Research map / Vision-language and multimodal / Human and embodied modeling
Cognitive Steering in Deep Neural Networks via Long-Range Modulatory Feedback Connections
Talia Konkle, George A. Alvarez
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 |
|---|---|---|---|
| add long-range modulatory feedback pathways (LRM3 vs plain backbone) | AlexNet · ImageNet | top-1 accuracy | Improved |
| add LRM feedback pathways, feed-forward pass only | AlexNet · Brain-Score IT | neural predictivity (r): r = 0.358 -> r = 0.400 | Improved |
| add LRM feedback pathways | AlexNet · Brain-Score V2 | neural predictivity (r): r = 0.353 -> r = 0.333 | Hurt |
| add LRM feedback pathways | AlexNet · Brain-Score V1 | neural predictivity (r): r = 0.507 -> r = 0.531 | Improved |
| category-prototype steering instead of no steering (alpha=3) | LRM3 · Composite side-by-side | top-1 accuracy: 24.5% -> 68.9% | Improved |
| category-prototype steering instead of no steering (alpha=3) | LRM3 · Composite overlay | top-1 accuracy: 5.1% -> 51.8% | Improved |
| CLIP-text steering instead of no steering (alpha=3) | LRM3 · Composite side-by-side | top-1 accuracy: 24.5% -> 70.4% | Improved |
| one-hot steering instead of no steering (alpha=3) (ablation) | LRM3 · Composite side-by-side | accuracy: 0.245 -> 0.235 | No clear effect |
| instance steering vs category-prototype steering (ablation) | LRM3 · Composite side-by-side | accuracy: 0.689 -> 0.507 | Hurt |
| more feedback pathways: LRM1 to LRM2 (category steering) (ablation) | LRM · Composite side-by-side | accuracy: 0.395 -> 0.743 | Improved |
| increase alpha gain beyond 4 (ablation) | LRM3 · Composite tests | accuracy | Hurt |
| additional modulatory passes beyond the first (ablation) | LRM · Composite tests | accuracy | No clear effect |
Closest papers by meaning
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.88)
- Pathways of Visual Information Flow in Vision-Language Models (NeurIPS 2026 · similarity 0.86)
- A Convolutional Neural Network Interpretable Framework for Human Ventral Visual Pathway Representation (AAAI 2024 · similarity 0.85)
- Testing the Limits of Fine-Tuning for Improving Visual Cognition in Vision Language Models (ICML 2025 · similarity 0.84)
- Dynamic Multimodal Activation Steering for Hallucination Mitigation in Large Vision-Language Models (ICLR 2026 · similarity 0.84)
- DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026 · similarity 0.84)
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.
