Research map / Architectures and efficiency / RNNs and normalization
GAIT-prop: A biologically plausible learning rule derived from backpropagation of error
Nasir Ahmad, Marcel A. J. van Gerven, Luca Ambrogioni
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 weak orthogonal regularizer (lambda 0 -> 0.1) at lr 1e-4 (ablation) | GAIT-prop · MNIST train, 4 hidden layers | final train accuracy: 26.36 -> 100.00 | Improved |
| add orthogonal regularizer (lambda 0 -> 10) at lr 1e-3 (ablation) | GAIT-prop · MNIST train, 4 hidden layers | final train accuracy: 17.94 -> 93.79 | Improved |
| add orthogonal regularizer (lambda 0 -> 0.1) at lr 1e-4 (ablation) | backpropagation · MNIST train, 4 hidden layers | final train accuracy: 100.00 -> 100.00 | No clear effect |
| add orthogonal regularizer (lambda 0 -> 0.1) at lr 1e-3 (ablation) | target propagation · MNIST train, 4 hidden layers | final train accuracy: 10.22 -> 9.87 | No clear effect |
| raise orthogonal regularizer to lambda 1000 (ablation) | GAIT-prop · MNIST train, 4 hidden layers | final train accuracy: 100.00 -> 96.99 | Hurt |
| lower Adam lr from 1e-3 to 1e-5 for target propagation with lambda 10 (ablation) | target propagation · MNIST train, 4 hidden layers | final train accuracy: 11.45 -> 92.16 | Improved |
| replace GAIT-prop/BP with target propagation in non-linear networks | target propagation · MNIST non-linear networks | accuracy | Hurt |
| compare GAIT-prop with backpropagation across depths and tasks | GAIT-prop · MNIST, KMNIST, Fashion-MNIST, up to 4 hidden layers | test accuracy | No clear effect |
| increase depth to six or eight hidden layers | GAIT-prop · MNIST, 6 and 8 hidden layers | test accuracy | No clear effect |
| use target propagation in linear networks | target propagation · linear network, 4 hidden layers, no regularizer | accuracy | No clear effect |
| reduce hidden layer widths with auxiliary units at every layer | GAIT-prop · MNIST variable-width networks | accuracy | No clear effect |
Closest papers by meaning
- A Theoretical Framework for Target Propagation (NeurIPS 2020 · similarity 0.90)
- Efficient Target Propagation by Deriving Analytical Solution (AAAI 2024 · similarity 0.90)
- Towards Scaling Difference Target Propagation by Learning Backprop Targets (ICML 2022 · similarity 0.89)
- Forward Learning with Top-Down Feedback: Empirical and Analytical Characterization (ICLR 2024 · similarity 0.85)
- Overcoming Rank Collapse in Feedback Alignment (NeurIPS 2026 · similarity 0.85)
- Improving equilibrium propagation without weight symmetry through Jacobian homeostasis (ICLR 2024 · similarity 0.84)
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.
