Related Experiment Video
Updated: Jun 6, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
468
Parallel development of object recognition in newborn chicks and deep neural networks
Lalit Pandey1, Donsuk Lee1, Samantha M W Wood1,2,3
1Informatics Department, Indiana University, Bloomington, Indiana, United States of America.
Plos Computational Biology
|December 2, 2024
Summary
Newborn visual development is a space-time fitting process. Deep neural networks trained on chick visual data show similar object recognition learning, validating this model for understanding how newborns learn to see.
Area of Science:
- Neuroscience
- Computer Science
- Developmental Biology
Background:
- Newborn visual systems develop through interaction with their environment.
- Understanding the mechanisms of visual learning is crucial for developmental neuroscience.
Purpose of the Study:
- To test the space-time fitter theory of visual development.
- To compare visual learning in newborn chicks and deep neural networks (DNNs).
Main Methods:
- Controlled-rearing experiments with newborn chicks in impoverished environments.
- Simulating environments and training DNNs (CNNs, transformers) on first-person visual data.
- Comparing viewpoint-invariant object recognition between chicks and DNNs.
Main Results:
- DNNs trained on chick visual data developed similar object recognition skills.
- Space-time fitter DNNs exhibited comparable successes and failures to chicks across viewpoints.
- DNNs successfully learned object recognition in simulated impoverished environments.
Conclusions:
- The space-time fitter model provides a viable framework for understanding newborn visual learning.
- DNNs can serve as computational models for studying visual development from raw experience.
- This research offers image-computable models for visual system development.
