Toward Equation of Motion for Deep Neural Networks: Continuous-time Gradient Descent and Discretization Error Analysis
Dec 6, 2022
Speakers
Taiki Miyagawa
Speaker · 0 followers
About
We derive and solve an “Equation of Motion” (EoM) for deep neural networks (DNNs), a differential equation that precisely describes the discrete learning dynamics of DNNs. Differential equations are continuous but have played a prominent role even in the study of discrete optimization (gradient descent (GD) algorithms). However, there still exist gaps between differential equations and the actual learning dynamics of DNNs due to discretization error. In this paper, we start from gradient flow (GF) and derive a counter term that cancels the discretization error between GF and GD. As a result, we obtain EoM, a continuous differential equation that precisely describes the discrete learning dynamics of GD. We also derive discretization error to show to what extent EoM is precise. In addition, we apply EoM to two specific cases: scale- and translation-invariant layers. EoM highlights differences between continuous and discrete GD, indicating the importance of the counter term for a better description of the discrete learning dynamics of GD. Our experimental results support our theoretical findings.We derive and solve an “Equation of Motion” (EoM) for deep neural networks (DNNs), a differential equation that precisely describes the discrete learning dynamics of DNNs. Differential equations are continuous but have played a prominent role even in the study of discrete optimization (gradient descent (GD) algorithms). However, there still exist gaps between differential equations and the actual learning dynamics of DNNs due to discretization error. In this paper, we start from gradient flow (G…
Organizer
NeurIPS 2022
Account · 952 followers
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Improving 3D-aware Image Synthesis with A Geometry-aware Discriminator
Watch later
Zifan Shi, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
A Robust Phased Elimination Algorithm for Corruption-Tolerant Gaussian Process Bandits
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Multilabel Classification Framework for Approximate Nearest Neighbor Search
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Learning Probabilistic Topological Representations Using Discrete Morse Theory
Watch later
Xiaoling Hu, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Formulating Robustness Against Unforeseen Attacks
Watch later
Sophie Dai, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Compressed information is all you need: unifying intrinsic motivations and representation learning
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%