Research map / Architectures and efficiency / CNN design
Task-Driven Convolutional Recurrent Models of the Visual System
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 the Reciprocal Gated Cell, combining gating and bypassing | 6-layer CNN · ImageNet | accuracy | Helped |
| Use vanilla RNN or LSTM cells instead of the feedforward architecture | 6-layer CNN · ImageNet | accuracy | No clear effect |
| Increase feedforward model width or depth rather than adding standard RNN cells | 6-layer CNN · ImageNet | accuracy | Helped |
| Optimize LSTM training and structural hyperparameters | LSTM ConvRNN · ImageNet | accuracy | Helped |
| Replace the searched recurrent architecture with a Time Decay model using learned time constants (ablation) | ResNet-18-based backbone · ImageNet | accuracy | No clear effect |
| Use the fully unrolled searched ConvRNN rather than its minimally unrolled control (ablation) | Random model from the initial search phase · ImageNet | accuracy | No clear effect |
| Add long-range feedback connections (ablation) | Searched ConvRNN · ImageNet | accuracy | Mixed |
| Use task-optimized ConvRNN features instead of feedforward features to predict neural dynamics | Model of macaque V4 and IT responses · Held-out neural response images | correlation with neural responses | Helped |
| Replace ConvRNN dynamics with learned time constants (ablation) | Time Decay model of neural responses · Macaque V4, pIT, cIT/aIT | neural response prediction | Hurt |
Closest papers by meaning
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.91)
- Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs (NeurIPS 2019 · similarity 0.90)
- Stable and expressive recurrent vision models (NeurIPS 2020 · similarity 0.90)
- The Uncanny Similarity of Recurrence and Depth (ICLR 2022 · similarity 0.89)
- Self-Attention-Based Contextual Modulation Improves Neural System Identification (ICLR 2025 · similarity 0.87)
- RTify: Aligning Deep Neural Networks with Human Behavioral Decisions (NeurIPS 2024 · similarity 0.87)
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.