Week of AI Schedule

Monday, March 31, 2025

---

---

---

---

---

---

---

---

---

Tuesday, April 1, 2025

---

---

---

Research Round-Up

noon to 12:45 p.m.

Amistad Conference Room (SP2 – 12.216), Virtual

Session

A Perspective on AI in Astrophysics, Publishing and the Humanities, Presented by Roger Malina

Former professional NASA astrophysicist, I have of experience in data analytics and algorithmic analysis, now often called AI.  40 years’ experience as Executive Editor of the Leonardo Publications at MIT press, I have experience of peer reviewing and fake publishing. At UT Dallas, I am part of a team developing, with humanities author Fred Turner AI, software called Fred the Heretic. Its results are based on a database of everything Turner has written in the 8o years of his life. I will compare and contrast the use of AI in astrophysics, publishing industry and the humanities.  I will argue that we should start developing AI that fails the Turing Test because we need a variety of ‘intelligences’ to solve future problems. I am on the editorial board of the special issue of Leonardo with Tsinghua University in China on “Augmented Wisdom”. I elaborate Fred Turner’s insights on how his best poetry had been heavily influenced by his ‘private life’ that we have not yet designed into AI. Prompts often rely on private information not available on the internet eg the dreams that a human has while asleep, and the level of trust one develops with people one has worked with for decades. AI is not yet an ‘adult”. Intended audience will be people struggling to integrate AI into their daily life in UT Dallas. I will summarize my experience and will try to develop new focus area for our Phase 2 of development of Fred the Heretic AI at the Off-Center for Emergence Studies at UT Dallas.

Audience: Graduate Students; Undergraduate Students; Faculty; Staff; visitors; 

---

Session, Research Presentation

Transcribing In-Home Natural Language Recordings with an Automatic Speech Recognition Model (ASR), Presented by Rowan Soo Levick

Aims to inform researchers and students interested in language research methods about advantages and disadvantages of Automatic Speech Recognition Models in comparison to human transcribing for studying natural language recordings. 

Audience: Faculty; Undergraduate Students; Graduate Students; Staff; 

---

Session

High-Throughput Chemical Screening Using Machine Learning, Presented by Meghraj Magadi Shivalingaiah

Artificial intelligence and machine learning have revolutionized scientific discovery, enabling data-driven approaches for material and chemical property predictions. This proposal focuses on leveraging machine learning techniques to predict and design molecules with high optical absorption, enabling advancements in biomedical imaging, materials science, and optoelectronics. By integrating experimental data from UV-Vis-NIR spectroscopy and cheminformatics databases such as ChEMBL and PubChem, we aim to develop predictive models that can accurately correlate chemical structures with their optical properties. This will significantly accelerate molecular discovery, replacing labor-intensive trial-and-error approaches with AI-driven design.

Audience: Undergraduate Students; Graduate Students; 

---

Session

Nanosheet LDMOS Study Through TCAD Simulation and Bayesian Optimization, Presented by Elham Mashhadi

The work aims to reduce incredibly the specific on-resistance RSP(on) in low voltage applications, achieving up to a five-time reduction in RSP(on) compared to conventional LDMOS structures[Ali] while maintaining or even enhancing slightly competitive breakdown voltage (BV). Consequently, results illustrate impressive increase in vital factor of efficiency for power devices represented by Baliga (FOM)[Baliga]. We combine TCAD simulations with Bayesian optimization to explore a large design space and tune device parameters, demonstrating a novel process for designing highly efficient power devices.

Audience: Graduate Students; Faculty; Undergraduate Students; Staff; 

---

---

---

---

---

Wednesday, April 2, 2025

---

---

---

Research Round-Up

1 to 1:45 p.m.

Amistad Conference Room (SP2 – 12.216), Virtual

• Large Language Models for Market Research: A Data-Augmentation Approach – Presented by Selene M. Wang 
  LLMs have revolutionized AI and market research, particularly in conjoint analysis. Traditional surveys face scalability and cost issues, making LLM-generated data a promising alternative. However, biases exist between LLM-generated and human data. This paper proposes a statistical data augmentation approach to integrate LLM-generated data with real data, using transfer learning to debias the LLM data with a small amount of human data. Validated through studies on COVID-19 vaccine preferences and sports car choices, this method reduces estimation error and saves data and costs by 24.9% to 79.8%. LLM-generated data complements human responses within a robust statistical framework.
• From Data to Dialogue: Using Artificial Intelligence and Natural Language Processing tools to Teach Wildfire Risk Management -Presented by Dr. Steven Haynes
  This presentation explores using AI, particularly NLP, to enhance property risk management education. By analyzing public comments on wildfire risks in Texas, the study used sentiment analysis and topic modeling to convert qualitative data into educational content. Sentiment analysis revealed public emotions on wildfires, insurance, governance, climate change, and preparedness. Topic modeling identified themes like political influence and financial issues. Integrating AI insights into classroom presentations connected theoretical concepts with real-world perceptions, engaging students and promoting critical discussions. This project demonstrates AI’s potential as a transformative educational tool, linking theory to practice and enhancing student comprehension.
• ConfliBERT: An LLM for Political Event Detection – Presented by Patrick Brandt 
  Most conflict event data are expensively coded by humans from news reports. This project relies on recent advances in artificial intelligence (AI) and large language models (LLM) to address that problem, and builds on earlier NSF efforts that created a publicly available large language model to study inter- and intra-state conflict, called ConfliBERT (Hu et al., 2022). This project expands ConfliBERT to multilingual settings, including Arabic and Spanish. It also focuses on creating network data for individuals, groups, locations, and events. 
• Using ML for cell segmentation -Presented by Nikhil Inturi & Hemanth Mydugolam
  This presentation explores the application of YOLO v11 for cell segmentation in Xenium transcript images, focusing on detecting neurons and delineating cell boundaries with high precision. We will discuss the challenges of segmenting cellular structures in transcriptomic data, the advantages of using YOLO v11’s real-time object detection capabilities, and its performance in comparison to traditional segmentation methods. The talk will highlight model training, evaluation metrics, and potential applications in neuroscience research and single-cell analysis. 
  
  Audience: Faculty; Graduate Students; Undergraduate Students; Staff

---

Research Round-Up

2 to 2:45 p.m.

Amistad Conference Room (SP2 – 12.216), Virtual

Session

• Intrusion Detection System Using AI Agents – Presented by Kunal Mukherjee
  Advanced Persistent Threats (APTs) infiltrate organizations through stealthy, prolonged attacks with catastrophic consequences. System Provenance-based Intrusion Detection Systems (PIDS) are crucial for comprehensive surveillance and defense against APTs. They capture information about system resources and their interactions, forming a provenance graph that models dynamic behaviors. PIDS are uniquely positioned to detect and thwart stealthy attacks. Ensuring transparency, reliability, and accountability in PIDS is essential. A new agentic framework for PIDS uses large language models (LLMs) to analyze threat reports, query provenance databases, and provide explainability, scalably detecting attacks and summarizing explanations from threat reports.
• Tafsiri: Revolutionizing Podcast Accessibility with AI – Presented by Sir Mbwika 
  Tafsiri is an AI-powered software developed to transcribe and translate podcasts from ArtSciLab’s Creative Disturbance Publishing project into over 100 languages, enhancing global accessibility. It uses the seamlessM4T V2 model for speech-to-speech translation, ensuring non-English speakers can access CDP’s content. Tafsiri democratizes knowledge, promoting inclusivity and diversity. It highlights AI’s role in content accessibility, showcases technological integration, and encourages cross-disciplinary collaboration. Tafsiri aims to inspire industry professionals, educators, students, and university staff to explore AI applications in solving global challenges. This project exemplifies AI’s potential to make digital content more inclusive.
• MedCeptor: How AI Can Revolutionize Medical First Responder Training – Presented by Ivan Tong
  MedCeptor is an AI-powered virtual EMT field training officer that bridges the gap between textbook knowledge and real-world application. It offers 24/7 access to realistic medical emergencies, dynamic skill assessments, and personalized feedback through professionally curated practice scenarios. MedCeptor ensures effective learning by requiring students to perform medical calls from start to finish, including interventions, differential diagnosis, and gathering medical history. This practical and accessible training method does not require expensive VR simulations or in-person FTOs. MedCeptor aims to revolutionize EMT training and the healthcare industry, with potential expansion to paramedic, nursing, and medical schools.
• Beyond the Gaze: Rethinking the Representation of Genocide in Media and AI – Presented by Maruf Rahman 
  My dissertation proposal studies the discourses and scholarships of the photographic and documentary representation of the Holocaust and conducts a comparative study on how the perpetrator’s/ Nazi’s gaze is still predominant in the visual representation of the Rohingya genocide both in mainstream media and digital media. For my dissertation, I am creating a six-to-eight-minute co-created documentary using video clips recorded by the victims of the Rohingya genocide. My second project examines the representation of the Holocaust and the Rohingya genocide in digital media and examines the future of the Holocaust studies and challenges harnessed from generative AI for Holocaust and Rohingya genocide testimony. 
  
  Audience: Faculty; Graduate Students; Undergraduate Students; Staff

---

---

---

Thursday, April 3, 2025

---

Research Round-Up

Session

10 to 10:45 a.m.

Amistad Conference Room (SP2 – 12.216), Virtual

• Fred The Heretic (FTH) AI – Presented by Venkatesh Prasad Ravichandran
  AI has advanced in creative fields like poetry. Paul Fishwick’s chatbot, FredTheHeretic, mimics Dr. Frederick Turner’s style. Evaluating AI-generated poetry is subjective, making comparisons with human-written poetry difficult. Traditional methods rely on subjective impressions, complicating meaningful comparisons. 
• Deep Optical Sensing – Presented by Fan Zhang
  Optical sensors measure the information of light, such as intensity, polarization state, wavelength or spectrum. They are traditionally mono-functional, bulky, and inefficient. By integrating advances in moiré quantum materials and deep learning, we demonstrate an all-in-one intelligent sensing scheme for light in mid-infrared regime at 79 K. By leveraging the top-gate and bottom-gate tunability of two-dimensional materials, each incoming light produces a unique nonlinear response map, which encodes all the light information. A trained convolutional neural network is shown to be able to decode the power, wavelength, and polarization state of the light simultaneously and instantaneously with a high precision level. This new scheme not only identifies a pathway for future intelligent sensing technologies in an extremely compact, on-chip manner but also opens a new horizon for deep quantum sensing schemes that can be generalized to include other frequency regimes.
• How Do NGOs Really Work? – Presented by Elizabeth A.M. Searing 
  Nonprofits, or NGOs, have been in the news recently. Despite public records, our understanding has been limited to finances for 45 years. AI now allows access to written records, revealing how nonprofits deliver services without owners or taxes. Projects include categorizing US nonprofits by SDGs, learning UK nonprofit investment rules, and mapping funding in US and Canadian nonprofits. Understanding nonprofits is crucial before applying data science. 
• The Art of Intelligence: Integrating Lessons from AI’s Commonsense Knowledge Problem – Presented by Lisa Whitsett  
  This talk examines the commonsense knowledge problem in AI history, showing how early assumptions shape current research. Early AI assumptions prioritized abstract reasoning over perception. Melanie Mitchell identifies a fallacy in AI: assuming easy tasks for humans are easy to program. Biomimicry shifts focus to embodied, ecological intelligence but retains anthropocentric assumptions. Revisiting the commonsense knowledge problem reveals AI’s narrow views of cognition and anthropocentric bias, suggesting a more ethically informed approach engaging diverse intelligences. Insights from biomimicry, embodied cognition, and multispecies studies invite us to see AI as part of a larger ecological and cognitive web. 
• Leveraging AI to Assess Social Attention in Young Autistic Children – Presented by Erin E. Kosloski 
  This presentation discusses using AI to analyze social attention in autistic toddlers. An AI model was trained on over 10,000 video frames from the Rollins Autism Corpus. Preliminary results showed limited success due to “neurotypical bias,” but the new model, fine-tuned with autism-specific data, significantly improved gaze detection accuracy. This research highlights the need for a more inclusive reference database. Supported by a UT Dallas Seed Program, the presentation is designed for a general audience. 

---

---

---

---

---

---

---

---

---

Friday, April 4, 2025

---

---

---

---

---

Watch: AI Showcase

1 to 1:45 p.m.

Amistad Conference Room (SP2 – 12.216), Virtual

Session

Audience: Staff; Graduate Students; Undergraduate Students; Faculty

• Leveraging Cursor.ai for Rapid Web Development and Data Processing – Presented by Tarun Raghu 
  AI has advanced in creative fields like poetry. Paul Fishwick’s chatbot, FredTheHeretic, mimics Dr. Frederick Turner’s style. Evaluating AI-generated poetry is subjective, making comparisons with human-written poetry difficult. Traditional methods rely on subjective impressions, complicating meaningful comparisons. 
• Enhancing Student Success with AI: A Live Demo of Campus Connect – Presented by Daniel Febres 
  Campus Connect is an AI-driven platform that personalizes student success by recommending campus resources, scholarships, and support services. The interactive demo will show how AI can revolutionize student engagement with data-driven recommendations and a chatbot for instant answers. Attendees will see how AI-driven personalization can improve student retention, academic support, and campus experience. Key topics include finding campus resources faster, personalized scholarship recommendations, and data-driven engagement strategies. Objectives are to demonstrate AI’s impact on student success, improve campus resource utilization, gather feedback, and explore partnerships with universities and AI researchers. 
• The Role of AI in Revolutionizing the Entertainment / Casino Industry – Presented by Srikanth Srinivas
  The integration of Artificial Intelligence (AI) in the casino industry is driving significant transformation, enhancing both the gaming experience and operational efficiency. This presentation explores how AI is reshaping various aspects of the casino environment, from personalized player experiences to advanced security measures. AI-driven systems are improving customer service through intelligent chatbots, providing targeted marketing strategies, and offering real-time player behavior analytics. In terms of security, AI-powered facial recognition and fraud detection tools are safeguarding both physical and online casinos.  
• Bubble-Bursting and Hype-Busting – Navigating The Realities of AI Profitability  – Presented by Paul Nichols

---

---

Research Round-Up

Session

2 to 2:45 p.m.  

Amistad Conference Room (SP2 – 12.216), Virtual

 Audience: Staff; Graduate Students; Undergraduate Students; Faculty 

• AI in Sustainability – Machine Learning enabled Trash Collectors – Presented by Areeb Iqbal 
  Waste management poses a significant ecological threat due to non-recyclable trash in landfills. This presentation explores AI/ML applications in waste management, focusing on robots and object detection models. It covers the problem statement, types of waste, disposal methods, and the role of robots in waste collection. The research proposes energy-efficient robots for deployment on campuses or large events, simplifying complex ideas for both beginners and experienced individuals. 
• Advancing AI in Surgery: Real-Time Computer Vision for Decision Support and Training – Presented by Sara Juneja
  At the Schumacher lab at Tufts University, we are developing AI algorithms to analyze real-time surgical videos. Our goal is to train models to recognize key anatomical structures, track workflows, and identify high-risk scenarios, enhancing patient safety and outcomes. Our research has shown AI can accurately identify critical structures like the pulmonary artery and inferior pulmonary vein. This innovation aims to improve surgical precision, reduce complications, and support trainee education through AI-assisted feedback. 
• Modeling social attributes of dynamic faces with deep neural networks -Presented by Suvel Muttreja 
  This proposal explores the application of machine learning (ML) and deep This study investigates if DCNNs can replicate human-like social attribute judgments from dynamic faces. A behavioral experiment with 12 students showed consistent human trustworthiness judgments but almost no correlation with DCNN representations. This suggests that while DCNNs excel at identity recognition, they fail to capture human-like social inferences. The findings highlight the need for models incorporating human-like cognitive frameworks, especially for ethical AI use in hiring or law enforcement.
• Research Presentation (Faculty/Student AI research review) by Presented by Wangzhou, Andi
  This proposal explores the application of machine learning (ML) and deep learning techniques to uncover the mechanisms of pain using medical and transcriptomic data from human donor tissues. By analyzing large-scale biological and neurophysiological datasets, our approach aims to identify novel pain biomarkers and potential drug development targets. This research will enhance our understanding of pain processing in humans, facilitating the discovery of more effective, targeted therapies for pain management.
• CBTI (Cognitive Behavioral Therapy for Insomnia) with AI -Presented by Mohsin Maqbool, MD.
  AI can revolutionize insomnia treatment by enhancing Cognitive Behavioral Therapy for Insomnia (CBTI) with AI-powered chatbots and personalized machine learning models. This approach offers real-time, adaptive therapy, expanding CBTI’s reach and optimizing patient outcomes. I propose presenting this at UT Dallas Week of AI to highlight the integration of neuroscience, behavioral therapy, and technology.

---

---