Related Experiment Video
Updated: Oct 4, 2025

Measuring Attention and Visual Processing Speed by Model-based Analysis of Temporal-order Judgments
Published on: January 23, 2017
A computational examination of the two-streams hypothesis: which pathway needs a longer memory?
Abolfazl Alipour1,2, John M Beggs2,3, Joshua W Brown1,2
1Department of Psychological and Brain Sciences, Indiana University, Bloomington, IN USA.
The dorsal visual stream uses absolute metrics for immediate action, while the ventral stream uses relative metrics and memory for complex tasks like object recognition. This study supports the two visual streams theory by showing object classification requires more memory than orientation determination.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Cognitive Science
Background:
- The two visual streams hypothesis posits distinct functions for the dorsal (action-oriented, absolute metrics) and ventral (perception-oriented, relative metrics) visual pathways.
- The dorsal stream is thought to process information for immediate visually-guided motor behavior with minimal memory requirements.
- The ventral stream is associated with object recognition, utilizing relative metrics and benefiting from substantial memory capacity.
Purpose of the Study:
- To computationally test the relationship between neural functional specialization, processing metrics, and memory demands in the visual system.
- To investigate whether tasks typically associated with the dorsal stream require less memory than those associated with the ventral stream.
Main Methods:
- Trained identical recurrent neural networks, including Long Short-Term Memory (LSTM) and leaky-integrator echo state networks (LiESN), on two distinct tasks.
- Task 1: Viewpoint-invariant object classification (putative ventral stream function) using relative metrics.
- Task 2: Orientation/size determination (putative dorsal stream function) using absolute metrics.
Main Results:
- LSTM networks required significantly more memory for the object classification task compared to the orientation/size determination task.
- Memory requirements for object classification increased with feature self-occlusion and when object classes were defined by visual similarity.
- LiESN models did not consistently replicate these findings, potentially due to limitations in fine-grained memory adjustments.
Conclusions:
- The findings support the two visual streams theory, indicating that viewpoint-invariant object classification benefits from longer memory for feature interpolation across viewpoints.
- Orientation/size determination tasks, associated with the dorsal stream, do not appear to benefit from extended memory.
- Computational models like LSTMs can effectively simulate and provide insights into the memory dynamics of distinct visual processing streams.
Related Concept Videos
Parallel Processing
Understanding Memory
Working Memory
Information Processing Approach
Role of Cerebellum and Prefrontal Cortex in Memory
Storage

