Related Experiment Videos
Dynamic model of visual recognition predicts neural response properties in the visual cortex
1Department of Computer Science, University of Rochester, NY 14627-0226, USA.
Neural Computation
|May 15, 1997
Summary
This study introduces a hierarchical network model explaining visual cortical neuron responses. The model uses an extended Kalman filter and minimum description length (MDL) principle to integrate bottom-up and top-down signals for recognition.
Area of Science:
- Computational Neuroscience
- Visual Perception
- Machine Learning
Background:
- Visual cortical neurons respond to stimuli outside their classical receptive field.
- Modulatory effects on neural responses are observed during natural scene viewing.
Purpose of the Study:
- To present a hierarchical network model of visual recognition.
- To explain experimental observations of neural responses using computational principles.
Main Methods:
- Utilized an extended Kalman filter based on the minimum description length (MDL) principle.
- Integrated bottom-up and top-down signals for dynamic state prediction.
- Employed Hebbian learning for synaptic weight adaptation, akin to an expectation-maximization (EM) algorithm.
Main Results:
- The model explains neural responses modulated by stimuli beyond the classical receptive field.
- Demonstrated the role of feedback from higher cortical areas in mediating these effects.
- Simulations validated the model's ability to explain experimental data in both fixation and free viewing conditions.
Conclusions:
- Reciprocal connections between visual cortical areas play an active role in neural response properties.
- The model provides a framework for understanding visual recognition and neural processing.
- Feedback mechanisms are crucial for integrating contextual information in visual perception.