Related Experiment Video
Updated: Aug 6, 2026
09:39
Spatial Temporal Analysis of Fieldwise Flow in Microvasculature
Published on: November 18, 2019
Perceptual distortions in PredNet and quantification of top-down/bottom-up flow
Hiroki Kojima1, Keisuke Suzuki2, Yuichi Yamashita1
1Department of Information Medicine, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo, Japan.
Frontiers in Computational Neuroscience
|July 22, 2026
Summary
This study explores visual distortions in Autism Spectrum Disorder (ASD) using predictive coding models. Findings suggest that adjusting prediction error weighting can mitigate these distortions by improving the model's ability to process global visual features.
Area of Science:
- Neuroscience
- Computational Psychiatry
- Artificial Intelligence
Background:
- Perceptual distortions are common in psychiatric conditions like Autism Spectrum Disorder (ASD).
- Bayesian models propose "aberrant precision" as a core mechanism in psychiatric and learning disorders.
- Existing models struggle to explain visual distortions due to limitations with high-dimensional data and precision definition.
Purpose of the Study:
- To apply predictive coding and advanced analysis to model visual distortions.
- To investigate the role of aberrant precision in visual processing within a computational framework.
- To address limitations of previous models in handling complex visual inputs.
Main Methods:
- Utilized the predictive coding-based deep neural network, PredNet.
- Developed a novel analysis method inspired by the precision account using the Hilbert-Schmidt Independence Criterion (HSIC).
- Trained the model with extended temporal contexts and manipulated top-layer prediction error weighting.
Main Results:
- Visual distortions emerged when the model was trained with longer temporal contexts.
- Increasing the weight of the top layer's prediction error mitigated these distortions.
- HSIC analysis revealed that this weighting adjustment enhanced top-down information flow and global feature capture (e.g., rotation, brightness, hue).
Conclusions:
- Aberrant precision, modeled via prediction error weighting in PredNet, plays a role in visual distortions.
- Computational models can simulate and offer insights into perceptual abnormalities in ASD.
- Enhanced top-down processing improves the model's ability to interpret complex visual scenes, potentially relevant for understanding psychiatric disorders.
