[AAAI 2025] [LQ03] Developing explainable multimodal AI models with hands-on lab on the life-cycle of rare event prediction in manufacturing

Introduction

Welcome to our interactive lab session at AAAI 2025! 🎉 This lab is designed to provide participants with hands-on experience in tackling the challenges of rare event prediction in manufacturing. Through a structured approach covering data generation, preprocessing, model development, and evaluation, attendees will gain practical insights into multimodal AI modeling and its explainability.

Lab Schedule

Part 1 - Introduction and Overview

• Overview of rare event prediction and anomaly prediction in smart manufacturing.
• Discussion of objectives and expected takeaways from the lab.

Part 2 - Exploring Datasets for Rare Event Analysis

• Overview of publicly available datasets for rare event prediction.
• Deep dive into manufacturing-specific datasets (Pulp-and-paper manufacturing and Future Factories (FF) datasets) and its role in rare event prediction.

Part 3 - Addressing Data Scarcity and Improving Data Quality

• Hands-on session on using data augmentation techniques and modeling approaches to improve rare event prediction.

Part 4 - Model Selection, Development, and Evaluation

• Hands-on session with anomaly prediction using robust modeling techniques, including dependency capture between rare events.

Part 5 - Development and Use of Process Ontology

• Introduction to process ontology for rare event prediction.
• Hands-on session on using process ontology for generating user-level explanations.

Summary and Q/A

• Recap of the session and open discussion with participants.

---

Prerequisites

Participants are expected to have a basic understanding of:

• AI and Machine Learning concepts.
• Python programming and familiarity with libraries such as PyTorch, TensorFlow, and Scikit-learn.
• Basic knowledge of data processing techniques.

Resources

• Presentation slides
• Additional reading materials
• Datasets used in this lab

---

FAQs

Q1: Do I need prior experience with rare event prediction or multimodal AI?
A: While prior experience is beneficial, it is not mandatory. The lab is structured to guide participants through key concepts and hands-on exercises.

---

Contributors

• Chathurangi Shyalika (University of South Carolina)
• Ruwan Wickramarachchi (University of South Carolina)
• Revathy Venkataramanan (University of South Carolina)
• Dhaval Patel (IBM Research)
• Amit Sheth (University of South Carolina)

---

Contact Information

For any questions or feedback, please reach out to:
📎 Chathurangi Shyalika: jayakodc@email.sc.edu
📎 Ruwan Wickramarachchi: ruwan@email.sc.edu
📎 Revathy Venkataramanan: revathy@email.sc.edu
📎 Dhaval Patel: pateldha@us.ibm.com
📎 Amit Sheth: amit@sc.edu

---

Contributing to the Lab

We encourage contributions and feedback to improve the lab and its materials.

---

Citation

If you wish to cite this lab, please use the following format:

Chathurangi Shyalika, Ruwan Wickramarachchi, Revathy Venkataramanan, Dhaval Patel, & Amit Sheth (2025, February). Developing Explainable Multimodal AI Models: A Lab on the Lifecycle of Rare Event Prediction in Manufacturing. In AAAI Conference on Artificial Intelligence.

@article{shyalika2025lab,
  title={Developing Explainable Multimodal AI Models: A Lab on the Lifecycle of Rare Event Prediction in Manufacturing},
  author={Shyalika, Chathurangi and Wickramarachchi, Ruwan and Venkataramanan, Revathy and Patel, Dhaval and Sheth, Amit},
  booktitle={AAAI Conference on Artificial Intelligence},
  year={2025}
}