AI In Research & Education

AI Expertise: Machine Learning algorithms, Reinforcement Learning, Optimization (discrete and continuous), Duality Theory, Probability and Randomization, Provable guarantees in algorithm design, Game Theory, Algorithmic Mechanism Design

Research Interests: Foundations of AI and Machine Learning, Multi-Agent Systems, Large Language Models, Computational Economics, Auctions, Blockchains, Computational Chemistry

Research Ideas and Projects:
My research centers around foundational aspects of algorithms and their applications, which is manifested in the design and analysis of algorithms with provable guarantees on their properties and used resources (such as correctness, running time, communication, randomization, etc). Examples of related projects:

• Truthful Combinatorial Auctions: The design of online auctioning systems where potential buyers express their complex preferences over bundles of goods/services and the auction designer has to design an auction that is computationally efficient, approximately optimizes the system’s social welfare and is truthful, i.e., it incentivises the buyers to declare their preferences truthfully. I have been using Reinforcement Learning, Randomization and Non-cooperative Game Theory to design such auctioning systems with provable guarantees.
• Blockchain Mechanism Design: Blockchains are examples of decentralized distributed systems which are used to record and process information, and executing computer programs (called smart contracts) in a trustworthy way, without presence of third parties. I have been using Game Theory (such as Equilibrium Theory) and Randomization to design blockchain mechanisms and protocols which are decentralized, communication efficient and have provable guarantees.
• Computational Chemistry: I have been designing continuous and discrete optimization algorithms for the solid crystal structure prediction problems, aiming at the computational design of new solid materials on the atomic level. These explorations involve the development of new optimization methods based on mathematical programming (e.g., quadratic programming), enhancements of gradient descent algorithms and design and implementation of second order approximation methods to compute locally optimal solutions to non-convex energy functions.

AI Expertise: Machine learning and its applications, with a focus on adversarial machine learning, generative models, and AI for systems.

Projects: Training-free test-time method designed to enhance the zero-shot adversarial robustness of vision-language models (VLMs). While VLMs exhibit impressive zero-shot capabilities, they remain vulnerable to adversarial perturbations that lead to significant prediction error. Our approach introduces a cosine-guided adaptive anchor movement-based framework, which operates during inference by shifting input features towards a "feature anchor" derived from noise-perturbed inputs. This method effectively corrects adversarial examples without compromising clean accuracy, achieving state-of-the-art robust and clean accuracy without any model retraining.

AI Expertise: Machine Learning, Deep Learning, Generative AI, Computer Vision, Sequence Models

Research Interests: Machine learning/deep learning, Biomedical signal processing, Medical imaging, Medical Devices, Neuromorphic computing

Research Topics:
AI algorithms to detect/predict neurological disorders. In this project, our goal is to improve and automate the detection and prediction of neurological disorders by encompassing different modalities. These modalities are physiological signals (ECG, EMG, EEG, etc.), imaging data (CT, MRI, NIRS, PET, etc.), and omics data. Integration of these modalities gives a full and comprehensive picture of brain disorders and greatly augments reliable data extraction.
AI in cancer imaging. The main focus of this project is to develop novel integrated deep learning models that will help in early diagnosis of cancer through carrying out localization, segmentation, and classification tasks on MRI/CT scans. Additionally, we develop risk assessment models that predict cancer comorbidity like cardiovascular disease associated with some cancer treatment modalities.
AI tools to accelerate drug discovery. In this project, we develop new AI-based methodologies to identify promising drug candidates that raise the hit rate through predicting the binding affinity between molecules and proteins. The goal is to find molecules that can chemically bind to the target proteins and modulate them so that they no longer contribute to the disease of interest.

AI Expertise and Interests: AI-Driven Forensics, NLP Techniques, Large Language Models, Knowledge Graphs, Topic modelling techniques
 
Projects: AI-Powered Historical Warrant Analysis
Funding Source: NSF DASS 2131509 (Oct-1st, 2021 till Sep-30, 2025) 
Description: This project automates the retrospective analysis of search warrant documents using AI and retrieval-augmented generation to extract patterns in digital evidence requests and judicial approval processes across thousands of historical cases. The system enables evidence-based legal scholarship by making novel large-scale empirical warrant studies feasible.

AI Expertise:

• High-Performance Computing (HPC) and distributed systems design for scalable AI and data analytics.
• Blockchain and distributed ledger technologies for secure, auditable, and privacy-preserving computation.
• Distributed data management and consistency protocols for scientific and multidimensional data.
• Integration of trustworthy and resilient AI with large-scale computing and storage systems.

Research Interests:

• Trustworthy AI, connecting AI reliability, interpretability, and verifiability through blockchain-based provenance and accountability mechanisms.
• Federated and privacy-preserving learning across heterogeneous and resource-constrained computing environments.
• Scalable AI pipelines on HPC and cloud-edge continuum for scientific and healthcare data analytics.
• Efficient and secure data sharing frameworks enabling cross-institutional AI collaboration.

Research Ideas:
My current direction is to bridge trustworthy AI with distributed ledger and HPC infrastructures. I aim to design blockchain-assisted AI systems that ensure data integrity, transparency, and robustness against adversarial behavior while maintaining the scalability of scientific and healthcare workloads. A near-term goal is to develop lightweight consensus and validation protocols that make federated and decentralized AI training verifiable and efficient across compute clusters and edge nodes.

Projects:

Title: CRII: CSR: Enhancing Eventual Data Consistency in Multidimensional Scientific Computing through Lightweight In-Memory Distributed Ledger System
Funding Agency: National Science Foundation (NSF)
Description:
This project develops a lightweight distributed ledger system to ensure consistency, fault tolerance, and verifiable provenance in multidimensional scientific data environments. While primarily focused on distributed systems, its techniques contribute directly to trustworthy AI by enabling reliable data pipelines and transparent model training environments over large-scale computing infrastructures.

AI Expertise and Interest: Human-AI Interaction, Explainable AI, AI (Machine Learning and Large Language Models) System Development

Research Ideas: My research integrates natural language processing (NLP), AI system development, and algorithmic explainability. I focus on building machine learning and large language model–based systems that are explainable and transparent, and on understanding how people communicate and collaborate with these intelligent systems.

AI Expertise and Interest: Applied AI, which includes fine-tuning models for specific tasks; Natural Language Processing
Research Ideas: Cognitive-aware generative LLM for training and education. Multimodal deep learning models applied to scientific tasks. 

AI Expertise: Development and application of deep neural networks for neuroimaging data analysis. Integration of multimodal Large Language Models (LLMs) for multimodal data analysis. Advancement of quantum machine learning technologies for drug discovery. Development of gradient-based randomized optimizers for efficient model training.

Research Interests: 

• Applying AI techniques to analyze medical images, including MRI, fMRI, and CT.
• Using multimodal LLMs to assist in interpreting clinical imaging and textual data.
• Exploring quantum machine learning approaches for drug design, educational data mining, and transportation optimization.

Research Topics and Collaborations:

• Collaborating with physicians at the Medical College of Georgia (MCG) to optimize radiosurgery dose planning.
• Developing a hybrid LLM-driven framework and “Medical Imaging-of-Thought” to assist physicians in lesion interpretation within medical images.
• Partnering with external researchers to design novel quantum machine learning models that support clinicians in determining optimal radiosurgery dosage.

 Research Ideas:

I aim to develop intelligent sensing systems that combine physical sensors with advanced AI to create more natural human-computer interactions. One promising direction is using LLMs to interpret acoustic and physiological signals for continuous health monitoring while preserving user privacy. I'm also interested in building AI systems that can understand complex human behaviors through multimodal sensing, with applications ranging from human computer interaction to security systems that are resilient against emerging threats like Deepfake and Generative AI. These systems would leverage both traditional deep learning approaches and newer LLM-based methods to provide robust solutions.

AI Expertise: 

• Deep learning and neural network architectures for biometric security, authentication, and deepfake detection 

• Multimodal AI systems that integrate and interpret acoustic, visual, and sensor data streams 

• Large Language Model (LLM) applications for sensor data interpretation and time-series analysis 

• AI-driven wearable/earable computing for mobile healthcare and activity recognition 

Research Interests: 

• Foundation models for acoustic signal processing and physical world understanding 

• LLM-based reasoning for multimodal sensor fusion in healthcare and precision agriculture 

• Privacy-preserving AI for ambient intelligence and continuous monitoring systems 

• Embodied AI for human behavior understanding and context-aware interactions 

Research Ideas:  

I aim to develop intelligent sensing systems that combine physical sensors with advanced AI to create more natural human-computer interactions. One promising direction is using LLMs to interpret acoustic and physiological signals for continuous health monitoring while preserving user privacy. I'm also interested in building AI systems that can understand complex human behaviors through multimodal sensing, with applications ranging from human computer interaction to security systems that are resilient against emerging threats like Deepfake and Generative AI. These systems would leverage both traditional deep learning approaches and newer LLM-based methods to provide robust solutions. 

 

AI Interests: Learning-enabled cyber-physical systems.

Research Projects:

1. NSF CAREER: Enabling Trustworthy Upgrades of Machine-Learning Intensive Cyber-Physical Systems, https://www.nsf.gov/awardsearch/show-award?AWD_ID=2143351. This project targets unique machine-learning-intensive CPS upgrade challenges by developing scalable verification and monitoring methods for upgrades as well as safe upgrade procedures to enable trustworthy upgrades and achieve lifetime safety assurance in machine-learning-intensive CPS.
2. NSF CPS: Small: Data-Driven Modeling and Control of Human-Cyber-Physical Systems with Extended-Reality-Assisted Interfaces, https://www.nsf.gov/awardsearch/show-award?AWD_ID=2223035. This project will enable the synergistic integration of data-driven modeling and control methods such as neural networks, reinforcement learning, and model-based methods such as hybrid systems, and model predictive control. This project will also explore the benefits of extended reality (XR) in building human-machine interfaces for effective communication and interaction between human users, machines, and environments.
3. NSF Collaborative Research: SLES: Foundations of Qualitative and Quantitative Safety Assessment of Learning-enabled Systems, https://www.nsf.gov/awardsearch/show-award?AWD_ID=2331938. This project targets the foundational challenges of developing qualitative and quantitative safety assessment methods capable of capturing uncertainties from environments and providing timely, comprehensive, and accurate safety evaluations at the system level.

 

AI Expertise: Machine learning theory, Provable algorithmic robustness, Learning with data efficiency.

Research Interests: Machine learning in the presence of data corruptions, Designing algorithms with provable data efficiency, Interactive learning with queries.

Research Ideas:

• We study interactive machine learning with adaptive queries​, where the queries come in different forms and could be potentially corrupted. In this project, we apply rigorous analysis on how machine learning algorithms utilize different types of queries effectively and efficiently, with asymptotic upper and lower bounds.
• We consider fundamental machine learning problems such as robust classification tasks via nonlinear hypothesis classes. Robust learning problems have been at central of machine learning theory in the past decades. We follow this research line and design algorithms tolerant to different types of noise, e.g. Massart, adversarial label noise, or malicious noise. In addition, classification via nonlinear function classes is challenging due to the nature of nonlinearity, especially when modern data sets are usually corrupted with noise. In this project, we seek necessary and appropriate learning conditions under which robust and efficient algorithms can be established.
• We consider the fundamental multiclass learning problems and design algorithms with provable guarantees. While binary classification has been well understood by the machine learning theory community, extensions to multiclass learning scenarios are falling short. In this project, we seek sound theory for multiclassification by properly extending the existing literature of binary classification with robustness and data-efficiency guarantees.

AI Expertise: Side-channel analysis (acoustic, cache, and microarchitectural leakage), Machine learning for detecting and classifying side-channel patterns (feature engineering, supervised and unsupervised methods), Red-team research: discovering and characterizing novel attack vectors

Research Interests:

• Using ML to discover previously unknown information-leakage patterns and weaknesses
• Using ML to improve existing side channel attacks
• Translating offensive findings into practical defenses, mitigations, and secure design guidance
• Applying AI for threat hunting and attribution, including lawful use against criminal infrastructure

Research Ideas: 

Every computational process produces observable side effects. Each process consumes resources such as power, memory, cache, and CPU time, and may also generate heat, noise, electromagnetic emissions, and network traffic. We study these side effects to gain insights into the underlying process and the data being handled. By applying artificial intelligence, we detect subtle patterns in these signals to design and launch more effective and sophisticated attacks—primarily to expose vulnerabilities and strengthen defensive mechanisms.

AI Expertise: Data visualization and visual analytics, Human–AI teaming, Uncertainty quantification and visualization, Explainable AI (XAI)

Research Interests: Large language models (LLMs), Knowledge graphs, AI-assisted diagnosis and decision support 

Research Ideas: 

• Natural-language interface for data analysis: Use LLMs as the user interface to lower the technical barrier for data analysis. Build a chatbot-based workflow where users issue natural language commands to generate charts aligned with their analytical intents.
• Multimodal agent support for diagnosis: Orchestrate computer-vision and LLM agents to assist disease diagnosis. CV models recommend regions of interest; LLMs parse physicians’ requests to refine/manipulate ROIs and retrieve visually similar cases. The system is designed for human-in-the-loop labeling and diagnostic reasoning, with provenance tracking and calibrated uncertainty to reduce hallucinations and overconfidence.

 

AI Expertise: Intelligent User Interfaces, Incremental Learning, AI-supported XR, Genetic Algorithms, Interdisciplinary Research 

Research Interests: Practical XR, Eye tracking, Healthcare, Visualization, Telepresence

Research Ideas: 

Dr. Orlosky would like to explore AI-enabled XR systems, particularly vision language models for object detection and interaction, hybrid vision-visualization systems, semi-autonomous telepresence, and agentic learning systems for languages and STEM. 

My research focuses on the theoretical and algorithmic foundations of Artificial Intelligence and Machine Learning, particularly in:
- Optimization and reinforcement learning theory
- Multi-agent and distributed learning
- Robust machine learning
- Generative modeling and change detection

AI Expertise: 

• Evidential uncertainty quantification and reasoning for complex structural data
• Reliable and generalizable graph representation learning
• Trustworthy reasoning and explainability in large language and vision models

Research Interests: 

• Robust AI for out-of-distribution detection and generalization.
• Uncertainty quantification and reasoning in structural data.
• Trustworthy AI for downstream applications, including but not limited to autonomous driving, remote sensing, and health informatics.

Research Ideas: 

Dr. Yu’s research focuses on building trustworthy AI systems that stay reliable across real-world settings. Her work studies how uncertainty arises and shapes the behavior of complex models, including large language and vision models, multi-agent systems, and data with structured signals such as hyperspectral images and MRI scans. She also investigates uncertainty in varied learning settings, including active learning and continuous learning. Her goal is to classify, explain, and address key sources of uncertainty so that AI systems can make stable and reliable decisions.

Expertise related to AI: security and adversarial Machine Learning, AI-driven Topic Prediction, forensic data provenance

Expertise related to AI: Cybersecurity and AI, Machine Learning, Reinforcement Learning, Big Data, Smart Computing, Cloud Computing, Internet of Things

Expertise related to AI: Mobile Computing and AI, Machine Learning, Distributed Sensing, Internet of Things

Expertise related to AI: AI-driven mobile computing, ubiquitous computing, wireless sensor networks, sensor-enhanced computing, edge computing 

Expertise related to AI: AI and Computer Vision, assistive technologies, social media, open-source software, pedagogy and educational technology

Expertise related to AI: AI Health Assistants, AI-driven Telemedicine, information security, mobile payment systems

Expertise related to AI: AI-driven information technology, information quality, mis/disinformation, cybersecurity