Research map / Architectures and efficiency / RNNs and normalization
Adaptive recurrent vision performs zero-shot computation scaling to unseen difficulty levels
Vijay Veerabadran, Srinivas Ravishankar, Yuan Tang, Ritik Raina, Virginia R. de Sa
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 stability-based halting with learned ACT halting (ablation) | hConvGRU · PathFinder-21 | accuracy (%): 50.0 -> 82.63 | Improved |
| replace stability-based halting with learned ACT halting (LocRNN) (ablation) | hConvGRU · PathFinder-24 | accuracy (%): 50.0 -> 85.81 | Improved |
| replace stability-based halting with LocRNN ACT (ablation) | hConvGRU · Mazes-19 | % mazes solved: 50.26 ± 6.04 -> 86.83 ± 2.94 | Improved |
| replace stability-based halting with LocRNN ACT (ablation) | hConvGRU · Mazes-25 | % mazes solved: 21.36 ± 2.82 -> 49.99 ± 4.48 | Improved |
| replace fixed-iteration hConvGRU with LocRNN (ACT) | hConvGRU · PathFinder-21 | accuracy (%): 64.21 -> 92.89 ± 0.9 | Improved |
| replace ConvGRU with LocRNN (interneuron population) | ConvGRU (ACT) · PathFinder-Mixed | accuracy (%): 95.26 -> 97.13 | Improved |
| replace ConvGRU with LocRNN | ConvGRU (ACT) · PathFinder-24 | accuracy (%): 74.14 ± 6.52 -> 85.81 ± 5.57 | Improved |
| replace ConvGRU with LocRNN | ConvGRU (ACT) · Mazes-Mixed | % mazes solved: 98.4 -> 98.4 | No clear effect |
| replace ConvGRU with LocRNN | ConvGRU (ACT) · Mazes-25 | % mazes solved: 46.93 ± 4.2 -> 49.99 ± 4.48 | No clear effect |
| use weight-tied residual network instead of gated recurrent cells | R-ResNet-30 (ACT) · PathFinder-Mixed | accuracy (%): 95.26 -> 49.37 | Hurt |
| use weight-tied residual network instead of gated recurrent cells | R-ResNet-30 (ACT) · Mazes-Mixed | % mazes solved: 98.4 -> 0.0 | Hurt |
| add ACT halting to ConvGRU and extrapolate with more iterations vs fixed iterations | ConvGRU · Mazes-19 | % mazes solved: 16.2 ± 2.66 -> 75.1 ± 11.96 | Improved |
Closest papers by meaning
- DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026 · similarity 0.87)
- RTify: Aligning Deep Neural Networks with Human Behavioral Decisions (NeurIPS 2024 · similarity 0.86)
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.86)
- Diversity Matters: Revisiting Test-Time Compute in Vision-Language Models (ICML 2026 · similarity 0.86)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.86)
- Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (NeurIPS 2024 · similarity 0.86)
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.
