Research map / Architectures and efficiency / CNN design
Neural system identification for large populations separating “what” and “where”
David Klindt, Alexander S Ecker, Thomas Euler, Matthias Bethge
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 |
|---|---|---|---|
| Use a sparse factorized readout that jointly learns receptive-field locations and feature weights | CNN neural system identification · Mouse V1 scans 1–3 | correlation: 0.47 -> 0.50 | Improved |
| Replace the factorized readout with a fully connected readout (ablation) | CNN neural system identification · Mouse V1 average | correlation: 0.50 -> 0.43 | Hurt |
| Fix receptive-field masks using spike-triggered averages instead of optimizing them jointly (ablation) | CNN neural system identification · Mouse V1 average | correlation: 0.50 -> 0.42 | Hurt |
| Use an LNP model instead of a shared convolutional feature model with factorized readout | Mouse V1 neural response prediction · Mouse V1 average | correlation: 0.36 -> 0.50 | Hurt |
| Use a shared feature space across neurons rather than fitting each neuron independently | CNN system identification · Homogeneous linear population | FEV: same level of performance with less than a quarter of the data | Improved |
| Estimate receptive-field locations in advance with spike-triggered averages rather than learning them jointly (ablation) | CNN system identification · Synthetic nonlinear VGG feature population | FEV | Hurt |
| Use a fully connected readout instead of the factorized readout (ablation) | CNN system identification · Synthetic nonlinear VGG feature population | FEV | Hurt |
| Increase the diversity of neuron types in the population (ablation) | CNN with factorized readout · Synthetic VGG population | FEV | No clear effect |
| Increase the diversity of neuron types with a fully connected readout (ablation) | CNN with fully connected readout · Synthetic VGG population | FEV | Hurt |
| Use L1 sparsity on feature weights to encourage each neuron to select a feature channel (ablation) | CNN with factorized readout · Two-type linear population | cell-type interpretability | Improved |
| Use the ground-truth convolution kernel and optimize only location masks (ablation) | CNN system identification · Homogeneous linear population | FEV | No clear effect |
Closest papers by meaning
- Generalization in data-driven models of primary visual cortex (ICLR 2021 · similarity 0.89)
- Comparison Against Task Driven Artificial Neural Networks Reveals Functional Properties in Mouse Visual Cortex (NeurIPS 2019 · similarity 0.87)
- Neural Regression, Representational Similarity, Model Zoology & Neural Taskonomy at Scale in Rodent Visual Cortex (NeurIPS 2021 · similarity 0.87)
- Self-Attention-Based Contextual Modulation Improves Neural System Identification (ICLR 2025 · similarity 0.86)
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.85)
- STSBench: A Large-Scale Dataset for Modeling Neuronal Activity in the Dorsal Stream of Primate Visual Cortex (NeurIPS 2025 · 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.
