Speaker

Ananya Joshi

Abstract

As systems in critical domains become increasingly complex, effective data monitoring is essential for detecting real-time faults and system changes. Traditional methods, however, often rely on unrealistic assumptions, such as consistent, high-fidelity data, unlimited human attention, and abundant computational resources.

This talk introduces a human-in-the-loop framework for system monitoring in resource-constrained settings, incorporating novel fault detection methods grounded in extreme value theory. By leveraging human domain expertise in a statistically rigorous manner, this approach identifies significant anomalies in large datasets, while remaining computationally efficient. Applied to monitoring a public health data system for over a year, this framework enabled the Delphi Group at Carnegie Mellon University to improve fault analysis efficiency by 54x. These results demonstrate the potential for scalable, robust system monitoring in resource-limited environments and how health care-specific settings can drive strong computational research.

Bio

Ananya Joshi is a PhD candidate in Computer Science at Carnegie Mellon University. Her research develops interpretable solutions for data-intensive problems in resource-constrained settings, especially with applications in computational epidemiology. Her research has been supported by the NSF GRFP and a Fulbright Research Award. In addition, she has served as a project manager to oversee how her methods are deployed in practice and collaborated with public health agencies, including the CDC. She has interned at IBM Research on trustworthy machine learning and previously conducted research on software-defined networks, caching algorithms, sensor networks, model fine-tuning, and resource forecasting.