← Open in Shiplog

Interpretability and training dynamics

theme machine learning developing

Layer-wise interpretability via identity initialization, implicit bias of gradient regularization, and selective forgetting / unlearning.

Findings (3)

Papers (3)
- Understanding Gradient Regularization in Deep Learning: Efficient Finite-Difference Computation and Implicit Bias2023· ICML
- Layer-Wise Interpretation of Deep Neural Networks Using Idneity Initialization2021· IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
- Selective Forgetting of Deep Networks at a Finer Level than Samples2020· AAAI RSEML

Connections

This topic …

usesOrthogonal initialization

contrasts withDynamical isometry