Machines inspired by physiological and anatomical constraints to improve pattern completion
by Kris Brewer
When we open our eyes in the morning and take in that first scene of the day, we don’t give much thought to our surroundings and the fact that our brain is processing the objects within our field of view with such efficiency and lack of information to allow us to go about our daily functions. The glass of water you left on the nightstand when preparing for bed is now partially blocked from your line of sight by your alarm clock, yet you know that it is a glass.
This seemingly simple ability for humans to recognize partially occluded objects (defined in this situation as the effect of one object blocking another object from view) has been a complicated problem for the computer vision community. Martin Schrimpf, currently a graduate student in the DiCarlo lab in the Brain and Cognitive Sciences Department at MIT and co-author of this paper, explains that machines have become increasingly adept at recognizing whole items quickly and confidently, but when something covers part of that item from view, this task becomes increasingly difficult for the models to accurately recognize the article.
How are we as humans able to repeatedly do this everyday task without putting much thought and energy into this action, ingesting pieces that allow us to identify the whole quickly and accurately? They started with the human visual cortex as a model for how to improve the performance of machines in this setting, describes Gabriel Kreiman, Professor of Ophthalmology at Boston Children’s Hospital and Harvard Medical School and lead Principal Investigator on this study.
In their new paper Recurrent computations for visual pattern completion as published in the August 2018 Proceedings of the National Academy of Sciences (PNAS), the team shows how they developed a computational model, inspired by physiological and anatomical constraints, that was able to capture the behavioral and neurophysiological observations during pattern completion. This model provides initial insights towards understanding how we can make inferences from minimal information. This work was done within the Center for Brains, Minds and Machines (CBMM).
Hear from Kreiman and Schrimpf about the process and method that led to this development.