Bridging Data and Example-Aided Ideation
Speaker: Hannah Bako (School of Data Science)
Abstract: Automation is reshaping creative work, but when applied without care, it can overlook how designers truly ideate, particularly their reliance on examples as sources of inspiration. My research reveals how designers use existing data visualization examples to generate and refine ideas, challenging assumptions in current tools and informing new methods for retrieval, exploration, and implementation. This talk shares these insights and outlines a future where human-AI collaboration supports more contextual, creative, and designer-driven data visualization.
Creating Critical Material Networks for Studying Supply Chain Disruptions
Speaker: Mandy Wilson
Abstract: Supply chain analysis for critical materials often employs top-down methods, tracing backward from finished products. These methods can lack visibility into intermediate dependencies due to proprietary manufacturing and opaque supplier networks. This presentation presents a bottom-up agentic AI approach for constructing a Material Evolution Knowledge Hypergraph (MEKH), an essential building block for supply chain networks that document the transformation of raw materials into advanced technological products.
The approach uses the Material Processing Dependency Network Framework (MP-DNF), a multi-agent system in which specialized language models iteratively discover, classify, and validate materials and industrial processes. Multiple knowledge sources are integrated via the Model Context Protocol (MCP), including harmonized trade code databases, semantic reference caches, web searches, and structured encyclopedic sources. The resulting networks are represented as directed hypergraphs that capture the multi-input, multi-output nature of industrial processes.
Promote A Safe Use of Vision-Language Models for Healthcare
Speaker: Miaomiao Zhang (ECE/CS, SEAS)
Abstract: Vision-Language Models (VLMs) have rapidly gained attention for medical applications including image and video synthesis, classification, segmentation, and object recognition. Despite their promise, current VLMs largely focus on image intensity and texture manipulation and often fail to preserve the underlying topology and geometry of human anatomical structures. This limitation poses significant risks in healthcare settings, where anatomical accuracy is critical for diagnosis, treatment planning, and clinical decision-making. In this talk, I will discuss our recent research aimed at addressing these challenges and establishing frameworks that ensure VLMs operate safely, reliably, and anatomically faithfully in clinical imaging applications without compromising patient care.