Return Landscape

Return landscapes, which map policy parameters to performance, are a crucial area of study in reinforcement learning and federated learning. Current research focuses on understanding and manipulating these landscapes to improve algorithm stability and generalization, particularly in noisy environments and under privacy constraints. This involves developing algorithms that navigate away from unstable regions of the landscape and promote flatter minima, often employing techniques like Sharpness Aware Minimization (SAM). These advancements aim to enhance the robustness and efficiency of machine learning models, leading to more reliable and privacy-preserving applications.

Papers

September 26, 2023

Policy Optimization in a Noisy Neighborhood: On Return Landscapes in Continuous Control
Nate Rahn, Pierluca D'Oro, Harley Wiltzer, Pierre-Luc Bacon, Marc G. Bellemare
Native Robustness Policy OpTimization Deep Reinforcement Continuous Control Policy Parameterization Policy Space Return Landscape

May 19, 2023

Dynamic Regularized Sharpness Aware Minimization in Federated Learning: Approaching Global Consistency and Smooth Landscape
Yan Sun, Li Shen, Shixiang Chen, Liang Ding, Dacheng Tao
Optimization Purpose Sharpness Aware Minimization Local Optimum Global Consistency Generalization Benefit Return Landscape

May 1, 2023

Towards the Flatter Landscape and Better Generalization in Federated Learning under Client-level Differential Privacy
Yifan Shi, Kang Wei, Li Shen, Yingqi Liu, Xueqian Wang, Bo Yuan, Dacheng Tao
Better Generalization Differentially Private Perturbation Robustness Differentially Private Federated Learning Client Level Differential Privacy Return Landscape

March 20, 2023

Make Landscape Flatter in Differentially Private Federated Learning
Yifan Shi, Yingqi Liu, Kang Wei, Li Shen, Xueqian Wang, Dacheng Tao
Private Federated Learning Differentially Private Differentially Private Federated Learning Return Landscape

Return Landscape

Papers

Policy Optimization in a Noisy Neighborhood: On Return Landscapes in Continuous Control

Dynamic Regularized Sharpness Aware Minimization in Federated Learning: Approaching Global Consistency and Smooth Landscape

Towards the Flatter Landscape and Better Generalization in Federated Learning under Client-level Differential Privacy

Make Landscape Flatter in Differentially Private Federated Learning