Speaker 1: Abhishek Guin

Title

 Sequential Probability Ratio Test

Abstract

In several real life problems, Neyman-Pearson's fixed sample size methodology either would not work or can be beaten by a sequential design.  In this talk, we discuss motivations for probing sequential methods and provide a setup for testing simple hypotheses (SPRT). We will look at its properties to properly characterize it, view applications, examples and demonstrate its optimality property among all tests.

Speaker 2: Bryce Carey

Title

Developing a Computational Model of Neural Networks into a Learning Machine

Abstract

Temporal hierarchical probabilistic associative memory (THPAM) is a functional model of biological neural networks which learns and predicts based on orthogonalization of bipolar vector inputs rather than conventional optimization techniques popular in machine learning. The orthogonalization procedure provides convenient learning, predictive, and generalization mechanisms based on relative hamming distance and frequency of learned inputs, but also incurs exponential computational complexity with the input dimension. THPAM is applied to sample small datasets to demonstrate its application towards categorical and real-valued data. An alternative generalization scheme is proposed which provides a strictly offline learning scheme that avoids the orthogonalization procedure, permitting the application of THPAM to more interesting data.