Graduate Student Seminar

Speaker 1: Qing Ji

Title

Applying Hidden Markov Models to Stock Market Data

Abstract

The terms bull and bear are used to describe the direction of the movement of the stock market.  These market trends are recognized after the events meaning they are unobserved at the time.  Hence, the market trend is a latent variable.  The hidden Markov model (HMM) is often used to analyze stock market data because its structure suits this problem well.

In this talk, an example of a HMM is shown using S&P500 index values and then we extend the HMM to the stock price data and test a proposed stock evaluation criteria.  The result shows that proposed method generates higher a capital return in percentages compared to the market return in average in the same period.

Speaker 2: Michael Retzlaff

Title

Data Compression for Optimization of a Molecular Dynamics System -- Preserving Basins of Attraction

Abstract

The growing disparity between speed of computation and writing in the next generation of supercomputers heightens the need for the use of data compression within their software.  A highlighting example, found in an application under development for the Exascale Computing Project (ECP), finds and catalogs the exponentially many optimal states of a molecular dynamics system.  To reduce time required for writing to storage, the optimizers must be compressed in a lossy fashion whilst guaranteeing the compressed optimizer remains in the same basin of attraction under an optimization algorithm.

We develop a framework to determine the acceptable level of compression of an optimizer by applying variants of the Kantorovich Theorem, using binary digit rounding as our compression technique.  Choosing the Lennard-Jones potential function as a model problem, we devise a compression scheme then show numerical results which validate and show its scalability to systems of the size under consideration for ECP.