Prof. Galit Yovel
How do humans recognize faces? Insights from biological and artificial face recognition systems
Face recognition is a computationally challenging classification task that is critical for intact social interaction. The question of how to resolve this task has occupied both cognitive and computer scientists for many years. In recent years, machine learning algorithms have reached human-level performance. However, it is not clear whether they attain a similar solution as the human brain. In my talk I will propose a model that accounts for human face recognition. I will show that this model is consistent with the representation that is generated by deep convolutional neural networks (DCNNs) that are optimized for face recognition. This human-like face representation emerges at higher layers of a face-trained but not an object-trained DCNN, akin to the divergence to a face and an object system in high-level visual cortex. I will further claim that the representation that enables human face recognition primarily depends on conceptual (supervised) learning rather than pure perceptual (unsupervised) learning. This enables the system to accomplish its main goal to recognize socially significant people.