Research map / Architectures and efficiency / RNNs and normalization
Assessing the Scalability of Biologically-Motivated Deep Learning Algorithms and Architectures
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 |
|---|---|---|---|
| alternate forward and inverse optimization instead of parallel optimization (ablation) | DTP · MNIST, fully connected | test error (%): 2.86 -> 1.83 | Helped |
| alternate forward and inverse optimization instead of parallel optimization (ablation) | DTP · CIFAR-10, fully connected | test error (%): 59.14 -> 42.32 | Helped |
| alternate forward and inverse optimization instead of parallel optimization (ablation) | DTP · ImageNet, locally connected | Top-1 test error (%): 98.34 -> 99.36 | Hurt |
| remove gradient propagation from DTP (ablation) | SDTP · MNIST, fully connected | test error (%): 1.83 -> 3.52 | Hurt |
| remove gradient propagation from DTP (ablation) | SDTP · CIFAR-10, fully connected | test error (%): 42.32 -> 55.32 | Hurt |
| add auxiliary output to SDTP (ablation) | SDTP · CIFAR-10, fully connected | test error (%): 55.32 -> 47.11 | Helped |
| add auxiliary output to SDTP (ablation) | SDTP · CIFAR-10, locally connected | test error (%): 46.63 -> 40.05 | Helped |
| increase auxiliary output size (ablation) | AO-SDTP · CIFAR-10 | test performance | Helped |
| use auxiliary output targets rather than basic SDTP targets (ablation) | SDTP · CIFAR-10 | test error (%) | Helped |
| use larger receptive fields or fully connected inverse mappings (ablation) | DTP · CIFAR-10, locally connected | test error (%) | No clear effect |
| replace pooling with strided locally connected layers (ablation) | TP architectures · Image classification | performance | Helped |
| use shared-weight convolutions rather than locally connected layers (ablation) | BP · CIFAR-10, locally connected | test error (%): 32.41 -> 31.87 | Helped |
Closest papers by meaning
- Biologically-Plausible Learning Algorithms Can Scale to Large Datasets (ICLR 2019 · similarity 0.89)
- Efficient Target Propagation by Deriving Analytical Solution (AAAI 2024 · similarity 0.88)
- Overcoming Rank Collapse in Feedback Alignment (NeurIPS 2026 · similarity 0.87)
- $\mu$PC: Scaling Predictive Coding to 100+ Layer Networks (NeurIPS 2025 · similarity 0.86)
- Improving equilibrium propagation without weight symmetry through Jacobian homeostasis (ICLR 2024 · similarity 0.85)
- Counter-Current Learning: A Biologically Plausible Dual Network Approach for Deep Learning (NeurIPS 2024 · similarity 0.85)
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.