Research map / Architectures and efficiency / RNNs and normalization
A Theoretical Framework for Target Propagation
Alexander Meulemans, Francesco Carzaniga, Johan Suykens, João Sacramento, Benjamin F. Grewe
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 |
|---|---|---|---|
| Replace layer-wise reconstruction training with DRL for DTP feedback (ablation) | DTP · MNIST | test error: 2.39 \pm 0.19\% -> 2.21 \pm 0.09\% | Improved |
| Replace layer-wise reconstruction training with DRL for DTP feedback (ablation) | DTP · Frozen-MNIST | test error: 10.64 \pm 0.53\% -> 6.10 \pm 0.17\% | Improved |
| Replace layer-wise reconstruction training with DRL for DTP feedback (ablation) | DTP · Fashion-MNIST | test error: 11.49 \pm 0.23\% -> 11.22 \pm 0.20\% | Improved |
| Replace layer-wise reconstruction training with DRL for DTP feedback (ablation) | DTP · CIFAR10 | test error: 51.74 \pm 0.30 \% -> 50.80 \pm 0.43 \% | Improved |
| Add direct linear output-to-hidden feedback trained with DRL (ablation) | DTP · MNIST, Frozen-MNIST, Fashion-MNIST, CIFAR10 | test error: 2.39 \pm 0.19\% -> 2.04 \pm 0.08\%; 10.64 \pm 0.53\% -> 6.42 \pm 0.17\%; 11.49 \pm 0.23\% -> 11.11 \pm 0.35\%; 51.74 \pm | Mixed |
| Use an output reconstruction loop without DRL difference correction (ablation) | DDTP-linear · Frozen-MNIST | test error: 6.42 \pm 0.17\% -> 9.70 \pm 0.31\% | Hurt |
| Use a random hidden feedback layer instead of direct linear feedback (ablation) | DDTP-linear · Frozen-MNIST | test error: 6.42 \pm 0.17\% -> 5.11 \pm 0.19\% | Improved |
| Use a random hidden feedback layer instead of direct linear feedback (ablation) | DDTP-linear · CIFAR10 | test error: 50.36 \pm 0.26\% -> 51.94 \pm 0.49 \% | Hurt |
| Pre-train DTP feedback weights (ablation) | DTP · MNIST | test error: 2.39 \pm 0.19\% -> 2.26 \pm 0.18\% | Improved |
| Pre-train DTP feedback weights (ablation) | DTP · CIFAR10 | test error: 51.74 \pm 0.30 \% -> 52.20 \pm 0.50 \% | Hurt |
| Add DRL-trained direct linear feedback | BP · CIFAR10, small CNN | test error: 24.38 \pm 0.29\% -> 23.99 \pm 0.31\% | Improved |
Closest papers by meaning
- Towards Scaling Difference Target Propagation by Learning Backprop Targets (ICML 2022 · similarity 0.94)
- Efficient Target Propagation by Deriving Analytical Solution (AAAI 2024 · similarity 0.92)
- Forward Learning with Top-Down Feedback: Empirical and Analytical Characterization (ICLR 2024 · similarity 0.86)
- Improving equilibrium propagation without weight symmetry through Jacobian homeostasis (ICLR 2024 · similarity 0.85)
- Local Loss Optimization in the Infinite Width: Stable Parameterization of Predictive Coding Networks and Target Propagation (ICLR 2025 · similarity 0.85)
- Two Tales of Single-Phase Contrastive Hebbian Learning (ICML 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.
