Research map / Image generation / Fast diffusion sampling
Energy Guided Diffusion for Generating Neurally Exciting Images
Paweł A. Pierzchlewicz, Konstantin Friedrich Willeke, Arne Nix, Pavithra Elumalai, Kelli Restivo, Tori Shinn, Cate Nealley, Gabrielle Rodriguez, Saumil Patel, Katrin Franke, Andreas S. Tolias, Fabian H. Sinz
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 EGG diffusion instead of gradient ascent to generate MEIs | V4 encoding models · MEI cross architecture (Gaussian) | mean unit activation: 5.51 -> 6.53 | Improved |
| use EGG diffusion instead of gradient ascent to generate MEIs | V4 encoding models · MEI cross architecture (Attention) | mean unit activation: 4.42 -> 5.50 | Improved |
| use EGG diffusion instead of gradient ascent to generate MEIs | Attention V4 model · MEI within architecture (Attention) | mean unit activation: 7.59 -> 10.56 | Improved |
| use EGG diffusion instead of gradient ascent to generate MEIs | Gaussian V4 model · MEI within architecture (Gaussian) | mean unit activation: 11.43 -> 11.76 | No clear effect |
| diffusion guidance instead of 1000-step gradient ascent | MEI generation · MEI generation time | generation time (s): 219s -> 46s | Improved |
| use Attention readout instead of Gaussian readout (task-driven core) (ablation) | V4 encoding model · Macaque V4 neural prediction | correlation to average: 0.262 -> 0.276 | Improved |
| use Attention readout instead of Gaussian readout (data-driven core) (ablation) | V4 encoding model · Macaque V4 neural prediction | correlation to average: 0.229 -> 0.294 | Improved |
| use data-driven core instead of task-driven core (Attention readout) (ablation) | V4 encoding model · Macaque V4 neural prediction | correlation to average: 0.276 -> 0.294 | Improved |
| use data-driven core instead of task-driven core (Gaussian readout) (ablation) | V4 encoding model · Macaque V4 neural prediction | correlation to average: 0.262 -> 0.229 | Hurt |
| use EGG diffusion instead of gradient descent for image reconstruction | Gaussian model reconstruction · Reconstruction, cross architecture | images with lower L2 distance (%): 98% | Improved |
| AdamW instead of SGD for gradient ascent baseline (ablation) | GA MEIs · MEI generation | relative activation: r = 0.69 | Hurt |
| 100 instead of 1000 gradient ascent steps (ablation) | GA MEIs · MEI generation | relative activation: r = 0.95 | Hurt |
Closest papers by meaning
- MEIcoder: Decoding Visual Stimuli from Neural Activity by Leveraging Most Exciting Inputs (NeurIPS 2025 · similarity 0.88)
- NEvo: Neural-Guided Evolutionary Video Synthesis for Dynamic Visual Selectivity (NeurIPS 2026 · similarity 0.87)
- Benchmarking Out-of-Distribution Generalization Capabilities of DNN-based Encoding Models for the Ventral Visual Cortex. (NeurIPS 2024 · similarity 0.85)
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.85)
- Retrospective for the Dynamic Sensorium Competition for predicting large-scale mouse primary visual cortex activity from videos (NeurIPS 2024 · similarity 0.85)
- DiffEye: Diffusion-Based Continuous Eye-Tracking Data Generation Conditioned on Natural Images (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.
