Debugging and security analysis (Malware Analysis)
- 22 views
Talk by Andrew Honig. Ex-Google Security Head. Ex-NSA. Author of Practical Malware Analysis. LinkedIn
DISSERTATION DEFENSE
Department of Computer Science and Engineering
University of South Carolina
Author : Vipula Rawte
Advisor: Dr. Amit Sheth
Date: March 18, 2025
Time: 9:00 am – 11:00 am
Place: Zoom and AI Institute, Seminar Room 529
Meeting Link: https://sc-edu.zoom.us/j/83442966750
Meeting ID: 834 4296 6750
Abstract
Deception is inherent in human interactions, and AI systems increasingly exhibit similar tendencies, mainly through hallucinations - plausible yet incorrect outputs stemming from their design, memory limitations, and statistical nature. As AI progresses into Wave 2 - Generative AI, as outlined by Mustafa Suleyman in The Coming Wave, models like GPT and DALL-E are revolutionizing fields like healthcare and education. However, their rapid adoption brings misinformation, safety, and ethics challenges. Notable cases, such as Air Canada’s chatbot providing false information, highlight the real-world impact of AI hallucinations, a phenomenon so prevalent that hallucinate was named Cambridge Dictionary’s Word of the Year for 2023.
This dissertation tackles AI hallucinations through six key areas: (i) Characterization - developing a taxonomy and benchmark (HILT); (ii) Quantification - introducing evaluation metrics (HVI and HVI_auto); (iii) Detection - proposing a span-based Factual Entailment method to improve accuracy; (iv) Avoidance - creating techniques like “Sorry, Come Again?” (SCA) and [PAUSE] injection for better responses; (v) Mitigation - developing RADIANT, a retrieval-augmented framework for entity-context alignment; and (vi) Multi-modal - constructing VHILT and ViBe datasets for hallucination analysis in image-to-text and text-to-video models. This research makes generative AI more reliable and trustworthy by systematically addressing AI hallucinations.
Abstract
Reinforcement learning (RL) has emerged as a powerful paradigm for decision-making in complex environments. However, many RL tasks exhibit inherent structural properties—such as homogeneity, symmetry, and linguistic patterns—that are often underutilized, leading to inefficiencies in learning and generalization. This dissertation systematically exploits these structures to improve the efficiency, scalability, and robustness of RL algorithms across multi-agent and sequential decision-making settings.
Abstract
Natural Language Understanding (NLU) faces both opportunities and challenges as the amount of social media and healthcare data grows. This is particularly evident in context-sensitive applications such as evaluating cognitive health, identifying mental health symptoms, and monitoring drug abuse. Even though traditional NLU models work well for processing language in a wide range of areas, they often lack the ability to understand language in a specific domain, reason in context, and incorporate structured external knowledge. This dissertation talks about the Knowledge and Ontology Enhanced Approach to Natural Language Understanding (KOE-NLU), a new framework that is meant to make NLU systems better at understanding semantic depth, contextual awareness, and drawing conclusions in computational social media and healthcare informatics. The KOE-NLU framework is built upon three fundamental pillars: There are three types of methods. The first is Ontology-Driven Semantic Representation, which puts domain-specific knowledge into formalized ontologies to help with making sense of things based on their context and drawing conclusions from them. The second is Knowledge Graph Integration, which links different types of data sources to improve structured representation and reasoning over unstructured text. The third type is transformer-based language models with knowledge augmentation. This type of modeling uses domain-specific prompts to teach transformer based models, generative AI models structured knowledge sources. These components work in synergy to enable more accurate and context-aware natural language predictions. To demonstrate the efficacy of the KOE-NLU framework, this dissertation presents four major application areas. In the first place, the framework is used in cognitive health informatics to predict Mild Cognitive Impairment (MCI) by putting together MoCA test scores and automated discourse analysis. This is done by using a Cross-Cognitive Domain Attention (CCDA) model to pull out linguistic markers that show cognitive decline. Second, we present process knowledge-guided language prompting as a way to automate Main Concept Analysis (MCA) for discourse assessment. It is shown that process-aware language models are better than traditional text-based classifiers at detecting speech coherence in people with neurodegenerative conditions. Third, in mental health analytics, a knowledge-infused multi-task learning framework is developed to extract mental health symptoms linked to cardiovascular disease (CVD) from social media discourse, employing hierarchical attention networks combined with expert-curated knowledge bases. Finally, the dissertation introduces a Drug Abuse Ontology (DAO) that was created using semi-automated ontology engineering methods. We can use this ontology to identify patterns in substance use disorders, monitor illicit drug trends on social media, and examine the evolution of drug-related emotions over time. A rigorous experimental framework is implemented to evaluate the KOE-NLU models across these applications. The role of ontology-enhanced reasoning in model performance is looked at by comparing supervised and self-supervised learning paradigms. The results show that knowledge-infused transformer architectures do better than baseline deep learning models by as much as 17% in the F1 score and are better at interpreting clinical discourse in terms of context. Structured ontological constraints also make it much easier to classify substance use disorders and pull out mental health symptoms. This cuts down on false positives in automated detection systems by over 12%. This dissertation makes computational social media analytics and healthcare informatics better by giving us a knowledge-driven NLU framework that can be used on a large scale. This framework connects data-driven machine learning with symbolic AI methods. The KOE-NLU framework creates the foundation for the next generation of AI models that can be explained. These models will use process knowledge, ontological reasoning, and knowledge graphs to help computers understand language better in important healthcare settings. More work can be built on top of this by including speech, physiological, and neuroimaging signals, making domain-adaptive self-learning ontologies, and setting up real-time clinical decision support systems. This research marks a significant advancement in KOE-NLU, with important real-world applications in healthcare AI, clinical decision support, and public health monitoring on social media.
Below is a tentative agenda for the day. If you plan to attend the luncheon, please RSVP by at this link https://forms.office.com/r/2fayB7SPqA by Tuesday, March 4 to ensure we have an accurate lunch count.
Agenda:
10:00 am - 12:00 pm- Industrial Advisory Board meeting in room 2277
10:30 am - 12:00 pm- Even number graduate student poster session, 2nd floor hallway
12:00 pm - 1:30 pm- Lunch in room 2277
1:30 pm - 2:30 pm- Industrial Advisory Board meeting continues in room 2277
1:30 pm - 3:00 pm- Odd number graduate student poster session, 2nd floor hallway
3:00 pm - 4:00 pm - Talk and poster session winners, room 1400 (1st floor)
Number | Time | Presenting Author(s) | Poster Title |
AI18 | Morning | Rojina Panta | Finding Reaction Mechanism Pathways with Deep Reinforcement Learning and Heuristic Search |
AI19 | Morning | Mahsa Majdzadeh Ardakani | LLMPi: Optimizing LLMs for Realtime Conversational AI on Raspberry Pi |
AI20 | Morning | Bharath Muppasani | On Generalized Planning for Controlling Opinion Networks: Interpreting Human-AI Dialog States and Beliefs |
AI21 | Morning | Protik Nag | Targeted Layer Recalibration: Enhancing Interpretability with TCAV-Based Retraining |
AI22 | Morning | Sadman Sadeed Omee | Polymorphism-Aware Crystal Structure Prediction via Neural Network Potential and Genetic Algorithm with Adaptive Space Group Control |
CHS2 | Morning | Md Hasibul Amin | CrossNAS: A Cross-Layer Neural Architecture Search Framework for PIM Systems |
CHS3 | Morning | Suyash Vardhan Singh;Iftakhar Ahmad; | Resource Scheduler for Real-time Machine Learning |
CHS4 | Morning | Jinendra Malekar | PIM-LLM: A High-Throughput Hybrid PIM Architecture for 1-bit LLMs |
CV5 | Morning | Huaqiang Guo | An End-to-end Flooding Depth Detection System Based on SAM Model |
CV6 | Morning | Lili Wang | Diffusion-based CT Image Segmentation for Intracerebral Hemorrhage |
CV7 | Morning | Renjith Prasad | Pic2Prep:A multimodal conversational agent for cooking assistance |
CV8 | Morning | Hongpeng Yang | Unlocking Dark Vision Potential for Medical Image Segmentation |
HB10 | Morning | Alireza Bagheri Rajeoni | Deep Learning For Human Vascular Analysis |
HB11 | Morning | Hamed Abdollahi | Genome Methylation Array Analysis Using Iterative Random Forest Classification |
HB12 | Morning | Andrew Smith; Kuba Jerzmanowski | Monitoring Opioid Use Disorder Treatment Adherence Using Smartwatch Gesture Recognition |
HB13 | Morning | Christopher Lee | Reducing Diagnostic Barriers Using Neural Networks for Trypanosoma cruzi Detection in Blood Smears |
HB14 | Morning | Savannah Noblitt and Kayly Tran | The Creation and Implementation of a Novel Database for Medical School, Research Tracking and Student Research Opportunities |
QN32 | Morning | Erik Connerty | Predicting Chaotic Systems with Quantum Echo-state Networks |
QN33 | Morning | Rabins Wosti | Quantum fanout and GHZ states using spin-exchange interactions |
QN34 | Morning | Hasti Zanganeh, Blake Seekings | SpikeFed: Exploring the Landscape of Federated Learning for Spiking Neural Networks |
RA27 | Morning | Ritirupa Dey | Dynamic Role Transition in Human-in-the-Loop Robotic System with Neural Network Controllers |
RA28 | Morning | Mohammadreza Mohammadi; Aishneet Juneja | TRACE: Efficient Object Tracking via In-Sensor Compression and Near-Sensor Template-Matching |
RA29 | Morning | Chathurangi Shyalika | Time Series Foundational Models: Their Role in Anomaly Detection and Prediction |
RA30 | Morning | Jian Liu | Reinforcement Learning Driven Channel Switching Strategy for Connected Vehicles under Extreme Weather |
AI16 | Afternoon | Jinzhu Luo | Reinforcement Learning with Euclidean Data Augmentation for State-Based Continuous Control |
AI17 | Afternoon | Nitin Gupta, Bharath Muppasani | Towards Enhancing Road Safety in South Carolina Using Insights from Traffic and Driver-Education Data |
AI23 | Afternoon | Qinyang Li | Out-of-distribution materials property prediction using adversarial learning based fine-tuning |
AI24 | Afternoon | Yong Yang;Xiang Guan | Mimicking Aphasia through Controlled Neural Perturbations in Multimodal Large Language Models |
AI25 | Afternoon | Andrew Smith | Toward Concurrent Identification of Human Activities with a Single Unifying Neural Network Classification: First Step |
AI35 | Afternoon | Vansh Nagpal, Rae Sara Jones, Kausik Lakkaraju | Disseminating Authentic Public Messages using Chatbots - A Case Study with ElectionBot-SC to Understand and Compare Chatbot Behavior for Safe Election Information in South Carolina |
CHS1 | Afternoon | Tiffany Yu | Enhancing Security and Efficiency in Computational Hardware Through Intelligent Optimization |
HB9 | Afternoon | Savannah Noblitt | Web Application for Searching and Displaying Cancer Patient Data |
HB15 | Afternoon | Musa Azeem | Dihedral Angle Adherence: Evaluating Protein Structure Predictions in the Absence of Experimental Structures |
QN31 | Afternoon | Michael Cluver, Eli Hatcher | Efficient Quantum Circuit Synthesis with DeepXube |
RA26 | Afternoon | Misagh Soltani | Planning in Large Action Spaces Using Invertible Policy Networks |
DISSERTATION DEFENSE
Department of Computer Science and Engineering
University of South Carolina
Abstract
On the internet, where the number of available choices is exponentially growing, there is a need to prioritize and efficiently deliver relevant results to users, on demand. Recommendation systems (RSs) address this need by searching and filtering through large amounts of dynamically generated information and providing users with recommendations tailored to them. These systems have primarily focused on (1) single-user models, where recommendations are tailored towards a specific individual, or (2) single-item models, where items are recommended based on a broader appeal to users and similarities in item metadata, in the past. In many real-world scenarios, however, recommendation systems are often tailored to groups of users rather than individual users. This shift from single-user and towards group recommendation brings forth unique challenges such as addressing the diverse needs of a group, resolving conflicts and reaching consensus, assuring that groups receive fair recommendations without bias towards protected characteristics such as race, gender, and socioeconomic status; and ensuring that recommendations remain stable and adaptable even in response to dynamic and evolving environments. Such systems are growing increasingly relevant in disciplines such as team management, healthcare, online communities, and entertainment.
This dissertation explores and addresses the unique challenges of group recommendation systems (GRSs), such as increased computational complexity, varying group dynamics, and principles of trustworthy artificial intelligence (AI) related to bias, fairness, and robustness. The research makes contributions towards developing novel methods and metrics for GRSs, evaluating them with respect to performance through real-world case studies, such as team formation (project management) and meal planning (healthcare), and improving fairness and robustness of group recommendations. For team recommendation, the built system, called ULTRA, has been evaluated to demonstrate its effectiveness and generality in the context of research funding at academic institutions in the United States (US) and India. This work releases comprehensive datasets to aid in developing adaptable and reliable GRSs that optimize both performance and user satisfaction in dynamic environments. Beyond research, the dissertation is opening new avenues in entrepreneurship, intellectual property, and teaching resources.
DISSERTATION DEFENSE
Department of Computer Science and Engineering
University of South Carolina
Abstract
Wireless networks have become an integral aspect of our daily lives. Over the years, earlier generations of wireless networks have enabled some innovative applications, such as wireless gaming, fast internet browsing, and home automation, which were previously unattainable. However, to support emerging technologies like autonomous driving, virtual reality, telemedicine, and intelligent manufacturing, which require high data throughput and low latency, there is a need for advanced wireless networks. Legacy networks like WiFi/LTE, which operate below 6 GHz, have limited bandwidth and are insufficient to fulfill the high data throughput demands of several applications.
Millimeter-wave (mmWave) networks, operating between 30 GHz to 300 GHz, promise to offer a broader contiguous spectrum and low latency and might solve various applications’ data throughput and latency requirements. However, due to the directional nature of mmWave, it is prone to frequent outages while using Line-of-Sight (LoS) paths only, and the network deployment needs to consider Non-Line-of-Sight (NLoS) paths to provide reliable network connectivity. Additionally, due to the high operating frequency of mmWave network devices, these devices can also work as sensors to detect objects in their surroundings and work in all weather conditions. However, the specularity of mmWave and the weak reflectivity of objects cause most object details to be lost, making objects imperceptible. Besides, sharing a single mmWave device for networking and sensing applications reduces data throughput and sensing accuracy.
In this dissertation, we first design and evaluate a deep learning-based model to find optimal locations of 5G base stations (called “picocell”) for indoor and outdoor environments. Due to the lack of open-source data samples at high frequency, we collect actual data samples and verify our deep-learning models. Our model uses semantic features of the environment, such as the wall, ceiling, floor, and door, to make it generalizable and requires only a few data samples from the new environment to fine-tune, and it removes the need for an extensive site survey for optimal deployment. We then enable accurate sensing and high-throughput networking with a single mmWave device. We use conditional Generative Adversarial Networks to get an accurate depth image of the objects and residual networks to predict unobservable data samples from previously observed ones during networking. Finally, we design and implement a deep learning framework to detect the 3D bounding boxes of vehicles and pedestrians from mmWave devices in harsh weather conditions and compare their performance with vision-based methods. We use the non-anchor bounding box prediction method to enable 3D object detection for outdoor environments.
DISSERTATION DEFENSE
Department of Computer Science and Engineering
University of South Carolina
Author : Pingping Cai
Advisor: Dr. Song Wang
Abstract
3D computer vision is a promising research field with the potential to revolutionize future lifestyles. Among various 3D representation formats, point clouds stand out for their efficiency in depicting 3D objects using a set of coordinates, enabling advancements in fields such as autonomous driving, virtual reality, and robotics. Due to the limitations of sensor fields of view and scanning trajectories, the collected point clouds are usually sparse, noisy, and incomplete, impeding the performance of many downstream applications. Thus, the tasks of low-level point cloud processing are proposed to refine and generate dense, clean, and complete point clouds.
You are invited to attend an AI Roundtable event this Friday, November 1. This event is entitled “Ethical and effective use of AI in academic settings. How should we use AI effectively in our classrooms?”
We will have representatives from various departments across campus (Law/Academic Integrity/Library Sciences) and the panel will discuss the use of Gen AI in classrooms. You will have the opportunity to provide feedback to them as both faculty and students.
If you are interested in attending, please register here: https://forms.office.com/r/n4UznxruNg
Held online and in person at the AI Institute
Room 513, 1112 Greene St. Columbia, SC 29208 (Science and Technology Building)
10 am ET to 12 PM ET (refreshments will be provided)
*Limited to 50 in-person participants
DISSERTATION DEFENSE
Department of Computer Science and Engineering
University of South Carolina
Author : Fengyao Yan
Advisor : Dr. Yan Tong, Dr. Jijun Tang
Date : Oct 24th, 2024
Time: 9:30 am
Place : Zoom Meeting
Link: https://us05web.zoom.us/j/82857019481?pwd=ueKsnxBVTLySbXb4yj4Z93pAzb7va…
Meeting ID: 828 5701 9481
Passcode: 747882
One tap mobile
+12532158782,,82857019481#,,,,*747882# US (Tacoma)
+13462487799,,82857019481#,,,,*747882# US (Houston)
Dial by your location
• +1 253 215 8782 US (Tacoma)
• +1 346 248 7799 US (Houston)
• +1 669 444 9171 US
• +1 669 900 9128 US (San Jose)
• +1 719 359 4580 US
• +1 253 205 0468 US
• +1 507 473 4847 US
• +1 564 217 2000 US
• +1 646 558 8656 US (New York)
• +1 646 931 3860 US
• +1 689 278 1000 US
• +1 301 715 8592 US (Washington DC)
• +1 305 224 1968 US
• +1 309 205 3325 US
• +1 312 626 6799 US (Chicago)
• +1 360 209 5623 US
• +1 386 347 5053 US
Meeting ID: 828 5701 9481
Passcode: 747882
Find your local number: https://us05web.zoom.us/u/leCEnSLjPT
Abstract
Genes can be treated as a graph that can be mapped. Tremendous information is coded in genes to ensure a complex functioning organism. Decoding this information is critical to understanding our biology and developing treatments for various diseases including cancer. Deep learning, a new branch of computer science, has gained traction over the past decade. It offers more insight into the data that is processed by the deep-learning models. Our study has shown that deep-learning models can be an effective tool in decoding genetic data such as gene tissue-deconvolution, gene graph mapping and genomic imputation. In tasks such as tissue deconvolution, our research has demonstrated the superior capability of deep learning-based approaches in capturing sample variations compared to traditional numerical analytical methods. While our approach requires large relevant datasets for effective deep learning training, this challenge can be addressed with increasing data availability. In gene graphing and mapping, our Graph Neural Network based approach consistently outperforms traditional regression techniques by a significant margin. The primary challenge here lies in the demand for substantial computing resources; however, the ongoing growth in average computing power and the enhanced accessibility of computational resources are expected to alleviate this constraint over time. Moreover, in the realm of generating and imputing missing biological data, cutting-edge generative AI models have proven to be invaluable. We are actively exploring the potential of generative AI to aid in imputing common missing biological data such as gene expression or methylation states. Overall, the evolution of advanced deep learning models has introduced fresh perspectives and possibilities to the field of biology and medicine, albeit accompanied by certain challenges. By addressing these challenges, deep learning models exhibit remarkable efficacy in resolving complex biomedical issues. In the foreseeable future, these advancements hold the promise of unveiling novel biological insights and facilitating the development of innovative treatments, thereby propelling biomedical research forward and ultimately benefiting humanity.
Zoom and AI Institute, Seminar Room 529