Computational challenges and stochastic algorithms in large scale Machine Learning - from 10:25am to 11:00am

 Pr. Naftlati Tishbi - The Hebrew University of Jerusalem, Israel

 

Modern Machine Learning is based on local optimization algorithms in high dimensions which are often implemented by stochastic gradient decent methods. Such algorithms use diffusion in high dimensional spaces to find global or local error optima  or invariant manifolds of such optima. This is a computationally demanding task when done on big data and millions of dimensions. I will discuss the importance of enhancing such algorithms and the potential of using stochastic quantum Hamiltonians of the type proposed by the Copac project for scalable quantum diffusion enhancement. I will also discuss possible applications for training large deep neural networks with such an optical co-processor.