About

Aggelos K. Katsaggelos is the Joseph Cummings Professor of Electrical and Computer Engineering at Northwestern University, with courtesy appointments in Computer Science and Radiology. He serves as the Director of the Image and Video Processing Lab (IVPL), Co-Director of the Center for Scientific Studies in the Arts (NU-ACCESS), and Deputy Director of Computation at the Center for Computational Imaging and Signal Analytics in Medicine. His research interests encompass artificial intelligence, machine learning, multimedia signal processing, multimedia communications, computational imaging, computer vision, and medical and biological signal processing, including DNA signal processing and scientific studies in the arts. Katsaggelos has contributed to advancing methods in these fields through his extensive research, which impacts various societal and daily life aspects.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Medicine
• Internal medicine
• Algorithm
• Radiology
• Cartography
• Engineering
• Ecology
• Geography
• Data science
• Speech recognition
• Pathology
• Mathematics
• Statistics
• Emergency medicine
• Remote sensing

Selected publications

• MetaSR: Content-Adaptive Metadata Orchestration for Generative Super-Resolution

  arXiv (Cornell University) · 2026-04-29

  preprintOpen accessSenior author

  We study generative super-resolution (SR) in real-world scenarios where content and degradations vary across domains, genres, and segments. For example, images and videos may alternate between text overlays, fast motion, smooth cartoons, and low-light faces, each benefiting from different forms of side information. Existing metadata-guided SR methods typically use a fixed conditioning design, which is suboptimal when useful cues are content dependent and transmission budgets are limited. We propose MetaSR, a Diffusion Transformer (DiT)-based framework that selects and injects task-relevant metadata to guide SR under resource constraints. Specifically, we use the DiT's own VAE and transformer backbone to fuse heterogeneous metadata, and adopt an efficient distillation strategy that enables one-step diffusion inference. Experiments across diverse content buckets and degradation regimes show that MetaSR outperforms reference solutions by up to 1.0~dB PSNR while achieving up to 50\% transmission bitrate saving at matched quality. We assess these gains under a rate--distortion optimization (RDO) framework that jointly accounts for sender-side bitrate and receiver/display quality metrics (e.g., PSNR and SSIM).

  PublisherDOI

• Automatic Question Generation for Intuitive Learning Utilizing Causal Graph Guided Chain of Thought Reasoning

  arXiv (Cornell University) · 2026-01-02

  preprintOpen accessSenior author

  Intuitive learning is crucial for developing deep conceptual understanding, especially in STEM education, where students often struggle with abstract and interconnected concepts. Automatic question generation has become an effective strategy for personalized and adaptive learning. However, its effectiveness is hindered by hallucinations in large language models (LLMs), which may generate factually incorrect, ambiguous, or pedagogically inconsistent questions. To address this issue, we propose a novel framework that combines causal-graph-guided Chain-of-Thought (CoT) reasoning with a multi-agent LLM architecture. This approach ensures the generation of accurate, meaningful, and curriculum-aligned questions. Causal graphs provide an explicit representation of domain knowledge, while CoT reasoning facilitates a structured, step-by-step traversal of related concepts. Dedicated LLM agents are assigned specific tasks such as graph pathfinding, reasoning, validation, and output, all working within domain constraints. A dual validation mechanism-at both the conceptual and output stages-greatly reduces hallucinations. Experimental results demonstrate up to a 70% improvement in quality compared to reference methods and yielded highly favorable outcomes in subjective evaluations.

  PublisherDOI

• Hunting for new glitches in LIGO data using community science

  Journal of Physics Conference Series · 2026-02-01

  articleOpen accessSenior author

  Abstract Data from ground-based gravitational-wave detectors like LIGO contain many types of noise. Glitches are short bursts of non-Gaussian noise that may hinder our ability to identify or analyse gravitational-wave signals. They may have instrumental or environmental origins, and new types of glitches may appear following detector changes. The Gravity Spy project studies glitches and their origins, combining insights from volunteers on the community-science Zooniverse platform with machine learning. Here, we study volunteer proposals for new glitch classes, discussing links between these glitches and the state of the detectors, and examining how new glitch classes pose a challenge for machine-learning classification. Our results demonstrate how Zooniverse empowers non-experts to make discoveries, and the importance of monitoring changes in data quality in the LIGO detectors.

  PublisherDOI

• Hallucination-Free Automatic Question & Answer Generation for Intuitive Learning

  arXiv (Cornell University) · 2026-01-13

  preprintOpen accessSenior author

  Hallucinations in large language models (LLMs), defined as fluent yet incorrect or incoherent outputs, pose a significant challenge to the automatic generation of educational multiple-choice questions (MCQs). We identified four key hallucination types in MCQ generation: reasoning inconsistencies, insolvability, factual errors, and mathematical errors. To address this, we propose a hallucination-free multi-agent generation framework that breaks down MCQ generation into discrete, verifiable stages. Our framework utilizes both rule-based and LLM-based detection agents, as well as hallucination scoring metrics to optimize question quality. We redefined MCQ generation as an optimization task minimizing hallucination risk while maximizing validity, answerability, and cost-efficiency. We also introduce an agent-led refinement process that uses counterfactual reasoning and chain-of-thought (CoT) to iteratively improve hallucination in question generation. We evaluated a sample of AP- aligned STEM questions, where our system reduced hallucination rates by over 90% compared to baseline generation while preserving the educational value and style of questions. Our results demonstrate that structured multi-agent collaboration can mitigate hallucinations in educational content creation at scale, paving the way for more reliable LLM-powered learning tools.

  PublisherDOI

• Learning the Stellar Structure Equations via Self-supervised Physics-Informed Neural Networks

  arXiv (Cornell University) · 2026-04-06

  preprintOpen accessSenior author

  Stellar astrophysics relies critically on accurate descriptions of the physical conditions inside stars. Traditional solvers such as \texttt{MESA} (Modules for Experiments in Stellar Astrophysics), which employ adaptive finite-difference methods, can become computationally expensive and challenging to scale for large stellar population synthesis ($>10^9$ stars). In this work, we present an self-supervised physics-informed neural network (PINN) framework that provides a mesh-free and fully differentiable approach to solving the stellar structure equations under hydrostatic and thermal equilibrium. The model takes as input the stellar boundary conditions (at the center and surface) together with the chemical composition, and learns continuous radial profiles for mass $M_r(r)$, pressure $P(r)$, density $ρ(r)$, temperature $T(r)$, and luminosity $L_r(r)$ by enforcing the governing structure equations through physics-based loss terms. To incorporate realistic microphysics, we introduce auxiliary neural networks that approximate the equation of state and opacity tables as smooth, differentiable functions of the local thermodynamic state. These surrogates replace traditional tabulated inputs and enable end-to-end training. Once trained for a given star, the model produces continuous solutions across the entire radial domain without requiring discretization or interpolation. Validation against benchmark \texttt{MESA} models across a range of stellar masses yields a Mean Relative Absolute Error of $3.06\%$ and an average $R^2$ score of $99.98\%$. To our knowledge, this is the first demonstration that the stellar structure equations can be solved in a fully self-supervised and data-free fashion employing PINNs. This work establishes a foundation for scalable, physics-informed emulation of stellar interiors and opens the door to future extensions toward time-dependent stellar evolution.

  PublisherDOI

• Automatic Question Generation for Intuitive Learning Utilizing Causal Graph Guided Chain of Thought Reasoning

  ArXiv.org · 2026-01-02

  articleOpen accessSenior author

  Intuitive learning is crucial for developing deep conceptual understanding, especially in STEM education, where students often struggle with abstract and interconnected concepts. Automatic question generation has become an effective strategy for personalized and adaptive learning. However, its effectiveness is hindered by hallucinations in large language models (LLMs), which may generate factually incorrect, ambiguous, or pedagogically inconsistent questions. To address this issue, we propose a novel framework that combines causal-graph-guided Chain-of-Thought (CoT) reasoning with a multi-agent LLM architecture. This approach ensures the generation of accurate, meaningful, and curriculum-aligned questions. Causal graphs provide an explicit representation of domain knowledge, while CoT reasoning facilitates a structured, step-by-step traversal of related concepts. Dedicated LLM agents are assigned specific tasks such as graph pathfinding, reasoning, validation, and output, all working within domain constraints. A dual validation mechanism-at both the conceptual and output stages-greatly reduces hallucinations. Experimental results demonstrate up to a 70% improvement in quality compared to reference methods and yielded highly favorable outcomes in subjective evaluations.

  PublisherOA PDF

• Learning the Stellar Structure Equations via Self-supervised Physics-Informed Neural Networks

  arXiv (Cornell University) · 2026-04-06

  articleOpen accessSenior author

  Stellar astrophysics relies critically on accurate descriptions of the physical conditions inside stars. Traditional solvers such as \texttt{MESA} (Modules for Experiments in Stellar Astrophysics), which employ adaptive finite-difference methods, can become computationally expensive and challenging to scale for large stellar population synthesis ($>10^9$ stars). In this work, we present an self-supervised physics-informed neural network (PINN) framework that provides a mesh-free and fully differentiable approach to solving the stellar structure equations under hydrostatic and thermal equilibrium. The model takes as input the stellar boundary conditions (at the center and surface) together with the chemical composition, and learns continuous radial profiles for mass $M_r(r)$, pressure $P(r)$, density $ρ(r)$, temperature $T(r)$, and luminosity $L_r(r)$ by enforcing the governing structure equations through physics-based loss terms. To incorporate realistic microphysics, we introduce auxiliary neural networks that approximate the equation of state and opacity tables as smooth, differentiable functions of the local thermodynamic state. These surrogates replace traditional tabulated inputs and enable end-to-end training. Once trained for a given star, the model produces continuous solutions across the entire radial domain without requiring discretization or interpolation. Validation against benchmark \texttt{MESA} models across a range of stellar masses yields a Mean Relative Absolute Error of $3.06\%$ and an average $R^2$ score of $99.98\%$. To our knowledge, this is the first demonstration that the stellar structure equations can be solved in a fully self-supervised and data-free fashion employing PINNs. This work establishes a foundation for scalable, physics-informed emulation of stellar interiors and opens the door to future extensions toward time-dependent stellar evolution.

  PublisherOA PDF

• MetaSR: Content-Adaptive Metadata Orchestration for Generative Super-Resolution

  arXiv (Cornell University) · 2026-04-29

  articleOpen accessSenior author

  We study generative super-resolution (SR) in real-world scenarios where content and degradations vary across domains, genres, and segments. For example, images and videos may alternate between text overlays, fast motion, smooth cartoons, and low-light faces, each benefiting from different forms of side information. Existing metadata-guided SR methods typically use a fixed conditioning design, which is suboptimal when useful cues are content dependent and transmission budgets are limited. We propose MetaSR, a Diffusion Transformer (DiT)-based framework that selects and injects task-relevant metadata to guide SR under resource constraints. Specifically, we use the DiT's own VAE and transformer backbone to fuse heterogeneous metadata, and adopt an efficient distillation strategy that enables one-step diffusion inference. Experiments across diverse content buckets and degradation regimes show that MetaSR outperforms reference solutions by up to 1.0~dB PSNR while achieving up to 50\% transmission bitrate saving at matched quality. We assess these gains under a rate--distortion optimization (RDO) framework that jointly accounts for sender-side bitrate and receiver/display quality metrics (e.g., PSNR and SSIM).

  PublisherOA PDF

• Hallucination-Free Automatic Question & Answer Generation for Intuitive Learning

  ArXiv.org · 2026-01-13

  articleOpen accessSenior author

  Hallucinations in large language models (LLMs), defined as fluent yet incorrect or incoherent outputs, pose a significant challenge to the automatic generation of educational multiple-choice questions (MCQs). We identified four key hallucination types in MCQ generation: reasoning inconsistencies, insolvability, factual errors, and mathematical errors. To address this, we propose a hallucination-free multi-agent generation framework that breaks down MCQ generation into discrete, verifiable stages. Our framework utilizes both rule-based and LLM-based detection agents, as well as hallucination scoring metrics to optimize question quality. We redefined MCQ generation as an optimization task minimizing hallucination risk while maximizing validity, answerability, and cost-efficiency. We also introduce an agent-led refinement process that uses counterfactual reasoning and chain-of-thought (CoT) to iteratively improve hallucination in question generation. We evaluated a sample of AP- aligned STEM questions, where our system reduced hallucination rates by over 90% compared to baseline generation while preserving the educational value and style of questions. Our results demonstrate that structured multi-agent collaboration can mitigate hallucinations in educational content creation at scale, paving the way for more reliable LLM-powered learning tools.

  PublisherOA PDF

• Optimizing atrial fibrillation detection through ECG feature selection using Extra-Trees and statistical association measures

  Journal of Electrocardiology · 2026-01-24 · 1 citations

  articleOpen access

  INTRODUCTION: Atrial fibrillation (AFib) is the most prevalent abnormal heart rhythm, significantly increasing the risk of stroke and heart failure. Accurate and timely detection remains challenging, particularly due to the complexity of 12‑lead electrocardiogram (ECG) interpretation. While machine learning (ML) and deep learning (DL) models have demonstrated high accuracy in AFib detection, selecting the optimal input features is often non-trivial. This study aims to develop a hybrid feature selection methodology that objectively identifies the most discriminative ECG-based features for distinguishing AFib from normal sinus rhythm (NSR). MATERIAL & METHODS: We propose a hybrid framework that combines Extremely Randomized Trees (Extra-Trees) with statistical association measures to identify physiologically meaningful ECG features. Our analysis evaluates morphological, entropy-based and spectral hand-crafted features extracted from 12‑lead ECG recordings of patients who underwent catheter ablation for AFib. Two novel metrics, the feature importance score (FIS) and overall feature importance score (OFIS), are introduced to quantify feature relevance. RESULTS: The proposed approach ranked 97 extracted features and identified the 10 most important per ECG lead and 20 most relevant overall, with high consistency across leads. The interquartile range of RR-intervals achieved the highest normalized OFIS value (0.064), followed by other rhythm-related and entropy-based measures, confirming their strong discriminative power. The dimensionality of the feature space was thus reduced by nearly 80% while preserving interpretability and physiological meaning. CONCLUSIONS: This methodology provides a reproducible, interpretable and statistically grounded framework for ECG-based feature discovery, offering a preprocessing step for ML/DL models and aiding clinicians in real-time AFib detection.

  PublisherDOI

Recent grants

• Next-Generation Cardiovascular MRI powered by Artificial Intelligence

  NIH · $420k · 2021–2024

• Collaborative Research: HCC: Medium: Intelligent support for non-experts to navigate large information spaces

  NSF · $445k · 2021–2025

Frequent coauthors

• Rafael Molina

  253 shared

• Oliver Cossairt

  107 shared

• Javier Mateos

  Universidad de Granada

  92 shared

• G.M. Schuster

  Ostschweizer Fachhochschule OST

  64 shared

• Miguel Vega

  Universidad de Granada

  58 shared

• Manuel Ballester

  55 shared

• S. Derin Babacan

  48 shared

• Reza Borhani

  Northwestern University

  44 shared

Labs

• Image and Video Processing Lab (IVPL)PI

Education

• PhD, EE

  Georgia Institute of Technology

  1985

• MS, EE

  Georgia Institute of Technology

  1981

• Diploma, Electrical and Mechanical Engineering

  Aristotle University of Thessaloniki

  1979