Talk by David Weston. VP of Security at Microsoft. LinkedIn.

DISSERTATION DEFENSE

Author : Revathy Venkataramanan Chandrasekaran
Advisor: Dr. Amit Sheth
Date: March 21, 2025
Time: 10:30 am
Place: Zoom and AI Institute, Seminar Room 529
Meeting Link: https://sc-edu.zoom.us/j/8440139296
Meeting ID: 844 013 9296

Abstract

Can I eat this food or not, and why? Which AI pipeline is best for a task and dataset? These questions differ from factual questions and answering tasks as they involve processes with interacting entities. Recipes consist of ingredients, methods, and interactions, while AI pipelines include datasets, models, and tasks. Each entity must be analyzed independently, and a collective inference, known as compositional reasoning, is required to draw the conclusion.

Existing process recommendation methods rely on the availability of structured data but struggle with unstructured data like recipes and AI pipelines. These datasets are often lengthy and noisy, making it hard to capture interactions and derive relevant insights. Additionally, natural language descriptions don’t provide necessary domain knowledge. For example, recipes don’t state that potatoes are healthy carbs with a high glycemic index. Domain-specific knowledge is needed for effective analysis and recommendations.
While neural networks excel in pattern recognition, they struggle with compositional reasoning. This work introduces a neurosymbolic framework for explainable process recommendation using Dynamic Multimodal Process Knowledge Graphs (DMPKGs). DMPKGs provide structured process representations grounded in multi-contextual knowledge for reasoning, explainability, and traceability while utilizing neural networks for pattern recognition. They enable modular entity inference and capture interactions for dynamic decision-making. DMPKGs allow continuous updates and store multimodal data, improving recommendation accuracy and explainability. Two use cases, recipe suitability analysis and AI pipeline recommendation, are explored to demonstrate the effectiveness of this approach in process recommendation.

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.

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)

Posters:

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.