Please join us for the ISP Forum where we will be featuring Zhengbo Zhou and Arun Narayanan.   Come enjoy lunch and refreshments starting at noon, followed by presentations from 12:30–1:30 PM. The event will be held in person in SENSQ 5317.

Zhengbo Zhou’s Presentation Information

Presentation Title: Modeling Spatiotemporal Asymmetries in Longitudinal Mammography for Breast Cancer Risk Prediction

Abstract: Breast cancer risk prediction from longitudinal mammograms requires modeling subtle changes not only over time but also between the left and right breasts. In this talk, I will present STA-Risk, a deep learning framework that explicitly models spatiotemporal asymmetries in longitudinal mammography for future breast cancer risk prediction. The framework incorporates side and temporal encodings together with an asymmetry-aware learning objective to capture bilateral differences and longitudinal changes across screening examinations. I will also discuss challenges in cross-cohort domain shift and how joint training across datasets can improve robustness and generalization. These findings highlight the potential of explicitly modeling spatial and temporal asymmetries for personalized breast cancer risk assessment from routinely acquired screening mammograms.

Bio:  Zhengbo Zhou is a PhD in the Intelligent Systems Program at the University of Pittsburgh, advised by Prof. Shandong Wu. His research focuses on artificial intelligence for medical imaging, particularly longitudinal representation learning, breast cancer risk prediction, and multimodal learning.

Arun Narayanan’s Presentation Information

Title: Addressing a Bias in Evaluating of Student Self-Explanations of Worked Programming Examples

Abstract: Worked examples are step-by-step solutions to problems in a specific domain, offered to students to acquire domain-specific problem-solving skills. The power of worked examples could be magnified by combining them with self-explanations, which ask students to explain rather than passively study each problem-solving step. The main challenge of this approach is assessing the correctness of the student's explanations. In the current approach, student explanations are judged by their semantic similarity to an explanation provided by an instructor or domain expert. However, recent studies of example explanations in the domain of programming demonstrated that many students express themselves very differently from domain experts. In this situation, a traditional semantic similarity approach might introduce bias against students who correctly explain worked examples but are considerably different from expert explanations. In this paper, we use a recently published dataset to compare several explanation-assessment approaches based on semantic similarity with alternative approaches based on direct Large Language Model prompting. Our results show that the use of Large Language Models enables worked example systems that follow an active learning approach to reduce bias in evaluating example explanations.

Bio: Arun is a PhD Student in the Intelligent Systems Program at the University of Pittsburgh. His current research is on developing personalized educational technologies for students in undergraduate computer science. He works with using AI and LLMs to build Intelligent User interfaces for students to receive feedback on their interactions and in introductory programming interfaces. His research is in the topic of applied AI in education, particularly in the topics of data analysis, data mining, machine learning and natural language processing. He also works with developing intelligent textbooks.