He will give a talk titled "Maximum Entropy and Massive Matrices”, which is scheduled to start at 15:00 in IDSIA’s meeting room in Manno, Galleria 1 building

Abstract:

Scalability is a key concern for machine learning in the big data era, whereby inference schemes are expected to yield optimal results within a constrained computational budget. Underlying these algorithms, linear algebraic operations with high computational complexity, such as the log determinant and matrix inversion, pose a significant bottleneck to scalability. We consider the theoretical and practical implications of an information theoretic approximation using soley matrix vector multiplication.

Best regards,

Alessio