Purdue researchers' groundbreaking study published in Frontiers in Computational Neuroscience

Anne Sereno is professor of psychological sciences in the College of Health and Human Sciences and of biomedical engineering in the Weldon School of Biomedical Engineering.
Anne Sereno is professor of psychological sciences in the College of Health and Human Sciences and of biomedical engineering in the Weldon School of Biomedical Engineering.
Looking for an apple in a bowl of fruit, how do you locate the apple? Your brain must correctly bind together the location and features of each fruit. This “binding problem” is a significant challenge for artificial neural networks for object or scene recognition, which often have to add in attention mechanisms or memory-augmented architectures to mitigate the problem.

A groundbreaking study from Anne Sereno’s Purdue lab, led by her student Zhixian Han, has demonstrated, using artificial neural networks, that location is always a better choice than other kinds of features to solve the binding problem. The study, entitled “A Spatial Map: A Propitious Choice for Constraining the Binding Problem,” was published in Frontiers in Computational Neuroscience.

Previous work from their own and other labs has shown that two-pathway, segregated networks can determine objects’ identities and locations more accurately and efficiently than one-pathway networks. However, when using such networks to process multiple objects' identities and locations, a binding problem arises because the networks may not associate each object's features with its location correctly. Typically the brain needs to process and link many attributes of an object (e.g., shape, color, location), and any of these attributes could be used to constrain the binding problem. In their current study, Han and Sereno tried to find the best attribute to constrain the binding problem when presenting multiple visual objects with multiple attributes (shape, luminance, orientation, location) to such two-pathway networks. They found that location is always the best compared to the other attributes for constraining the binding problem.

The computational findings perfectly agree with previous neurophysiological findings that show that the organization or map in many visual cortical areas is primarily a spatial map.

“Such a modeling approach provides insights into visual cortical organization, suggesting that there is a computational advantage for these retinotopic spatial maps in the brain” said Sereno.

Anne Sereno is professor of psychological sciences in the College of Health and Human Sciences and of biomedical engineering in the Weldon School of Biomedical Engineering, and Zhixian Han a PhD candidate in Mathematical and Computational Psychology, in the Department of Psychological Sciences.

Source: Anne Sereno (asereno@purdue.edu)