Research map / Architectures and efficiency / RNNs and normalization
Learning From the Past with Cascading Eligibility Traces
Tokiniaina Raharison Ralambomihanta, Ivan Anokhin, Roman Pogodin, Samira Ebrahimi Kahou, Jonathan Cornford, Blake Aaron Richards
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 |
|---|---|---|---|
| increase number of CET states (vs single-state standard ET) (ablation) | CET with delayed error signal · MNIST (behavioral delays, 3-layer MLP) | accuracy | Improved |
| increase number of CET states (vs single-state standard ET) (ablation) | CET with delayed error signal · CIFAR-10 (behavioral delays, small CNN) | accuracy | Improved |
| increase number of CET states (ablation) | Actor-Critic actor with CET · CartPole and LunarLander (behavioral delays) | mean episodic return | Improved |
| increase number of CET states (ablation) | Actor-Critic actor with CET · MinAtar/SpaceInvaders (behavioral delays) | mean episodic return | Mixed |
| increase number of CET states (ablation) | CET gradient approximation · CIFAR-10 (delays 1, 2, 4 s) | cosine similarity to true gradient | Improved |
| increase delay with higher-order CETs (ablation) | CET gradient approximation · CIFAR-10 (long delay) | cosine similarity to true gradient | Hurt |
| increase CET order with stacked retrograde delays (minutes) (ablation) | MLP on MNIST and CNN on CIFAR-10 · retrograde delays, stacked across layers | test accuracy | Improved |
| increase CET order with stacked retrograde delays (ablation) | MLP on MNIST and CNN on CIFAR-10 · retrograde delays, stacked across layers | training speed | Improved |
| increase CET order with retrograde delays (ablation) | CET gradient approximation · MNIST and CIFAR-10, retrograde | cosine similarity to true gradient | Mixed |
| increase CET states with retrograde delays (ablation) | Actor-Critic actor with CET · CartPole and LunarLander (retrograde) | episodic return | Improved |
| increase network depth under stacked retrograde delays | CETs · retrograde delays | gradient estimate and learning | Hurt |
| apply CET at retrograde timescale on SpaceInvaders | CET · MinAtar/SpaceInvaders (8 s and 120 s delay) | episodic return | Hurt |
Closest papers by meaning
- Model Based Inference of Synaptic Plasticity Rules (NeurIPS 2024 · similarity 0.85)
- Dis-inhibitory neuronal circuits can control the sign of synaptic plasticity (NeurIPS 2023 · similarity 0.85)
- Biologically-plausible backpropagation through arbitrary timespans via local neuromodulators (NeurIPS 2022 · similarity 0.85)
- Discovering plasticity rules that organize and maintain neural circuits (NeurIPS 2024 · similarity 0.84)
- Learning efficient task-dependent representations with synaptic plasticity (NeurIPS 2020 · similarity 0.84)
- Feedback control guides credit assignment in recurrent neural networks (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.
