Paul Chichura

Research

My research lies at the intersection of astronomy, data science, and artificial intelligence. I develop and apply novel computational techniques to expand the scientific reach of large astronomical surveys, with a special interest in time-varying events and telescope operations. This page describes my past and current scientific interests:

AI for Telescope Control

Guiding Question: How can we use AI to make complex scientific instruments like telescopes operate more efficiently and accurately?

A telescope's performance is not static; it's a dynamic system influenced by its environment and strategic goals. My research focuses on building and deploying AI models that integrate directly with telescope control systems to enhance real-time performance and optimize long-term scientific strategy.

My work with the South Pole Telescope (SPT) addressed a decades-old problem where pointing accuracy was influenced by structural deformations caused by extreme Antarctic weather. I developed and deployed XGBoost models on the SPT control system to predict and correct these pointing errors in real time. By engineering predictive features from a sensor suite, I built models that reduced the average pointing error by over 30% during in situ tests, an improvement that will enable future observations with the Event Horizon Telescope. These results were published in JAI.

Currently, my research focuses on AI for survey scheduling, i.e. deciding when and where to point a telescope to maximize scientific return of observations. Traditional survey scheduling relies on bespoke, hand-tuned algorithms, an approach that is inefficient for the multi-objective, resource-constrained environment of modern surveys. I am applying reinforcement learning methods to design interpretable scheduling agents capable of dynamic decision-making. This AI scheduler agent will help balance diverse science goals, adapt to changing observing conditions, and reduce the human cost of survey design.

Technical Skills: Machine learning (XGBoost, Reinforcement Learning) | Real-Time Forecasting | Feature Engineering | Optimization & Resource Allocation | Model Deployment & Validation

Variable and Transient Astronomy

Guiding Question: How can we reuse large, archival datasets to find and characterize time-varying astronomical events?

The universe is full of objects that change, flash, or move over time, from the supermassive black hole at the center of our galaxy (Sagittarius A*, or Sgr A*) to nearby asteroids. Detecting and analyzing these variable and moving sources requires sophisticated time-series analysis and signal processing, especially when the signals are buried in terabytes of data.

I led the first targeted analysis of asteroids in cosmic microwave background (CMB) survey data, repurposing archival observations from the SPT. To do this, I developed a new analysis pipeline, implementing difference imaging to remove the static sky and a shift-and-stack algorithm to coherently boost the faint signals of moving asteroids. This work resulted in the first-ever published detections of their kind and established a new methodology that has since been adopted as the standard by other international collaborations.

I designed and led a new survey to monitor Sgr A*, the supermassive black hole at the center of our galaxy. This campaign required developing entirely new observing strategies for the SPT. The most critical innovation was a novel real-time calibration system that repurposed an internal instrument source to operate during science observations, allowing real-time detector monitoring. This produced a uniquely long-duration, multi-frequency light curve of Sgr A*, providing a rare benchmark for constraining models of black hole accretion physics. Analysis of these data is ongoing.

Technical Skills: Time Series & Fourier Analysis | Large-Scale Data Analysis (>600 TB) | Signal Processing & Noise Characterization | Analysis Pipeline Development

Telescope Operations and Survey Design

Guiding Question: How do we strategically design and execute large-scale astronomical surveys to answer new scientific questions?

A successful astronomical survey requires a holistic strategy that bridges high-level scientific goals with the operational realities of a complex instrument. My research involves designing these strategies, creating new methods for data collection, and ensuring the quality of final data products.

My project to monitor Sgr A* with the SPT camera SPT-3G, an instrument not originally designed for frequent targeted observations of a single source, required a rethinking of the telescope's operational mode. I designed the survey from the ground up, developing new schedules and real-time calibration methods to ensure the camera's detectors performed reliably and could be calibrated accurately. This work established a new operational mode for the SPT, demonstrating its capability for targeted time-domain astronomy.

My current research on an AI scheduling agent is the next evolution of this work. It aims to transform survey design from a static, human-driven plan into a dynamic, data-driven strategy. The AI agent will learn an optimal observing policy that adapts to changing weather and real-time data quality in order to maximize the scientific potential of the instrument over the course of a survey. This research is a foundational step toward fully autonomous observatories and represents a frontier of data-driven strategic planning in science.

Technical Skills: Experimental Survey Design | Data-Driven Strategic Planning | Systems Thinking & Operations | Instrument Calibration & Data Quality Assurance