Atish Agarwala

32 papers A* 10Journal 22

Year	Rank	Type	Title / Venue / Authors
2026	J	`jnl`	High dimensional theory of two-phase optimizers. CoRR Atish Agarwala
2026	J	`jnl`	What do near-optimal learning rate schedules look like? CoRR Hiroki Naganuma, Atish Agarwala, Priya Kasimbeg, George E. Dahl
2025	A*	`conf`	Avoiding spurious sharpness minimization broadens applicability of SAM. ICML Sidak Pal Singh, Hossein Mobahi, Atish Agarwala, Yann N. Dauphin
2025	J	`jnl`	Avoiding spurious sharpness minimization broadens applicability of SAM. CoRR Sidak Pal Singh, Hossein Mobahi, Atish Agarwala, Yann N. Dauphin
2025	A*	`conf`	Exact risk curves of signSGD in High-Dimensions: quantifying preconditioning and noise-compression effects. ICML Ke Liang Xiao, Noah Marshall, Atish Agarwala, Elliot Paquette
2025	J	`jnl`	How far away are truly hyperparameter-free learning algorithms? CoRR Priya Kasimbeg, Vincent Roulet, Naman Agarwal, Sourabh Medapati, Fabian Pedregosa, Atish Agarwala, George E. Dahl
2025	J	`jnl`	How far away are truly hyperparameter-free learning algorithms? Trans. Mach. Learn. Res. Priya Kasimbeg, Vincent Roulet, Naman Agarwal, Sourabh Medapati, Fabian Pedregosa, Atish Agarwala, George E. Dahl
2025	J	`jnl`	Per-example gradients: a new frontier for understanding and improving optimizers. CoRR Vincent Roulet, Atish Agarwala
2025	A*	`conf`	Scaling Collapse Reveals Universal Dynamics in Compute-Optimally Trained Neural Networks. ICML Shikai Qiu, Lechao Xiao, Andrew Gordon Wilson, Jeffrey Pennington, Atish Agarwala
2025	J	`jnl`	Scaling Collapse Reveals Universal Dynamics in Compute-Optimally Trained Neural Networks. CoRR Shikai Qiu, Lechao Xiao, Andrew Gordon Wilson, Jeffrey Pennington, Atish Agarwala
2025	A*	`conf`	To Clip or not to Clip: the Dynamics of SGD with Gradient Clipping in High-Dimensions. ICLR Noah Marshall, Ke Liang Xiao, Atish Agarwala, Elliot Paquette
2024	J	`jnl`	A Clipped Trip: the Dynamics of SGD with Gradient Clipping in High-Dimensions. CoRR Noah Marshall, Ke Liang Xiao, Atish Agarwala, Elliot Paquette
2024	J	`jnl`	Exact Risk Curves of signSGD in High-Dimensions: Quantifying Preconditioning and Noise-Compression Effects. CoRR Ke Liang Xiao, Noah Marshall, Atish Agarwala, Elliot Paquette
2024	J	`jnl`	Feature learning as alignment: a structural property of gradient descent in non-linear neural networks. Trans. Mach. Learn. Res. Daniel Beaglehole, Ioannis Mitliagkas, Atish Agarwala
2024	J	`jnl`	Gradient descent induces alignment between weights and the empirical NTK for deep non-linear networks. CoRR Daniel Beaglehole, Ioannis Mitliagkas, Atish Agarwala
2024	J	`jnl`	High dimensional analysis reveals conservative sharpening and a stochastic edge of stability. CoRR Atish Agarwala, Jeffrey Pennington
2024	J	`jnl`	Neglected Hessian component explains mysteries in Sharpness regularization. CoRR Yann N. Dauphin, Atish Agarwala, Hossein Mobahi
2024	A*	`conf`	Neglected Hessian component explains mysteries in sharpness regularization. NeurIPS Yann N. Dauphin, Atish Agarwala, Hossein Mobahi
2024	A*	`conf`	Stepping on the Edge: Curvature Aware Learning Rate Tuners. NeurIPS Vincent Roulet, Atish Agarwala, Jean-Bastien Grill, Grzegorz Swirszcz, Mathieu Blondel, Fabian Pedregosa
2024	J	`jnl`	Stepping on the Edge: Curvature Aware Learning Rate Tuners. CoRR Vincent Roulet, Atish Agarwala, Jean-Bastien Grill, Grzegorz Swirszcz, Mathieu Blondel, Fabian Pedregosa
2023	J	`jnl`	On the Interplay Between Stepsize Tuning and Progressive Sharpening. CoRR Vincent Roulet, Atish Agarwala, Fabian Pedregosa
2023	A*	`conf`	SAM operates far from home: eigenvalue regularization as a dynamical phenomenon. ICML Atish Agarwala, Yann N. Dauphin
2023	J	`jnl`	SAM operates far from home: eigenvalue regularization as a dynamical phenomenon. CoRR Atish Agarwala, Yann N. Dauphin
2023	A*	`conf`	Second-order regression models exhibit progressive sharpening to the edge of stability. ICML Atish Agarwala, Fabian Pedregosa, Jeffrey Pennington
2023	J	`jnl`	Temperature check: theory and practice for training models with softmax-cross-entropy losses. Trans. Mach. Learn. Res. Atish Agarwala, Samuel Stern Schoenholz, Jeffrey Pennington, Yann N. Dauphin
2022	A*	`conf`	Deep equilibrium networks are sensitive to initialization statistics. ICML Atish Agarwala, Samuel S. Schoenholz
2022	J	`jnl`	Deep equilibrium networks are sensitive to initialization statistics. CoRR Atish Agarwala, Samuel S. Schoenholz
2022	J	`jnl`	Second-order regression models exhibit progressive sharpening to the edge of stability. CoRR Atish Agarwala, Fabian Pedregosa, Jeffrey Pennington
2021	A*	`conf`	One Network Fits All? Modular versus Monolithic Task Formulations in Neural Networks. ICLR Atish Agarwala, Abhimanyu Das, Brendan Juba, Rina Panigrahy, Vatsal Sharan, Xin Wang, Qiuyi Zhang
2021	J	`jnl`	One Network Fits All? Modular versus Monolithic Task Formulations in Neural Networks. CoRR Atish Agarwala, Abhimanyu Das, Brendan Juba, Rina Panigrahy, Vatsal Sharan, Xin Wang, Qiuyi Zhang
2020	J	`jnl`	Learning the gravitational force law and other analytic functions. CoRR Atish Agarwala, Abhimanyu Das, Rina Panigrahy, Qiuyi Zhang
2020	J	`jnl`	Temperature check: theory and practice for training models with softmax-cross-entropy losses. CoRR Atish Agarwala, Jeffrey Pennington, Yann N. Dauphin, Samuel S. Schoenholz