Timezone: »

What You See is What You Get: Principled Deep Learning via Distributional Generalization
Bogdan Kulynych · Yao-Yuan Yang · Yaodong Yu · Jarosław Błasiok · Preetum Nakkiran

Tue Nov 29 02:00 PM -- 04:00 PM (PST) @ Hall J #332

Having similar behavior at training time and test time—what we call a “What You See Is What You Get” (WYSIWYG) property—is desirable in machine learning. Models trained with standard stochastic gradient descent (SGD), however, do not necessarily have this property, as their complex behaviors such as robustness or subgroup performance can differ drastically between training and test time. In contrast, we show that Differentially-Private (DP) training provably ensures the high-level WYSIWYG property, which we quantify using a notion of distributional generalization. Applying this connection, we introduce new conceptual tools for designing deep-learning methods by reducing generalization concerns to optimization ones: to mitigate unwanted behavior at test time, it is provably sufficient to mitigate this behavior on the training data. By applying this novel design principle, which bypasses “pathologies” of SGD, we construct simple algorithms that are competitive with SOTA in several distributional-robustness applications, significantly improve the privacy vs. disparate impact trade-off of DP-SGD, and mitigate robust overfitting in adversarial training. Finally, we also improve on theoretical bounds relating DP, stability, and distributional generalization.

Author Information

Bogdan Kulynych (EPFL SPRING Lab)

PhD candidate in Computer Science at EPFL, Fellow at Harvard SEAS. B.Sc. from Kyiv Mohyla Academy in Ukraine. Formerly an intern at Google, CERN. I study privacy, security, reliability, and broader societal harms of algorithmic systems.

Yao-Yuan Yang (DeepMind)
Yaodong Yu (University of California, Berkeley)
Jarosław Błasiok
Preetum Nakkiran (Harvard)

More from the Same Authors