Timezone: »
We study how robust current ImageNet models are to distribution shifts arising from natural variations in datasets. Most research on robustness focuses on synthetic image perturbations (noise, simulated weather artifacts, adversarial examples, etc.), which leaves open how robustness on synthetic distribution shift relates to distribution shift arising in real data. Informed by an evaluation of 204 ImageNet models in 213 different test conditions, we find that there is often little to no transfer of robustness from current synthetic to natural distribution shift. Moreover, most current techniques provide no robustness to the natural distribution shifts in our testbed. The main exception is training on larger and more diverse datasets, which in multiple cases increases robustness, but is still far from closing the performance gaps. Our results indicate that distribution shifts arising in real data are currently an open research problem.
Author Information
Rohan Taori (Stanford University)
Achal Dave (Carnegie Mellon University)
Vaishaal Shankar (UC Berkeley)
Nicholas Carlini (Google)
Benjamin Recht (UC Berkeley)
Ludwig Schmidt (Toyota Research Institute)
Related Events (a corresponding poster, oral, or spotlight)
-
2020 Poster: Measuring Robustness to Natural Distribution Shifts in Image Classification »
Wed Dec 9th 05:00 -- 07:00 AM Room Poster Session 2
More from the Same Authors
-
2020 Poster: On Adaptive Attacks to Adversarial Example Defenses »
Florian Tramer · Nicholas Carlini · Wieland Brendel · Aleksander Madry -
2020 Oral: Hogwild!: A Lock-Free Approach to Parallelizing Stochastic Gradient Descent »
Benjamin Recht · Christopher Ré · Stephen Wright · Feng Niu -
2019 Poster: Model Similarity Mitigates Test Set Overuse »
Horia Mania · John Miller · Ludwig Schmidt · Moritz Hardt · Benjamin Recht -
2019 Poster: Unlabeled Data Improves Adversarial Robustness »
Yair Carmon · Aditi Raghunathan · Ludwig Schmidt · John Duchi · Percy Liang -
2019 Poster: A Meta-Analysis of Overfitting in Machine Learning »
Rebecca Roelofs · Vaishaal Shankar · Benjamin Recht · Sara Fridovich-Keil · Moritz Hardt · John Miller · Ludwig Schmidt -
2019 Poster: Finite-time Analysis of Approximate Policy Iteration for the Linear Quadratic Regulator »
Karl Krauth · Stephen Tu · Benjamin Recht -
2019 Poster: Certainty Equivalence is Efficient for Linear Quadratic Control »
Horia Mania · Stephen Tu · Benjamin Recht -
2018 Poster: Simple random search of static linear policies is competitive for reinforcement learning »
Horia Mania · Aurelia Guy · Benjamin Recht -
2018 Poster: Regret Bounds for Robust Adaptive Control of the Linear Quadratic Regulator »
Sarah Dean · Horia Mania · Nikolai Matni · Benjamin Recht · Stephen Tu -
2017 Workshop: OPT 2017: Optimization for Machine Learning »
Suvrit Sra · Sashank J. Reddi · Alekh Agarwal · Benjamin Recht -
2017 Poster: The Marginal Value of Adaptive Gradient Methods in Machine Learning »
Ashia C Wilson · Becca Roelofs · Mitchell Stern · Nati Srebro · Benjamin Recht -
2017 Oral: The Marginal Value of Adaptive Gradient Methods in Machine Learning »
Ashia C Wilson · Becca Roelofs · Mitchell Stern · Nati Srebro · Benjamin Recht -
2017 Oral: Test of Time Award »
ali rahimi · Benjamin Recht -
2016 Poster: The Power of Adaptivity in Identifying Statistical Alternatives »
Kevin Jamieson · Daniel Haas · Benjamin Recht -
2016 Poster: Cyclades: Conflict-free Asynchronous Machine Learning »
Xinghao Pan · Maximilian Lam · Stephen Tu · Dimitris Papailiopoulos · Ce Zhang · Michael Jordan · Kannan Ramchandran · Christopher Ré · Benjamin Recht -
2015 Poster: Parallel Correlation Clustering on Big Graphs »
Xinghao Pan · Dimitris Papailiopoulos · Samet Oymak · Benjamin Recht · Kannan Ramchandran · Michael Jordan