Back COLT seminar - Monday, September the 13th 14:30h, Jeffrey Bowers

COLT seminar - Monday, September the 13th 14:30h, Jeffrey Bowers




Professor Jeffrey Bowers from the University of Bristol gave a talk about the difference between how humans and artificial neural networks visually process images. He is actually leading an ERC funded project, called Generalisation in Mind and Machine, that englobes this topic.


Title : ​Reevaluating the claim that deep neural networks identify objects like humans

Abstract: Deep neural networks (DNNs) have had extraordinary success in classifying photorealistic images of objects and are often described as the best models of biological vision.  This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioural benchmark datasets, and (3) DNNs do the best job in predicting brain signals in response to images taken from various brain benchmark datasets (e.g., single cell responses or fMRI data).  However, current behavioural and brain benchmarks report the outcomes of observational experiments that do not manipulate any independent variables, and I show that prediction on these datasets may be mediated by DNNs that share little or no overlap with biological vision.  More problematically, I show that DNNs account for almost no results from psychological research.   This contradicts the common claim that DNN are good, let alone the best, models of human object recognition.



SDG - Sustainable Development Goals:

Els ODS a la UPF