In the last decade, machine vision has seen tremendous progress using deep learning for both high-level and low-level tasks. Still, it cannot solve many problems as quickly and accurately as our brain, e.g., geometry understanding, dynamic scene parsing, few shot learning of novel objects, anomaly detection, stable reproduction of imaginary visuals, etc. Open world scenarios with unknown, evolving, and long tailed distributions pose a major challenge to the offline training setup. New inspirations from the brain could make machine vision more efficient, robust, and capable of continuous learning and adaptation.

This workshop seeks to understand the contexts in which the brain’s visual system can inform and improve current AI systems. What principles of the brain can be extracted and incorporated into AI and ML? Some relevant research questions are as follows:

1. What role do eye movements play in visual search, recognition, information gathering, social interaction, etc.? Can foveated vision be fruitful for artificial systems?

2. How to build machine vision systems that are explainable?

3. How can a computer achieve the dynamic scene understanding capabilities of the brain, such as the separation of self-motion and object-motion, and apply the knowledge to real world scenarios such as self-driving?

4. How do recurrent and top-down connections in the visual cortex predict outcomes? And why is this important?

5. How does spike based spatiotemporal encoding in the visual cortex increase information processing?

6. How does the brain extract complex social information from visual cues? Can we train machines to perform social cognition?

7. What are the different roles of attention in biological vision? Can the overt, covert, or feature based attention mechanisms in biology be applied in computer vision for efficiency?

8. How does the brain learn novel objects in a few exemplars and continue to learn with minimal loss?