Events

Name: Chenwei Zhang

Date: December 15, 2025

Time: 9:30 AM

Location: ICCS X836

Co-Supervisors: Anne Condon and Khanh Dao Duc

Title of Thesis: Applications of Deep Generative Models in DNA Reaction Kinetics and Cryogenic Electron Microscopy

Abstract:
This dissertation explores how deep generative models can advance the analysis of challenging biological problems by integrating domain knowledge with cutting-edge deep learning techniques. It focuses on two fundamental areas: DNA reaction kinetics and cryogenic electron microscopy (cryo-EM).

In the first part, we present ViDa, a biophysics-informed deep learning framework that leverages variational autoencoders (VAEs) and geometric scattering transform to generate biophysically-plausible embeddings of DNA reaction kinetics simulations. These embeddings are further reduced to a two-dimensional Euclidean space to visualize DNA hybridization and toehold-mediated three-way strand displacement reactions. By embedding simulated secondary structures and reaction trajectories into a low-dimensional representation, ViDa preserves structure and clusters trajectory ensembles into reaction pathways, making simulation results more interpretable and revealing new insights into reaction mechanisms.

In the second part, we address key challenges in cryo-EM density map interpretation and protein structure modeling. We first provide a comprehensive review and benchmarking of state-of-the-art deep learning methods for protein structure modeling (i.e., atomic model building). We propose improved evaluation metrics to assess the performance of these methods and provide guidance for researchers. We then present Struc2mapGAN, a generative adversarial network (GAN) that synthesizes high-fidelity experimental-like cryo-EM density maps from protein structures. We finally present CryoSAMU, a structure-aware multimodal U-Net that enhances intermediate-resolution cryo-EM density maps by integrating density features with structural embeddings from protein large language models through cross-attention mechanisms.

Overall, these contributions demonstrate the potential of deep generative models to interpret DNA reaction mechanisms and to advance cryo-EM density map analysis and protein structure modeling.

Speaker:  Dr. Guy E. Blelloch, U.A. and Hellen Whitaker University Professor, CMU

Title:  Parallel Algorithms : A Retrospective and Current Directions

Abstract:

Almost no computational devices with a single core are still being      
produced, and the vast majority of computational cycles that are used    
today are based on algorithms that make use of dozens to thousands of    
cores.  Yet when I ask my CS colleagues to list a few algorithms,        
parallel (multi-core) algorithms are rarely in the mix.  The community  
often views parallel algorithms as complicated, esoteric, and a topic    
for specialists.                                                        
                                                                         
The state of the art in parallel algorithms, however, has improved      
dramatically over the past decades, both in theory and practice.        
Today efficient parallel algorithms can be as simple as their            
sequential counterparts, and much faster on modern machines.  On the    
other hand, developing efficient parallel algorithms for some specific         
problems remains notoriously difficult.                                  
                                                                         
In this talk I will outline the current state of parallel algorithms,    
describe some techniques that have been developed over recent years,    
some problems that remain hard, and describe current and potential      
future directions in parallel algorithms.

Bio:

Guy Blelloch is a U.A. and Hellen Whitaker University Professor of  
Computer Science at Carnegie Mellon University. He received his     
PhD in Computer Science from MIT in 1988.  His research contributions    
have been in the interaction of practical and theoretical                
considerations in parallel algorithms and programming languages.        
Blelloch has received the ACM Paris Kanellakis Theory and Practice Award
(2024) for his work parallel graph algorithms, the inaugural SPAA        
Parallel Computing Award for his contributions in parallel algorithms,  
the IEEE Charles Babbage Award for his contributions to parallel        
programming and algorithms, and is an ACM Fellow.                                                    

Host:  Yuanhao Wei, UBC Computer Science

Name: Rohit Murali

Date: November 13^th, 2025

Time: 12:30 PM

Location: Room 200, Graduate Student Centre (6371 Crescent Road)

Supervisor: David Poole

Title of Thesis: Predictive models for personalized support using interaction data in open learning environments

Abstract: Personalization during student learning has been shown to improve learners' experience by responding to learners in real-time. Various learner signals can be leveraged for personalization during learning such as student's affective states or students' predicted future learning performance. The work in this PhD focuses on predicting learner signals that can drive personalization using real-time sequential interaction data in various open-ended learning environments (OELEs).

First, we look at predicting two co-occurring emotion pairs (student experiences Boredom and/or Frustration; student experiences Curiosity and/or Anxiety) using eye-tracking and interface interaction data in an Intelligent Tutoring System (ITS). Our work combines two datasets with different eye-trackers. We test the effect that combining datasets has on predictions using interaction and eye-tracking data, in isolation and in combination with each

other. We conduct a statistical analysis of classifier performance that provides insights into the effectiveness of the different data modalities in terms of predicting emotion pairs. Our analysis shows that combining data from different eye-trackers is feasible, and can be utilised for real-world affect detection where data from different eye-trackers. Moreover, for each emotion pair, we were able to isolate classifiers that outperform a baseline in terms of overall accuracy.

Second, we look at predicting student learning performance using interface interaction data in a game-based OELE. Our work builds a data-informed intelligent pedagogical agent (IPA) that predicts students' future learning performance during self-directed interaction. Our work tests the effectiveness of these predictions to trigger help interventions for struggling students. We built a student model by extracting meaningful student behaviors on real-world interface interaction data obtained during interaction in online classrooms and including expert insights. Our results show that our student model performs better than a baseline and has the potential for adaptive support in self-directed interaction with the OELE. To the best of our knowledge, we are the first to build and evaluate an IPA for in-the-wild interaction with OELEs. In this regard, our study provided empirical evidence for research in OELEs that adaptive scaffolding can improve student learning performance. 

Last, we systematically evaluated various predictive classifiers across multiple open-learning datasets. Our analysis provides insights into the strengths and weaknesses of different classifiers along with a discussion about the tradeoff between inherent interpretability and accuracy. These findings aim to contribute to the development of more effective adaptive support systems and scaffold research on building personalized systems in learning.