Research map / Architectures and efficiency / RNNs and normalization
Scaling Direct Feedback Learning with Theoretical Guarantees
Paul Caillon, Erwan Fagnou, Blaise Delattre, Alexandre Allauzen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add BP calibration (T=1) to GrAPE (ablation) | GrAPE on AlexNet · CIFAR-100 | accuracy: 45.45 -> 62.63 | Improved |
| add BP calibration (T=1) to GrAPE (ablation) | GrAPE on VGG-16 · CIFAR-100 | accuracy: 32.40 -> 56.93 | Improved |
| add BP calibration (T=1) to DFA (ablation) | DFA on VGG-16 · CIFAR-100 | accuracy: 1.00 -> 29.40 | Improved |
| add BP calibration (T=1) to DFA (ablation) | DFA on AlexNet · CIFAR-100 | accuracy: 42.59 -> 49.37 | Improved |
| replace DFA fixed feedback with GrAPE learned aligned feedback | DFA on AlexNet · CIFAR-100 | accuracy: 42.59 -> 45.45 | Improved |
| replace DFA fixed feedback with GrAPE learned aligned feedback | DFA with T=1 calibration on VGG-16 · CIFAR-100 | accuracy: 29.40 -> 56.93 | Improved |
| use GrAPE instead of DFA, no calibration | DFA on ResNet-20 · CIFAR-100 | accuracy: 20.94 -> 24.28 | Improved |
| add BP calibration (T=1) to GrAPE (ablation) | GrAPE on ResNet-20 · CIFAR-100 | accuracy: 24.28 -> 64.82 | Improved |
| increase calibration interval from T=1 to T=50 (ablation) | GrAPE + calibration on ResNet-20 · CIFAR-100 | accuracy: 64.82 -> 36.79 | Hurt |
| add BP calibration (T=1) to GrAPE (ablation) | GrAPE on Transformer-Base macro · WikiText-103 | validation perplexity: 42.3 -> 33.1 | Improved |
| add BP calibration (T=1) to GrAPE (ablation) | GrAPE on Transformer-Base micro · WikiText-103 | validation perplexity: 81.1 -> 67.3 | Improved |
| increase calibration interval from T=1 to T=10 (ablation) | GrAPE + calibration on Transformer-Base micro · WikiText-103 | validation perplexity: 67.3 -> 78.2 | Hurt |
Closest papers by meaning
- Overcoming Rank Collapse in Feedback Alignment (NeurIPS 2026 · similarity 0.88)
- One-Time Soft Alignment Enables Resilient Learning without Weight Transport (NeurIPS 2026 · similarity 0.87)
- Forward Learning with Top-Down Feedback: Empirical and Analytical Characterization (ICLR 2024 · similarity 0.85)
- Implicit Regularization in Feedback Alignment Learning Mechanisms for Neural Networks (ICML 2024 · similarity 0.85)
- Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks (NeurIPS 2025 · similarity 0.83)
- DFA-GNN: Forward Learning of Graph Neural Networks by Direct Feedback Alignment (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
