Picture for Roy Frostig

Roy Frostig

Learning from many trajectories

Mar 31, 2022
Stephen Tu, Roy Frostig, Mahdi Soltanolkotabi

Figure 1 for Learning from many trajectories

Figure 2 for Learning from many trajectories

Figure 3 for Learning from many trajectories

Access Paper or Ask Questions

Decomposing reverse-mode automatic differentiation

May 20, 2021
Roy Frostig, Matthew J. Johnson, Dougal Maclaurin, Adam Paszke, Alexey Radul

* Presented at the LAFI 2021 workshop at POPL, 17 January 2021

Access Paper or Ask Questions

The advantages of multiple classes for reducing overfitting from test set reuse

May 24, 2019
Vitaly Feldman, Roy Frostig, Moritz Hardt

Figure 1 for The advantages of multiple classes for reducing overfitting from test set reuse

Access Paper or Ask Questions

Toward Deeper Understanding of Neural Networks: The Power of Initialization and a Dual View on Expressivity

May 19, 2017
Amit Daniely, Roy Frostig, Yoram Singer

Figure 1 for Toward Deeper Understanding of Neural Networks: The Power of Initialization and a Dual View on Expressivity

Figure 2 for Toward Deeper Understanding of Neural Networks: The Power of Initialization and a Dual View on Expressivity

Figure 3 for Toward Deeper Understanding of Neural Networks: The Power of Initialization and a Dual View on Expressivity

Access Paper or Ask Questions

Estimation from Indirect Supervision with Linear Moments

Aug 10, 2016
Aditi Raghunathan, Roy Frostig, John Duchi, Percy Liang

Figure 1 for Estimation from Indirect Supervision with Linear Moments

Figure 2 for Estimation from Indirect Supervision with Linear Moments

Figure 3 for Estimation from Indirect Supervision with Linear Moments

Figure 4 for Estimation from Indirect Supervision with Linear Moments

* 12 pages, 7 figures, extended and updated version of our paper appearing in ICML 2016

Access Paper or Ask Questions