Backward Error Analysis
Cross-source consensus on Backward Error Analysis from 1 sources and 4 claims.
1 sources · 4 claims
How it works
Comparisons
Evidence quality
Highlighted claims
- The theoretical analysis uses backward error analysis to interpret a discrete optimizer as following a modified objective. — Implicit Regularization of Mini-Batch Training in Graph Neural Networks
- For sampled-subgraph mini-batch training, the modified loss is not simply the full-graph loss. — Implicit Regularization of Mini-Batch Training in Graph Neural Networks
- For full-graph gradient descent, the modified objective adds an epsilon-scaled squared-gradient correction to the full-graph loss. — Implicit Regularization of Mini-Batch Training in Graph Neural Networks
- The formal backward error analysis is limited to vanilla SGD, while Adam lacks a matching stochastic graph mini-batch derivation. — Implicit Regularization of Mini-Batch Training in Graph Neural Networks