January 20, 2026 (4-5 pm EST) | Self-Supervised Learning Workshop Slides Recording
Self-Supervised Learning (SSL) is a powerful machine learning paradigm that enables models to learn directly from unlabeled data by predicting parts of the data from other parts of the same input. This workshop introduces the fundamentals of SSL, explains why it has become essential for modern AI, and demonstrates how SSL supports strong representation learning for downstream tasks such as classification, segmentation, anomaly detection, and forecasting. This session includes real-world case studies (with notebooks), including SSL with LoRA fine-tuning and an application in subglacial mapping.
Facilitator: Dr. Bayu Tama,
Research Assistant Professor, iHARP, UMBC
Co-Facilitator: Dr. Tolulope Ale,
Data Scientist, Microsoft
January 27, 2026 (2-4 pm EST) | Reproducible AI: GitHub & Version Control Workshop Slides Recording
AI Reproducibility can be defined as the ability to utilize similar code and data to regenerate research findings. Alternatively, it can be perceived as achieving the same outcomes with distinct data and methods. With significant research in AI depicted by growth in number of publications, attempts to reproduce research findings are often met with challenges. These include accessibility to research artifacts such as code, communication hindrances with authors, time constraints, external randomness, under-reporting of hyper-parameters, inconsistency in computing environments, to mention a few. This workshop introduces the importance of AI reproducibility and takes a hands-on approach on leveraging version control systems like GitHub, which thus enable tracking changes in research artifacts like code, data, models, hyperparameters, dependencies, among others.
Facilitator: Dr. Josephine Namayanja,
Research Associate Professor & Executive Director, iHARP, UMBC
Co-Facilitator: Omar Faruque,
PhD Candidate, iHARP, UMBC
February 17, 2026 (4-5 pm EST) | Foundational Models in AI Workshop Slides Recording
This workshop introduces the fast-growing world of foundation models and shows how the same ideas are now transforming Earth and climate science through geo-foundation models. We start with an accessible overview of what foundation models are, why they matter, and how they differ from traditional machine learning. Then we’ll focus on geo-foundation models trained on massive collections of remote sensing data (satellite and aerial imagery), exploring what they can do for tasks such as land-cover mapping, change detection, and environmental monitoring. Along the way, we’ll discuss the basic building blocks (encoders, self-supervised learning) at a high level, practical challenges in working with satellite data, and open research questions. The session will conclude with a short live demo using the IBM–NASA geospatial foundation model on Hugging Face, plus pointers to tools and resources, aimed at researchers and students from all backgrounds, not just computer science.
Facilitator: Dr. Jianwu Wang,
Professor, Department of Information Systems, UMBC
Co-Facilitator: Mostafa Cham,
PhD Candidate, iHARP, UMBC
March 10, 2026 (4-5 pm EST) | Generative AI Workshop Slides Recording
Generative AI has rapidly advanced from early probabilistic models to modern approaches such as variational autoencoders, generative adversarial networks, and diffusion models. While widely known for generating images and text, these methods are increasingly important for scientific applications. This lecture provides a concise overview of the generative AI landscape, covering key concepts, major model families, and recent breakthroughs. It also highlights emerging applications in climate science, geoscience, and polar research. The talk will discuss both the opportunities and challenges of applying generative AI in scientific settings and explore how these techniques may help enable the next generation of data-driven Earth system modeling.
Facilitator: Dr. Sanjay Purushotham,
Associate Professor, Department of Information Systems, UMBC
March 24, 2026 (4-5 pm EST) | Explainable AI Workshop Slides Recording
eXplainable AI (XAI) is a crucial suite of tools and techniques designed to help human users understand, trust, and interpret the results produced by machine learning algorithms. This workshop explores the spectrum of XAI, from interpretable “glass-box” models like decision trees to post-hoc “black-box” explanations for complex deep learning systems. Participants will learn the essential role XAI plays in mitigating ethical bias, enabling scientific discovery in complex climate or genomics data, and fostering the trust necessary for AI adoption in critical fields like medicine and finance. The session provides a comprehensive taxonomy of prominent XAI approaches—including model-agnostic methods like LIME and SHAP, model-specific techniques such as Grad-CAM and Attention Maps, and causal or rule-based explanations—while introducing key evaluation metrics like faithfulness, stability, and compactness. Concluding with hands-on exercises, the workshop demonstrates how these techniques can be applied to real-world tasks, such as resolving semantic ambiguity in LLMs or identifying critical features in computer vision.
Facilitator: Dr. Osman Gani,
Assiatant Professor, Department of Information Systems, UMBC
Co-Facilitator: Emam Hossain,
PhD Candidate, iHARP, UMBC
March 31, 2026 (4-5 pm EST) | Physics-Informed Neural Networks Workshop Slides Recording
Estimating the future mass balance of the ice sheets is challenging due to partial information. Despite significant improvements in numerical modeling and the availability of data, we still lack a sufficient number of observations needed for ice sheet modeling and have a limited understanding of the physical processes. Physics-Informed Neural Networks (PINNs) are a viable alternative to traditional numerical methods as they close the data gap by learning from the fundamental laws of conservation in addition to observed data. Using PINNs we can learn complex, nonlinear relationships between variables, solve both forward and inverse problems, and incorporate new physics within our framework. Through this workshop, we will demonstrate our approach to setting up and applying PINNs for solving problems in ice sheet modeling.
Facilitator: Mansa Krishna,
PhD Candidate, iHARP, Dartmouth College
April 14, 2026 (2-4 pm EST) | Foundations of Large Language Models Workshop Slides Recording
The development of Large language models (LLMs) is one of the most significant advances in Natural Language Processing. It has enabled the creation of systems that can understand and generate natural languages like human beings. Recent research demonstrates that well-trained large language models are capable of handling a large number of tasks and can be adapted to perform new tasks. Specifically, LLMs have transcended conventional tasks, which include pre-training and post-tuning, to specialized tasks that apply reasoning and further to reinforcement learning. Overall, this suggests a step towards more advanced forms of artificial intelligence and inspires further exploration into developing more powerful language models. This workshop aims to introduce researchers, faculty, and students from all backgrounds to fundamental concepts that allow them to leverage LLMs to enhance research and learning through programming/coding, data analysis, and modeling.
Facilitator: Dr. Josephine Namayanja,
Research Associate Professor & Executive Director, iHARP, UMBC
Co-Facilitator: Rhoda Nankabirwa,
PhD Student, iHARP, UMBC
Technical Support: Dr. Maloy K. Devnath and Sai Vikas Amaraneni
May 6, 2026 (12-1:30 pm EST) | Machine Learning Models for Causal Inference Analysis + HPC
Facilitator: Dr. Eric Stokan,
Associate Professor, Department of Political Science, UMBC
Facilitator: Roy Prouty,
Asst. Dir. Research Computing, DoIT, UMBC
Facilitator: Sai Vikas Amaraneni,
PhD Student, iHARP, UMBC
Last updated April 21, 2026