Research map / Architectures and efficiency / RNNs and normalization
Hebbian Deep Learning Without Feedback
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 soft anti-Hebbian plasticity to SoftHebb (single layer) (ablation) | SoftHebb single conv layer · CIFAR-10 | test accuracy (%): 71.10 vs 62.69 for hard WTA | Helped |
| replace hard WTA plasticity with SoftHebb plasticity in a single conv layer (ablation) | Hebbian conv network · CIFAR-10 | test accuracy (%): +8.41 points (71.10 vs 62.69) | Helped |
| use SoftHebb learning instead of random weights in single conv layer (ablation) | SoftHebb · CIFAR-10 | test accuracy (%): 71.10 vs 63.93 | Helped |
| add more SoftHebb layers with growing width (deep stacking, width factor 4) (ablation) | SoftHebb multilayer · STL-10 | test accuracy (%): 76.23 vs 68.2 random weights | Helped |
| fine-tune SoftHebb network end-to-end with BP on labelled STL-10 data (ablation) | SoftHebb · STL-10 | test accuracy (%): 76.23 -> 78.5 | Helped |
| replace SoftHebb with hard WTA plasticity in multilayer network (ablation) | Hebbian conv network · STL-10 | top-1 accuracy (%): 54.8 vs 76.2 | Hurt |
| add depth with hard WTA plasticity (learned hard-WTA features do not improve with added layers) (ablation) | Hard WTA conv network · CIFAR-10 | test accuracy (%) | No clear effect |
| use RePU+Triangle forward activation instead of ReLU (ablation) | SoftHebb · CIFAR-10 | test accuracy (%): 80.31 vs 70.68 (ReLU) | Helped |
| use softmax as forward activation instead of RePU+Triangle (ablation) | SoftHebb · CIFAR-10 | test accuracy (%): 54.05 vs 80.31 | Hurt |
| use adaptive norm-dependent learning rate instead of linear decay (ablation) | SoftHebb · CIFAR-10 | convergence speed and robustness | Helped |
Closest papers by meaning
- Kernelized information bottleneck leads to biologically plausible 3-factor Hebbian learning in deep networks (NeurIPS 2020 · similarity 0.89)
- Rethinking Hebbian Principle: Low-Dimensional Structural Projection for Unsupervised Learning (NeurIPS 2025 · similarity 0.87)
- One-Time Soft Alignment Enables Resilient Learning without Weight Transport (NeurIPS 2026 · similarity 0.85)
- Hebbian Learning based Orthogonal Projection for Continual Learning of Spiking Neural Networks (ICLR 2024 · similarity 0.84)
- Two Tales of Single-Phase Contrastive Hebbian Learning (ICML 2024 · similarity 0.84)
- What Do Hebbian Learners Learn? Reduction Axioms for Iterated Hebbian Learning (AAAI 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.