by Theodore (Ted) Stark
When you see another person, your brain utilizes the fusiform facial area (FFA) to identify the face faster than anything else. Therefore, you can recognize faces before you can recall a person’s name. When viewing a face from the side (say 45 degrees), specific neurons fire within the FFA to recognize a face even at an angle. Researchers from MIT’s Center for Brains, Minds, and Machines (CBMM) have reported a computational model they have been building to recognize faces has spontaneously learned to identify faces at an angle.
This model was trained to spot visages from an extensive battery of images. As it learned, it created an intermediate step affording the model to identify faces from an angle in addition to straight on.
This advancement is exciting because it duplicates findings observed in investigating how primates process faces. This indicates that the computational model may be doing something similar to the brain. The authors of the paper, which appeared in Current Biology, are quick to point out that the computational model’s approach is probably a simplified version of how the brain processes faces, but it is a significant step forward.
MIT’s computational model relied on a neural network that has also been used by companies like Google, Facebook, and Apple to sort through photos. Better understanding how the brain recognizes faces can help artificial intelligence with facial perception tasks with heightened accuracy. Where this is clearly in its early stages, it represents another step forward for machine learning and artificial intelligence replicating how the brain works.
This was Article 72 from the Studio Quick Facts Series.
References
Guergiuev, J., Lillicrap, T. P., & Richards, B. A. (2016). Towards deep learning with segregated dendrites. arXiv preprint arXiv:1610.00161.
Kietzmann, T. C., McClure, P., & Kriegeskorte, N. (2017). Deep Neural Networks In Computational Neuroscience. bioRxiv, 133504.
Leibo, J. Z., Liao, Q., Anselmi, F., Freiwald, W. A., & Poggio, T. (2017). View-tolerant face recognition and Hebbian learning imply mirror-symmetric neural tuning to head orientation. Current Biology, 27(1), 62–67.
Martinez, A. M. (2017). Computational models of face perception. Current Directions in Psychological Science, 26(3), 263–269.
MIT’s AI figured out how humans recognize faces. (2017). Engadget. Retrieved 29 July 2017, from https://www.engadget.com/2016/12/02/mit-s-ai-figured-out-how-humans-recognize-faces/
Office, L. (2017). How the brain recognizes faces. MIT News. Retrieved 29 July 2017, from http://news.mit.edu.ezproxy.canberra.edu.au/2016/machine-learning-system-brain-recognizes-faces-1201
Tacchetti, A., Isik, L., & Poggio, T. (2016). Invariant recognition drives neural representations of action sequences. arXiv preprint arXiv:1606.04698.