AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

A tale of two tails: Preferred and anti-preferred natural stimuli in visual cortex

Rabia Gondur, Patricia L. Stan, Matthew A. Smith, Benjamin R. Cowley

ICLR 2026 · Vision-language and multimodal › Human and embodied modeling · approach: fMRI visual decoding

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 changedWhereMeasuredEffect
use pre-ReLU instead of post-ReLU ResNet50 features in a linear map (ablation)Linear mapping to V4 responses · V4 response predictionnoise-corrected predictionHurt
negate sign before ReLU (anti-preferred features only) (ablation)Linear mapping to V4 responses · V4 response predictionnoise-corrected predictionHurt
add 1x1 channel mixing plus ReLU before the linear readout (LRL mapping) (ablation)Pre-ReLU ResNet50 features · V4 response predictionnoise-corrected predictionImproved
add the same LRL mapping to post-ReLU features (ablation)Post-ReLU ResNet50 features · V4 response predictionnoise-corrected predictionImproved
train on both preferred and anti-preferred images instead of random images (fewer than 5k) (ablation)Student CNN predicting V4 model neurons · V4 model neuron tuningheld-out R2Improved
train on preferred only or anti-preferred only instead of random images (ablation)Student CNN predicting V4 model neurons · V4 model neuron tuningheld-out R2Hurt
train on non-preferred (near-median) images (ablation)Student CNN predicting V4 model neurons · V4 model neuron tuningheld-out R2Hurt
double candidate pool to 1M images for choosing extreme images (ablation)Preferred+anti-preferred training set · V4 model neuron tuningheld-out R2: 500k -> 1M imagesImproved
use gradient-synthesized images for training (ablation)Student CNN predicting V4 model neurons · V4 model neuron tuningheld-out R2Hurt
add anti-preferred images to preferred images when training (ablation)Student CNN predicting post-ReLU ResNet50 units · ResNet50 unit tuning (2k images)held-out R2Improved
give human subjects both preferred and anti-preferred prior images instead of one type or none (ablation)Human psychophysics · Predict V4 model neuronsP(correct)Improved
give human subjects anti-preferred prior images (ablation)Human psychophysics · Predict ResNet50 DNN unitsP(correct)No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy