AIAI Seminar - 24 April 2023 - Talks by James Vaughan and Chang Luo

Speaker:             James Vaughan

Title:                    Improving the navigation of formal mathematical libraries

Abstract:

The litany of formal mathematics provided with proof assistants is both a necessity and a curse. The presence of these libraries allows users to focus on their own proofs rather reinventing the wheel. However, knowing what to reuse is challenging when it is hidden amongst vast libraries of irrelevant material. In this talk, I will discuss how we are attempting to assist both users and automated tools navigate Isabelle libraries using network science.

Speaker:             Chang Luo  

Title:                    Spatial-Temporal Stock Movement Prediction and Portfolio Selection based on the Semantic Company Relational Graph

Abstract:

This study models the stock price movement as a spatial-temporal prediction task. The spatial dimension models the mutual effects of related companies, while the temporal dimension models the time-series effects of the historical price of companies. We first propose a Semantic Company Relational Graph (SCRG) to represent the spatial company relationships built upon vector embeddings based on company name co-occurrence statistics from a large financial news corpus dataset. We then propose a Non-Independent and Identically Distribution (Non-IID) Spatial-Temporal Graph Neural Network (NIST-GNN) to propagate the spatial information from neighbouring companies as well as the temporal domestic price sequences, for the task of stock movement prediction. Our results show that NIST-GNN outperforms the benchmarks, providing improved risk-adjusted returns. The empirical evaluation also provides insight into the effect of information diffusion from related companies in the US market.