exploit gaps in misspecified reward functions -- has been widely observed, but not yet systematically studied.
no code implementations • NeurIPS 2020 • Kush Bhatia, Ashwin Pananjady, Peter L. Bartlett, Anca D. Dragan, Martin J. Wainwright
Finally, we showcase the practical utility of our framework in a user study on autonomous driving, where we find that the Blackwell winner outperforms the von Neumann winner for the overall preferences.
no code implementations • 17 Apr 2021 • Kush Bhatia, Peter L. Bartlett, Anca D. Dragan, Jacob Steinhardt
This raises an interesting question whether learning is even possible in our setup, given that obtaining a generalizable estimate of utility $u^*$ might not be possible from finitely many samples.
no code implementations • NeurIPS 2020 • Kush Bhatia, Karthik Sridharan
In this setting, we study the problem of minimizing policy regret and provide non-constructive upper bounds on the minimax rate for the problem.
no code implementations • 27 Jul 2019 • Kush Bhatia, Yi-An Ma, Anca D. Dragan, Peter L. Bartlett, Michael. I. Jordan
We study the problem of robustly estimating the posterior distribution for the setting where observed data can be contaminated with potentially adversarial outliers.
no code implementations • 19 Mar 2019 • Arun Sai Suggala, Kush Bhatia, Pradeep Ravikumar, Prateek Jain
We provide a nearly linear time estimator which consistently estimates the true regression vector, even with $1-o(1)$ fraction of corruptions.
1 code implementation • NeurIPS 2018 • Aditya Kusupati, Manish Singh, Kush Bhatia, Ashish Kumar, Prateek Jain, Manik Varma
FastRNN addresses these limitations by adding a residual connection that does not constrain the range of the singular values explicitly and has only two extra scalar parameters.
no code implementations • 20 Dec 2018 • Dhruv Malik, Ashwin Pananjady, Kush Bhatia, Koulik Khamaru, Peter L. Bartlett, Martin J. Wainwright
We focus on characterizing the convergence rate of these methods when applied to linear-quadratic systems, and study various settings of driving noise and reward feedback.
no code implementations • NeurIPS 2018 • Kush Bhatia, Aldo Pacchiano, Nicolas Flammarion, Peter L. Bartlett, Michael. I. Jordan
In this paper, we study the problems of principle Generalized Eigenvector computation and Canonical Correlation Analysis in the stochastic setting.
no code implementations • 20 Nov 2018 • Kush Bhatia, Aldo Pacchiano, Nicolas Flammarion, Peter L. Bartlett, Michael. I. Jordan
In this paper, we study the problems of principal Generalized Eigenvector computation and Canonical Correlation Analysis in the stochastic setting.
no code implementations • 18 Oct 2018 • Sandy H. Huang, Kush Bhatia, Pieter Abbeel, Anca D. Dragan
In order to effectively interact with or supervise a robot, humans need to have an accurate mental model of its capabilities and how it acts.
Robotics
no code implementations • NeurIPS 2017 • Kush Bhatia, Prateek Jain, Parameswaran Kamalaruban, Purushottam Kar
We present the first efficient and provably consistent estimator for the robust regression problem.
no code implementations • 1 Jul 2016 • Kush Bhatia, Prateek Jain, Parameswaran Kamalaruban, Purushottam Kar
We illustrate our methods on synthetic datasets and show that our methods indeed are able to consistently recover the optimal parameters despite a large fraction of points being corrupted.
no code implementations • NeurIPS 2015 • Kush Bhatia, Himanshu Jain, Purushottam Kar, Manik Varma, Prateek Jain
The objective in extreme multi-label learning is to train a classifier that can automatically tag a novel data point with the most relevant subset of labels from an extremely large label set.
no code implementations • 9 Jul 2015 • Kush Bhatia, Himanshu Jain, Purushottam Kar, Prateek Jain, Manik Varma
Embedding based approaches make training and prediction tractable by assuming that the training label matrix is low-rank and hence the effective number of labels can be reduced by projecting the high dimensional label vectors onto a low dimensional linear subspace.
Extreme Multi-Label Classification General Classification +2
no code implementations • NeurIPS 2015 • Kush Bhatia, Prateek Jain, Purushottam Kar
In this work, we study a simple hard-thresholding algorithm called TORRENT which, under mild conditions on X, can recover w* exactly even if b corrupts the response variables in an adversarial manner, i. e. both the support and entries of b are selected adversarially after observing X and w*.