Related Experiment Video
Updated: Jul 9, 2025
10:05
Rodent Brain Microinjection to Study Molecular Substrates of Motivated Behavior
Published on: September 16, 2015
14.4K
Thinking beyond the ventral stream: Comment on Bowers et al.
Christopher Summerfield1, Jessica A F Thompson1
1Department of Experimental Psychology, University of Oxford, Oxford, UK christopher.summerfield@psy.ox.ac.uk jessica.thompson@psy.ox.ac.uk https://humaninformationprocessing.com/ https://thompsonj.github.io/about/.
The Behavioral and Brain Sciences
|December 6, 2023
Summary
Deep learning models struggle with visual perception constraints. The solution involves designing tasks that mimic biological vision
Area of Science:
- Computer Vision
- Neuroscience
- Artificial Intelligence
Background:
- Deep learning models often overlook established constraints in biological visual perception.
- Previous research has extensively documented these visual perception constraints.
Purpose of the Study:
- To address the limitations of deep learning in capturing visual perception constraints.
- To propose a method for improving deep learning models by aligning them with biological vision principles.
Main Methods:
- Designing novel stimuli and tasks for deep learning evaluation.
- Focusing on ecologically valid problems, such as scene understanding and action preparation.
Main Results:
- Deep learning models can be improved by incorporating biologically relevant tasks.
- Task design is crucial for enhancing the performance and generalizability of deep learning models in vision.
Conclusions:
- Discarding deep learning is not the answer; instead, refine its application.
- Aligning artificial systems with the evolutionary pressures on biological vision offers a promising path forward.
