Measuring Robustness to Natural Distribution Shifts in Image Classification
Rohan Taori, Achal Dave, Vaishaal Shankar, Nicholas Carlini, Benjamin Recht, Ludwig Schmidt
Advances in Neural Information Processing Systems (NeurIPS), 2020. Spotlight.
Code & Data  /
We study how robust current ImageNet models are to distribution shifts arising from natural variations in datasets.
Informed by a very large experimental testbed, we find that there is often little to no transfer of robustness from current synthetic to natural distribution shift.
Transposer: Universal Texture Synthesis Using Feature Maps as Transposed Convolution Filter
Guilin Liu, Rohan Taori, Ting-Chun Wang, Zhiding Yu, Shiqiu Liu, Fitsum Reda, Karan Sapra, Andrew Tao, Bryan Catanzaro
Short Video  /
We propose a new learned framework for texture synthesis that allows for fast, high-resolution, and diverse generation for arbitrary textures.
We utilize the transpose convolution operation as a soft way to stitch texels together.
Our method achieves state-of-the-art quality on various metrics.
Targeted Adversarial Examples for Black Box Audio Systems
Rohan Taori, Amog Kamsetty, Brenton Chu, Nikita Vemuri
2019 IEEE Deep Learning and Security Workshops.
We adopt a black-box approach to generate adversarial examples for a speech-to-text system.
Autoregressive Models: What Are They Good For?
Murtaza Dalal, Alexander Li, Rohan Taori
We investigate the use of autoregressive models as density estimators and find they are much less reliable than previously thought.
Data Science Class
I ran or was heavily involved with Machine Learning @ Berkeley's Data Science Class for a number of semesters. You can find complete course content (lecture slides, demos, & homeworks) from those semesters online:
Some lectures I gave:
Machine Learning Workshops & Reading Groups
I also organized with my friends a series of public workshops exploring topics in deep learning. Some of the workshops and other reading groups I've given:
- Intro to Deep Learning Workshop I gave with Sajel Shah at CalHacks 5.0.
- Intro to Reinforcement Learning Workshop I gave with Brenton Chu.
- Autogressive models presentation: MADE, PixelCNN, and more [ask for slides if interested].
- Brief tutorial on variational inference [ask for slides if interested].
- Inception + ResNet - common neural network architectures and why they're popular [ask for slides if interested].
- Robust machine learning presentations: adversarial examples, distribution shift, and more [ask for slides if interested].