Upcoming Seminars

Speaker

Dr. Bahareh Tolooshams, Amii Fellow and Assistant Professor in the Electrical & Computer Engineering Department at the University of Alberta

Title

Learning as Inference: Structure, Representation, and Interpretability

Abstract
Modern AI systems are shaped by their inductive biases. These biases become critical when models are used for scientific understanding. This is where representation learning and interpretability come into play, with the goal of extracting meaningful and potentially causal structure from data. In this talk, I present a perspective in which inference serves as a general framework for learning, connecting ideas across generative modelling, computational neuroscience, representation learning, and inverse problems. I then focus on mechanistic interpretability: how concepts are extracted from large AI models, and how the inductive biases of our methods shape what we uncover. Through examples of hierarchical structure extraction, I highlight the need for principled, theoretically grounded interpretability methods that better reflect the structure of data.

Presenter Bio
Dr. Tolooshams is the PI of NeuBahar Lab (Neuro–Bayesian AI for Human-interpretable Abstractions and Representation Learning). Dr. Tolooshams is an Assistant Professor at the University of Alberta and an Alberta Machine Intelligence Institute (Amii). Dr. Tolooshams received her PhD in May 2023 from the School of Engineering and Applied Sciences at Harvard University, where she was also an affiliate to the Center for Brain Science. Before joining the University of Alberta, Dr. Tolooshams was a postdoctoral researcher and held the Swartz Foundation Fellowship Theoretical Neuroscience for two years at AI for Science Lab at the California Institute of Technology (Caltech). Dr. Tolooshams' research interests lie at the intersection of representation learning, interpretability, generative models, and computational neuroscience.

Website

Bahareh Tolooshams

Timing & Location

Amii HQ 2nd floor event space (10065 Jasper Ave)

pizza from 11:30, seminar from noon to 1

Add event to calendar

Add event

Speaker

Calarina Muslimani, PhD student at the University of Alberta, supervised by Dr. Matthew Taylor

Title

Reward Design and Evaluation in Reinforcement Learning

Abstract
This talk will focus on the design and evaluation of reward functions in reinforcement learning. I will begin by discussing the key challenges in reward design and the difficulties in determining whether a reward function is properly specified. Then, I’ll introduce an approach to support RL practitioners in designing more effective and aligned reward functions.

Presenter Bio
Calarina (Callie) Muslimani is a fourth-year PhD student at the University of Alberta in the Reinforcement Learning and Artificial Intelligence (RLAI) Lab, advised by Matthew E. Taylor. Her research focuses on designing human-aligned reward functions for reinforcement learning, including developing metrics to evaluate reward functions and creating reward learning algorithms. 

Timing & Location

UComm Seminar Room 2-108

pizza from 11:30, seminar from noon to 1

Add event to calendar

Add event

Speaker

Prabhat Nagarajan, PhD student at the UofA, supervised by Dr. Marlos Machado

Title

Accelerating Deep Q-learning with the Mean-expansion Layer

Abstract
Efficient action-value learning is central to reinforcement learning (RL), as action-values underpin many control algorithms such as Q-learning. However, action-value learning can be slow, requiring many updates to move values from their initialization, typically near zero, to their true values, which may be far from zero. Moreover, action-value learning algorithms typically update each action's value individually, without learning a shared value component across actions within a state. In this paper, we address these inefficiencies by introducing the mean-expansion transformation, which accelerates action-value learning by sharing values across actions within a state and by changing the problem from directly learning potentially large action-values to learning a lower-norm representation of them. In deep RL, this transformation can be applied as a parameter-free modification to Q-network architectures without altering the underlying algorithm. Empirically, we show that it improves DQN and IQN's performance in aggregate across 57 Atari games while increasing action gaps and dramatically reducing value overestimation.

Presenter Bio
Prabhat Nagarajan is a PhD student at the University of Alberta working with Marlos C. Machado. In the past he received undergraduate and master's degrees from the University of Texas at Austin in Computer Science and has completed internships at Sony AI and Microsoft Research. His research is at the intersection of deep learning and reinforcement learning (RL). Specifically his recent research has aimed to understand and develop algorithms in value-based deep RL.

Timing & Location

UComm Seminar Room 2-108

pizza from 11:30, seminar from noon to 1

Add event to calendar

Add event

Speaker

Dr. Amber Simpson, Amii Fellow and Professor, Faculty of Medicine & Dentistry - Radiology & Diagnostic Imaging Department

Title

TBA

Abstract
TBA

Presenter Bio
TBA

Website

TBA

Timing & Location

Amii HQ 2nd floor event space (10065 Jasper Ave)

pizza from 11:30, seminar from noon to 1

Add event to calendar

TBA

Speaker

Dr. Liam McCoy, Neurology Resident Physician, University of Alberta, and Research Affiliate, Massachusetts Institute of Technology, hosted by Dr. Randy Goebel

Title

TBA

Abstract
TBA

Presenter Bio
TBA

Timing & Location

UComm Seminar Room 2-108

pizza from 11:30, seminar from noon to 1

Add event to calendar

TBA

Speaker

Dashti Ali, PhD student at Queens University, supervised by Dr. Amber Simpson

Title

Topological Data Analysis for Medical Imaging

Abstract
Reliable and robust feature extraction is a fundamental step in machine learning pipelines for deriving informative representations from medical images. Widely used feature extraction approaches in medical imaging domain are radiomics and convolutional neural networks (CNNs). Radiomic features rely on comparisons at the pixel level and can vary under noise and other clinical parameters. CNNs are computationally intensive and may learn spurious patterns that are not relevant to the target task, such as background noise or texture. Topological data analysis (TDA), a recent advancement based on the mathematical field of algebraic topology, addresses some of these limitations by capturing the structure of data across multiple scales through topological and geometric summaries. TDA has demonstrated promising results across various medical imaging applications. This presentation provides an overview of TDA and its potential in medical image analysis, discusses its integration with existing machine learning techniques, and highlights selected recent research projects.

Presenter Bio
Dashti Ali is currently a PhD candidate in the school of computing at Queens University under the supervision of Dr. Amber Simpson. He achieved his MSc in Scientific Computation from the university of Nottingham, UK. His research focus is on the field of machine learning and topological data analysis on medical imaging domain and particularly cancer image analysis.

Website

Dashti Ali

Timing & Location

UComm Seminar Room 2-108

pizza from 11:30, seminar from noon to 1

Add event to calendar

TBA

NO SEMINAR - Upper Bound

Speaker

Jiamin He, PhD student at the University of Alberta, supervised by Dr. Martha White

Title

TBA

Abstract
TBA

Presenter Bio
TBA

Timing & Location

UComm Seminar Room 2-108

pizza from 11:30, seminar from noon to 1

Add event to calendar

TBA