Research map / Architectures and efficiency / CNN design
A Unified Theory of Early Visual Representations from Retina to Cortex through Anatomically Constrained Deep CNNs
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 |
|---|---|---|---|
| Reduce retinal output channels from 32 to 1 | Retina-net + VVS-net · CIFAR-10 visual recognition and receptive-field formation | Retinal RF geometry | Helped |
| Reduce retinal output channels from 32 to 1 | Retina-net + VVS-net · CIFAR-10 visual recognition and receptive-field formation | Downstream RF orientation | Helped |
| Increase VVS-net depth | Retina-net + VVS-net · CIFAR-10 | Retinal response linearity | Helped |
| Increase VVS-net depth | Retina-net + VVS-net · CIFAR-10 | Image reconstruction from retinal representation | Helped |
| Increase VVS-net depth | Retina-net + VVS-net · CIFAR-10 | Linear separability of object classes at retina output | Hurt |
| Increase VVS-net depth | Retina-net + VVS-net · CIFAR-10 | Image classification accuracy | Helped |
| Increase retinal bottleneck channels | Retina-net + VVS-net · CIFAR-10 | Image classification accuracy | Helped |
| Increase retinal bottleneck channels | Retina-net + VVS-net · CIFAR-10 | Retinal response linearity | Hurt |
| Use an untied locally connected retina-net rather than convolutional parameter tying (ablation) | Retina-net + VVS-net · CIFAR-10 | Center-surround RF emergence | No clear effect |
| Add local response normalization at every layer (ablation) | Retina-net + VVS-net · CIFAR-10 | Center-surround retinal and oriented V1 RFs | No clear effect |
| Add local response normalization at every layer (ablation) | Retina-net + VVS-net · CIFAR-10 | Image classification performance | Hurt |
| Replace the trained nonlinear retina with its best linear convolutional approximation (ablation) | Deep VVS-net with N_BN=1 · CIFAR-10 | Linear separability in the first VVS-net layer | Helped |
Closest papers by meaning
- From deep learning to mechanistic understanding in neuroscience: the structure of retinal prediction (NeurIPS 2019 · similarity 0.93)
- Deep Learning Models of the Retinal Response to Natural Scenes (NeurIPS 2016 · similarity 0.92)
- Neural Regression, Representational Similarity, Model Zoology & Neural Taskonomy at Scale in Rodent Visual Cortex (NeurIPS 2021 · similarity 0.91)
- The emergence of multiple retinal cell types through efficient coding of natural movies (NeurIPS 2018 · similarity 0.90)
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.89)
- A Computational Framework for Modeling Emergence of Color Vision in the Human Brain (ICLR 2025 · similarity 0.88)
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.