MIT researchers show artificial neural networks can build their own learning machinery

Researchers at MIT's McGovern Institute for Brain Research and its K. Lisa Yang Integrative Computational Neuroscience (ICoN) Center published a paper in PNAS on October 9 about how neural networks learn. The paper asks what kinds of artificial neural networks can best help neuroscientists understand how the brain learns. Like the brain's circuits, artificial networks reorganize themselves as they learn, adjusting the strength of their connections to improve performance. The researchers argue that a crucial measure of whether a network's way of learning is biologically plausible is its ability to organize itself without detailed instructions from a developer. They show that it is possible for an artificial neural network to build its own learning machinery. MIT says networks that learn this way may offer insight into how the brain reorganizes itself during learning. For neuroscience, the work offers a way to choose which machine models are most informative. For AI research, it adds to the study of systems that develop their own learning procedures. The MIT announcement does not give performance figures or describe specific applications.