Allison M. Austin

Hello! I'm PhD student at the University of California, Davis advised by Dr. Kwan-Liu Ma at the VIDI Labs. I previously worked with Dr. Bo Fu in the D2 Lab at California State University, Long Beach.

I'm currently interested in applications of visual analytics for understanding complex algorithms and machine learning models to improve monitoring and usage of high-performance computing (HPC) systems.

On-going Projects

Visualizing Distributed Resilient Workflows and Infrastructure

Modern High-Performance Computing (HPC) infrastructures supporting large-scale scientific experiments and High Energy Physics (HEP) workflows generate massive volumes of complex telemetry and batch queue data. This project focuses on designing scalable visual analytics systems and interpretable machine learning frameworks to monitor, diagnose, and predict infrastructure and workflow performance across federated HPC environments.

Publications

Predictive Modeling of High Energy Physics Workflows on Federated HPC Batch Systems
Under Review
Allison Austin, Logan Tang, Justin Mueller, Andrew Norman, Kwan-Liu Ma

Interactive Visual Analytics for Generating and Exploring LLM-Predicted Protein-Protein Interactions
ACM BCB 2026
Allison Austin, Shilpika, Aditya Tanikanti, Venkatram Vishwanath, Michael E. Papka, Kwan-Liu Ma
[paper][code]

Understanding Large-Scale HPC System Behavior Through Cluster-Based Visual Analytics
ISC 2026
Allison Austin, Shilpika, Yan To Linus Lam, Yun-Hsin Kuo, Venkatram Vishwanath, Michael E. Papka, Kwan-Liu Ma
[paper][code]

Visualizing Mappings Between Pairwise Ontologies - An Empirical Study of Matrix and Linked Indented List in Their User Support During Class Mapping Creation and Evaluation
ISWC 2023
Bo Fu, Allison Austin, Max Garcia
[paper][talk][code]

Past Projects

Visual Analytics for Generating and Evaluating Protein-Protein Interactions

Molecular data has been widely used to enrich visual representations of protein-protein interaction (PPI) networks, and has been expanded to support large-language models (LLMs) for various tasks (e.g. protein generation, interaction prediction). This project explores visualization techniques for generation and evaluation of PPIs that originate from LLMs.

Visualizing Elastic Depths for Detecting Anomalies in Functional Time Series Data

Applying an interactive visualization layer to Functional Data Analysis (FDA) techniques that decouple phase and amplitude variations in high-dimensional time series curves. [code]

Transformers for Failure Prediction in High-Performance Computing System Logs

Exploring transformer-based architectures (GPT and LogBERT) for online anomaly detection and failure prediction in HPC system log sequences. We evaluate model performance across large-scale time-series log datasets from LANL (System 20) and LLNL (BlueGene/L) using probabilistic inference techniques and custom attention mechanism variations, including Additive Attention and Hierarchical Attention Networks (HAN). [paper][code]