"Bridging the Gap between Machine Learning and Neuroscience"
, 499 DSL
With deep learning making the news, there is a great deal of hype around artificial intelligence. But how close has AI come to human intelligence? In this talk, we put both machine learning and human learning under the microscope, compare them, and then introduce a new algorithm addressing a major difference. We begin with ideas from computational learning theory, the theory of machine learning and consequently deep learning. Following this, we head into the latest neuroscience research on biological learning theories and the algorithms they inspire. With this machine learning and neuroscience foundation, we analyze how each approach treats unsupervised learning. Then, in an attempt to bring these two approaches to intelligence closer, we derive a novel link between information theory and generalized Hebbian plasticity, the currently accepted biological model for learning. We prove the convergence of our derivation's associated stochastic approximation algorithm and show how numerous radial basis neurons, locally updating, self-organize to model the data generating distribution. Lastly, we take a preliminary look at how complex behavior might emerge from this self-organization.