Comparing humans and machines on robust visual inference (03)

• Subject Area: Cognitive, Systems and Behavioural Neurobiology ; General, Cognitive and Mathematical Psychology
• Term: from 2017 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 276693517

Biological visual systems facilitate navigation through complex environments and can recognise objects despite highly variable contextual conditions. This robust visual inference relies crucially on a suitably general feature space: the abstraction of meaningful environmental properties from a light field. In this project we will develop a psychophysical paradigm to directly compare human and machine vision systems (primarily, the feature spaces encoded by deep convolutional neural networks) on two robust visual inference tasks (scene understanding and style / content dissociation). We will develop two convolutional deep neural network models: one optimised to get the task correct, and one optimised to mimic human decisions. We will then compare the representational similarity of these two models to gain insight into which features are shared.

• DFG Programme: Collaborative Research Centres
• Subproject of: SFB 1233: Robust Vision - Inference Principles and Neural Mechanisms
• Applicant Institution: Eberhard Karls Universität Tübingen
• Project Heads: Professor Dr. Matthias Bethge; Professor Thomas Wallis, Ph.D., until 4/2019