Experiential Learning on Building Trustworthy Generative AI Systems: Consistency, Reliability, Explainability, Safety, and Trust (CREST)
By Dr. Manas Gaur
Course Objectives:
• Understand the challenges and limitations of current generative AI systems
• Master techniques for enhancing consistency in generative AI outputs
• Learn methods for improving reliability through knowledge integration and ensembling
• Implement user-level explainability techniques for generative AI
• Develop safety-oriented approaches for responsible AI deployment
• Apply the CREST framework to real-world applications
• Create NeuroSymbolic AI systems that combine the strengths of neural and symbolic approaches
| Week/Day | Topic |
|---|---|
| D01 |
Introduction to Generative AI & CREST framework • LLM capabilities, Challenges, NeuroSymbolic AI Knowledge sources |
| D02 |
Consistency challenges & Improving consistency • Paraphrasing, Knowledge-grounding, Evaluation methods, Self-consistency |
| D03 |
Ensemble approaches & Semi-deep ensembling • Knowledge infused learning, Shallow ensembling, Domain knowledge, Retrieval-augmentation |
| D04 |
Explainable AI & Evaluator pairing • User-level explainability, Attention visualization, Knowledge retrievers, Process knowledge |
| D05 |
Safety concepts & Process-guided safety • Red teaming, Contextual awareness, Safety constraints, Abstention |
| D06 |
Knowledge graphs & Knowledge-infused learning • Symbolic reasoning, KG-based LLMs, Mixture of experts, Performance maximization |
| D07 |
Healthcare AI challenges & Legal AI applications • Clinical knowledge, Healthcare safety, Legal reasoning, Legal explainability |
| D08 |
Bias detection & Fairness in AI • Attribution mechanisms, Ethics, Content attribution, Ethical guidelines |
| D09 |
Mental health considerations & Safety for vulnerable users • Clinical guidelines, Mental health explainability, Crisis detection, Human-AI collaboration |
| D10 |
Project Presentations & Emerging trends • Feedback, Research directions, Learning resources, Workshop conclusion |
Subject Matter Expert
Dr. Manas Gaur is an Assistant Professor in the Computer Science Department at the University of Maryland, Baltimore County (UMBC). His research focuses on the application of knowledge graphs, artificial intelligence, and natural language processing in areas such as conversational AI and recommendation systems. Prior to his academic role, Dr. Gaur served as a Senior Research Scientist at Samsung Research America and was a visiting researcher at the Alan Turing Institute in the United Kingdom. He completed his Ph.D. at the University of South Carolina under the guidance of Dr. Amit P. Sheth. His work has gained widespread attention and has been featured in media articles and podcasts.
