About

Klara Nahrstedt is the Swanlund Endowed Chair and Grainger Distinguished Chair in Engineering at the University of Illinois Urbana-Champaign, serving as a Professor in the Department of Computer Science and the Director of the Coordinated Science Laboratory. Her research interests are directed toward trustworthy multimedia distributed systems and networking, including quality of service (QoS) and resource management in Internet and mobile systems, real-time security in wireless networks for trustworthy power grids, edge-cloud systems, cyber-physical system security for electric vehicles, health systems, 3D tele-immersive systems, and advanced edge-cloud-based cyber-infrastructures for scientific instruments. She has received numerous awards such as the IEEE Communication Society Leonard Abraham Award, the IEEE Computer Society Technical Achievement Award, the ACM SIGMM Technical Achievement Award, and the Robert Piloty Prize, among others. Nahrstedt has held editorial positions in prominent journals and has been involved in organizing major conferences, including serving as general co-chair for ACM Multimedia 2006 and IEEE PerCom 2009. She is a Fellow of ACM, IEEE, and AAAS, a member of the Leopoldina German National Academy of Sciences, and a member of the National Academy of Engineering. Her academic background includes a Diploma in mathematics from Humboldt University, Berlin, and a Ph.D. in Computer and Information Science from the University of Pennsylvania.

Research topics

• Computer Science
• Computer Security
• Engineering
• Operating system
• Computer network
• Real-time computing
• Risk analysis (engineering)
• Business
• Data science
• Human–computer interaction
• Internet privacy
• World Wide Web

Selected publications

• Adaptive Unknown Fault Detection and Few-Shot Continual Learning for Condition Monitoring in Ultrasonic Metal Welding

  arXiv (Cornell University) · 2026-04-15

  preprintOpen access

  Ultrasonic metal welding (UMW) is widely used in industrial applications but is sensitive to tool wear, surface contamination, and material variability, which can lead to unexpected process faults and unsatisfactory weld quality. Conventional monitoring systems typically rely on supervised learning models that assume all fault types are known in advance, limiting their ability to handle previously unseen process faults. To address this challenge, this paper proposes an adaptive condition monitoring approach that enables unknown fault detection and few-shot continual learning for UMW. Unknown faults are detected by analyzing hidden-layer representations of a multilayer perceptron and leveraging a statistical thresholding strategy. Once detected, the samples from unknown fault types are incorporated into the existing model through a continual learning procedure that selectively updates only the final layers of the network, which enables the model to recognize new fault types while preserving knowledge of existing classes. To accelerate the labeling process, cosine similarity transformation combined with a clustering algorithm groups similar unknown samples, thereby reducing manual labeling effort. Experimental results using a multi-sensor UMW dataset demonstrate that the proposed method achieves 96% accuracy in detecting unseen fault conditions while maintaining reliable classification of known classes. After incorporating a new fault type using only five labeled samples, the updated model achieves 98% testing classification accuracy. These results demonstrate that the proposed approach enables adaptive monitoring with minimal retraining cost and time. The proposed approach provides a scalable solution for continual learning in condition monitoring where new process conditions may constantly emerge over time and is extensible to other manufacturing processes.

  PublisherDOI

• LLM-ADAM: A Generalizable LLM Agent Framework for Pre-Print Anomaly Detection in Additive Manufacturing

  ArXiv.org · 2026-05-05

  articleOpen access

  Additive manufacturing (AM) continues to transform modern manufacturing by enabling flexible, on-demand production of complex geometries across diverse industries. Fused filament fabrication (FFF) has extended AM to laboratories, classrooms, and small production environments, but this accessibility shifts process-planning responsibility to users who may lack manufacturing expertise. A syntactically valid slicer profile can still encode thermally or geometrically harmful settings, and subtle G-code edits can alter extrusion, cooling, or adhesion before a print begins. Pre-print G-code screening catches accidental or adversarial machine-program errors before material or machine time is wasted. This paper proposes LLM-ADAM as a generalizable LLM framework for pre-print anomaly detection in AM. The framework decomposes the task into three roles: Extractor-LLM maps a G-code file to a structured process-parameter schema; Reference-LLM converts printer and material documentation into aligned operating ranges; and Judge-LLM interprets a deterministic deviation table and G-code evidence to decide whether a part is non-defective or belongs to an anomaly class. Printers, materials, and LLM backbones are interchangeable test conditions, not fixed assumptions. We evaluate the framework on an N=200 FFF G-code corpus spanning two desktop printer families, two materials, and five classes including non-defective, under-extrusion, over-extrusion, warping, and stringing. The best framework configuration reaches 87.5% accuracy, compared with 59.5% for the strongest engineered single-LLM baseline. The results show that structured decomposition, rather than backbone strength alone, is the dominant source of improvement, with defect classes identified at or near ceiling for leading configurations while residual errors concentrate on conservative false alarms for non-defective samples.

  PublisherOA PDF

• Circuit Tracing in Vision-Language Models: Understanding the Internal Mechanisms of Multimodal Thinking

  Open MIND · 2026-02-23

  preprint

  Vision-language models (VLMs) are powerful but remain opaque black boxes. We introduce the first framework for transparent circuit tracing in VLMs to systematically analyze multimodal reasoning. By utilizing transcoders, attribution graphs, and attention-based methods, we uncover how VLMs hierarchically integrate visual and semantic concepts. We reveal that distinct visual feature circuits can handle mathematical reasoning and support cross-modal associations. Validated through feature steering and circuit patching, our framework proves these circuits are causal and controllable, laying the groundwork for more explainable and reliable VLMs.

  DOI

• Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning

  ArXiv.org · 2026-05-21

  articleOpen access

  Spreadsheet systems (e.g., Microsoft Excel, Google Sheets) play a central role in modern data-centric workflows. As AI agents grow increasingly capable of automating complex tasks, such as controlling computers and generating presentations, building an AI-driven spreadsheet agent has emerged as a promising research direction. Most existing spreadsheet agents rely on specialized prompting over general-purpose LLMs; while this design has potentials on simple spreadsheet operations, it struggles to manage the complex, multi-step workflows typical of real-world applications. We introduce Spreadsheet-RL, a reinforcement learning (RL) fine-tuning framework designed to train specialized spreadsheet agents within a realistic Microsoft Excel environment. Spreadsheet-RL features an automated pipeline for scalable collection of paired start-goal spreadsheets from online forums, as well as domain-specific evaluation tasks in areas such as finance and supply chain management, which we compile into the new Domain-Spreadsheet benchmark dataset. It also includes a Spreadsheet Gym environment designed for multi-turn RL: Spreadsheet Gym exposes extensive Excel functionality through a Python sandbox, along with a refined harness that incorporates a comprehensive tool set and carefully designed tool-routing rules for spreadsheet tasks. Through comprehensive experiments, we show that Spreadsheet-RL substantially enhances AI agent's performance on both general and domain-specific spreadsheet tasks: it improves Qwen3-4B-Thinking-2507's Pass@1 on SpreadsheetBench from 12.0% to 23.4%, and raises Pass@1 from 8.4% to 17.2% on our curated Domain-Spreadsheet dataset. These results highlight Spreadsheet-RL's strong potential for generalization and real-world adoption in spreadsheet automation, and broadly, its promise for advancing LLM-based interactions with data interfaces in everyday work.

  PublisherOA PDF

• Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning

  arXiv (Cornell University) · 2026-05-21

  preprintOpen access

  Spreadsheet systems (e.g., Microsoft Excel, Google Sheets) play a central role in modern data-centric workflows. As AI agents grow increasingly capable of automating complex tasks, such as controlling computers and generating presentations, building an AI-driven spreadsheet agent has emerged as a promising research direction. Most existing spreadsheet agents rely on specialized prompting over general-purpose LLMs; while this design has potentials on simple spreadsheet operations, it struggles to manage the complex, multi-step workflows typical of real-world applications. We introduce Spreadsheet-RL, a reinforcement learning (RL) fine-tuning framework designed to train specialized spreadsheet agents within a realistic Microsoft Excel environment. Spreadsheet-RL features an automated pipeline for scalable collection of paired start-goal spreadsheets from online forums, as well as domain-specific evaluation tasks in areas such as finance and supply chain management, which we compile into the new Domain-Spreadsheet benchmark dataset. It also includes a Spreadsheet Gym environment designed for multi-turn RL: Spreadsheet Gym exposes extensive Excel functionality through a Python sandbox, along with a refined harness that incorporates a comprehensive tool set and carefully designed tool-routing rules for spreadsheet tasks. Through comprehensive experiments, we show that Spreadsheet-RL substantially enhances AI agent's performance on both general and domain-specific spreadsheet tasks: it improves Qwen3-4B-Thinking-2507's Pass@1 on SpreadsheetBench from 12.0% to 23.4%, and raises Pass@1 from 8.4% to 17.2% on our curated Domain-Spreadsheet dataset. These results highlight Spreadsheet-RL's strong potential for generalization and real-world adoption in spreadsheet automation, and broadly, its promise for advancing LLM-based interactions with data interfaces in everyday work.

  PublisherDOI

• Adaptive Unknown Fault Detection and Few-Shot Continual Learning for Condition Monitoring in Ultrasonic Metal Welding

  ArXiv.org · 2026-04-15

  articleOpen access

  Ultrasonic metal welding (UMW) is widely used in industrial applications but is sensitive to tool wear, surface contamination, and material variability, which can lead to unexpected process faults and unsatisfactory weld quality. Conventional monitoring systems typically rely on supervised learning models that assume all fault types are known in advance, limiting their ability to handle previously unseen process faults. To address this challenge, this paper proposes an adaptive condition monitoring approach that enables unknown fault detection and few-shot continual learning for UMW. Unknown faults are detected by analyzing hidden-layer representations of a multilayer perceptron and leveraging a statistical thresholding strategy. Once detected, the samples from unknown fault types are incorporated into the existing model through a continual learning procedure that selectively updates only the final layers of the network, which enables the model to recognize new fault types while preserving knowledge of existing classes. To accelerate the labeling process, cosine similarity transformation combined with a clustering algorithm groups similar unknown samples, thereby reducing manual labeling effort. Experimental results using a multi-sensor UMW dataset demonstrate that the proposed method achieves 96% accuracy in detecting unseen fault conditions while maintaining reliable classification of known classes. After incorporating a new fault type using only five labeled samples, the updated model achieves 98% testing classification accuracy. These results demonstrate that the proposed approach enables adaptive monitoring with minimal retraining cost and time. The proposed approach provides a scalable solution for continual learning in condition monitoring where new process conditions may constantly emerge over time and is extensible to other manufacturing processes.

  PublisherOA PDF

• Revisiting Disaggregated Large Language Model Serving for Performance and Energy Implications

  2026-04-27

  articleSenior author
  PublisherDOI

• LLM-ADAM: A Generalizable LLM Agent Framework for Pre-Print Anomaly Detection in Additive Manufacturing

  arXiv (Cornell University) · 2026-05-05

  preprintOpen access

  Additive manufacturing (AM) continues to transform modern manufacturing by enabling flexible, on-demand production of complex geometries across diverse industries. Fused filament fabrication (FFF) has extended AM to laboratories, classrooms, and small production environments, but this accessibility shifts process-planning responsibility to users who may lack manufacturing expertise. A syntactically valid slicer profile can still encode thermally or geometrically harmful settings, and subtle G-code edits can alter extrusion, cooling, or adhesion before a print begins. Pre-print G-code screening catches accidental or adversarial machine-program errors before material or machine time is wasted. This paper proposes LLM-ADAM as a generalizable LLM framework for pre-print anomaly detection in AM. The framework decomposes the task into three roles: Extractor-LLM maps a G-code file to a structured process-parameter schema; Reference-LLM converts printer and material documentation into aligned operating ranges; and Judge-LLM interprets a deterministic deviation table and G-code evidence to decide whether a part is non-defective or belongs to an anomaly class. Printers, materials, and LLM backbones are interchangeable test conditions, not fixed assumptions. We evaluate the framework on an N=200 FFF G-code corpus spanning two desktop printer families, two materials, and five classes including non-defective, under-extrusion, over-extrusion, warping, and stringing. The best framework configuration reaches 87.5% accuracy, compared with 59.5% for the strongest engineered single-LLM baseline. The results show that structured decomposition, rather than backbone strength alone, is the dominant source of improvement, with defect classes identified at or near ceiling for leading configurations while residual errors concentrate on conservative false alarms for non-defective samples.

  PublisherDOI

• Circuit Tracing in Vision-Language Models: Understanding the Internal Mechanisms of Multimodal Thinking

  arXiv (Cornell University) · 2026-02-23

  articleOpen access

  Vision-language models (VLMs) are powerful but remain opaque black boxes. We introduce the first framework for transparent circuit tracing in VLMs to systematically analyze multimodal reasoning. By utilizing transcoders, attribution graphs, and attention-based methods, we uncover how VLMs hierarchically integrate visual and semantic concepts. We reveal that distinct visual feature circuits can handle mathematical reasoning and support cross-modal associations. Validated through feature steering and circuit patching, our framework proves these circuits are causal and controllable, laying the groundwork for more explainable and reliable VLMs.

  PublisherOA PDF

• LLM-Powered Data Annotation for Bridging the Semantic Gap in Air Quality Monitoring

  2025-11-11

  articleOpen accessSenior author

  Accurately annotating raw sensor data with Air Quality Index (AQI) categories presents significant challenges in traditional approaches, primarily due to the fundamental semantic gap between low-level sensor readings and high-level air quality interpretations. Rule-based systems require perfect domain expertise and are prone to labeling errors and inconsistencies, while supervised machine learning models demand extensive labeled datasets and computational resources. This work explores an alternative paradigm by leveraging Large Language Models (LLMs) as virtual annotators to bridge this semantic gap, enabling direct interpretation of raw sensor data without explicit formulaic programming. We develop a comprehensive framework using GPT-3.5 Turbo to classify AQI categories from multi-variate sensor data collected from Chicago's Array of Things network. Our systematic evaluation reveals several key findings: the LLM achieves 82% average accuracy using only pollutant data in zero-shot settings, improves to 88% when prompts are supported with bootstrap examples, and gains an additional 4% performance boost when augmented with environmental context (temperature, humidity). These results, validated through principled statistical analysis, demonstrate that LLMs can effectively overcome the semantic gap to comprehend complex sensor patterns and provide reliable AQI annotations. Our work establishes the viability of LLMs as scalable, context-aware virtual annotators for environmental monitoring, offering a promising solution to overcome the limitations of traditional annotation methods without relying on complex prompt-engineering or model fine-tuning.

  PublisherOA PDFDOI

Recent grants

• CC* Integration-Large: MAINTLET: Advanced Sensory Network Cyber-Infrastructure for Smart Maintenance in Campus Scientific Laboratories

  NSF · $1.0M · 2021–2025

• CC*Integration: BRACELET: Robust Cloudlet Infrastructure for Scientific Instruments' Lifetime Connectivity

  NSF · $500k · 2017–2021

• CNS Core: Medium: Collaborative Research: Scalable Dissemination and Navigation of Video 360 Content for Personalized Viewing

  NSF · $400k · 2019–2025

• EAGER: Collaborative Research: Augmented 360 Video for Situation Awareness in Firefighting

  NSF · $150k · 2021–2024

• NR: Resilient Internet Routing Framework

  NSF · $300k · 2003–2007

Frequent coauthors

• Ralf Steinmetz

  Technical University of Darmstadt

  38 shared

• Wanmin Wu

  Chengdu University of Information Technology

  31 shared

• Ahsan Arefin

  Google (United States)

  30 shared

• Yuan Xue

  Xi'an Jiaotong University

  27 shared

• Long Vu

  27 shared

• Wenbo He

  26 shared

• Raoul Rivas

  Intel (United Kingdom)

  26 shared

• Baochun Li

  University of Toronto

  24 shared

Education

• Ph.D., Computer Science

  University of California, Berkeley

  1995

• M.S., Computer Science

  University of California, Berkeley

  1991

• B.S., Computer Science

  University of California, Berkeley

  1988

Awards & honors

• IEEE Communication Society Leonard Abraham Award for Researc…
• 2008 University Scholar Award
• 2009 Humboldt Research Award
• 2012 IEEE Computer Society Technical Achievement Award
• 2014 ACM SIGMM Technical Achievement Award