Psychology undergraduate research receives global recognition

October 11, 2022

Justin Zhou

A master’s student in neuroscience at Western has received international recognition for research conducted as an undergraduate student.

Justin Zhou’s research on neural network models of vision has been recognized as a Highly Commended paper in the Global Undergraduate Awards. Zhou submitted work he completed during his bachelor’s degree in psychology, also at Western.

The paper, ‘Incorporating action information into computational models of the human visual system', documents research from Zhou’s honours thesis project. Zhou researched neural network models of vision, with a particular focus on models of the ventral visual stream.

In the human brain, the ventral visual stream is responsible for object recognition, and the dorsal visual stream is responsible for visually guided behaviour.

Previous research with neural networks had created models similar in structure and function with the ventral visual stream. However, these models did not reflect the exchange of information between dorsal and ventral streams that occurs in the real human brain. Zhou’s project aimed to incorporate dorsal stream information into models of the ventral stream to improve their realism.

Zhou’s approach was to use multi-task learning, with one learning task reflecting ventral stream function and one reflecting dorsal stream function. For dorsal stream function, he created a task to make a model generate grasp points, determining where to grab an object in order to lift it. This type of task requires more spatial computation, including representing the shape of the object, the shape of the hand, and the centre of mass. For ventral stream function, Zhou used a standard object categorization task.

To measure the similarity between the model and the human brain, Zhou used a technique called representational similarity analysis (RSA), which was co-developed by his supervisor, Dr. Marieke Mur. The method compares patterns of similarity in the representations produced by assortments of different visual stimuli. In doing so, it abstracts information encoding away from fundamental differences in structure and substrate.

The results indicated that multi-task learning yielded increased similarity between the model and the human brain, but had come at the cost of decreasing accuracy on object categorization tasks. Still, Zhou claims there is room for improvement: “Some tests showed it was not learning to its full potential… the results indicated there was a lot of work to do to optimize these models.”

On the experience of working on this project, Zhou says: “I had to learn a lot of this very quickly. I hadn’t done anything with machine learning prior to the end of third year.” Zhou credits the support of his supervisor and lab for his ability to learn all the information he needed for his research.

Zhou describes his interest in this topic as a product of a slow exploration through the field of psychology. Through his first- and second-year psychology courses, he was drawn toward cognitive psychology, focusing on how the brain processes information. Later, he realized the potential of computational approaches for implementing and testing the models of cognitive psychology.

Zhou is currently working towards an MSc in Neuroscience at Western and aims to continue his research. His long-term goal is to use what he has learned about the brain in the medical field. “Computer science has a lot to offer psychology in the future,” he said.

As for receiving the award itself, Zhou says that the recognition made him feel validated as a researcher. “Scientific research and scholarship is fundamentally international, collaborating with people all around the world… So I guess with this award, I feel validated as part of that community.”

The Global Undergraduate Awards is an academic awards program recognizing undergraduate work. Submissions are evaluated anonymously by a group of international academics, with the top 10% of entries from each region in each category named as Highly Commended.