*** 11:30 am Presenter:  Bavesh Balaji ***

Description: 2D pose estimation is crucial for hockey analytics, enabling tasks like action recognition and player assessment. However, motion blur, occlusions, and crowded scenes (multiple players) hinder accurate pose estimation for both players and their held objects (e.g., hockey sticks). Existing methods, limited to predicting keypoints within an image, struggle in these scenarios. This work addresses this gap by modeling the relationship between human joints and their extensions (like hockey sticks), enabling the estimation of out-of-image keypoints.

*** 12:00pm Presenter:  Harish Prakash ***

Tracking players in sports is the fundamental first step toward several strategic, analytical, and tactical insights. But ice hockey differs from most popular sports due to its fast-paced, chaotic player dynamics that are highly intense and unpredictable. It presents several inherent challenges to monocular tracking, including motion blurs, heavy player occlusions, and significant camera motion. In this seminar, I will discuss my work on overcoming these challenges to track players across a given broadcast sequence with high fidelity. Specifically, my discussion will revolve around building a complete pipeline for localization, detection, and association of players, and the challenges faced along the way. Additionally, I will also share several ‘key’ insights obtained by curating a dataset for the same, the bias in current evaluation metrics, and potential future works along this direction.

April 12, 2023, 11:30 am, EC4-2101A

Talk title:
Towards Trustworthy Human-Machine Collaboration

Abstract:
Advancements in technology such as the keyboard, mouse, touch-screens, voice-based communication, and today: data-enable interfaces (e.g., ChatGPT) have enabled more natural forms of interaction between humans and machines.

Despite seemingly magical experiences, many questions remain open:
* How does one recover from, or overturn, poor experiences via decisions made by AI?
* How does one assay the safety, factuality, and ethics of AI systems to foster trust in AI?
* How does one design systems that make use of the best of human and machine abilities?

I argue that the next step, and the solution to all these questions, is to continue the line of development above, and facilitate even more interaction, discussion, and communication between humans and intelligent agents with the ultimate goal of “intelligence augmentation” via “trustworthy human-machine collaboration.”

In this talk, I will describe efforts made to address these questions, as well as plans for future research.

Current bio:
Dr. Amir-Hossein Karimi is an Assistant Professor in the Electrical & Computer Engineering department at the University of Waterloo where he leads the Collaborative Human-AI Reasoning Machines (CHARM) Lab. The lab’s mission is to advance the state of the art in artificial intelligence and chart the path for trustworthy human-AI symbiosis. In particular, the group is interested in the development of systems that can recover from or amend poor experiences caused by AI decisions, assay the safety, factuality, and ethics of AI systems to foster trust in AI, and effectively combine human and machine abilities in various domains such as healthcare and education. As such, the lab’s research explores the intriguing intersection of causal inference, explainable AI, and program synthesis, among others.

Amir-Hossein’s research contributions have been showcased at esteemed AI and ML-related platforms like NeurIPS, ICML, AAAI, AISTATS, ACM-FAccT, and ACM-AIES, via spotlight and oral presentations, as well as through a book chapter and a highly regarded survey paper in the ACM Computing Surveys. Before joining the University of Waterloo, Amir-Hossein gained extensive industry experience at Meta, Google Brain, and DeepMind and offered AI consulting services worth over $250,000 to numerous startups and incubators. His academic and non-academic endeavours have been honoured with awards like the Spirit of Engineering Science Award (UofToronto, 2015), the Alumni Gold Medal Award (UWaterloo, 2018), the NSERC Canada Graduate Scholarship (2018), the Google PhD Fellowship (2021), and the ETH Zurich Medal (2024).

*** 11:30am Presenter:  Peter Lee ***

Title: Automated System for a Collaborative Robot Arm to Collect Nasopharyngeal Swab Samples

In the face of aging demographics and airborne spread illnesses, the development of robotics for close-contact healthcare tasks is of paramount important looking towards the future. In this talk, I will discuss my research towards using a standard collaborative robot arm to safely execute the nasopharyngeal swab test, a common diagnostic for detecting various types of respiratory illnesses. There are three major components to this research. The first involves the derivation of an optimization problem to find the best path to insert the swab along through the nasal cavity. Second, a visual servo system is designed that actuates a robotic arm to place the swab next to the nostril with the correct position and orientation. Third a compliant control system is formulated and implemented to insert the swab and collect samples from the nasopharynx, while regulating measured forces applied to the swab to ensure the robot is stable and responsive.  I will also discuss experiments featuring both human trials and trials with a realistic nasal cavity phantom that validate the proposed methods.

*** 12:00pm Presenter:  Neil Brubacher ***

This past Monday marked the 25th anniversary of Nunavut, Canada’s youngest territory. While Nunavut’s legal boundaries are relatively young, Inuit – peoples Indigenous to northern Canada – have lived in this challenging Arctic landscape for thousands of years. Today, Inuit are continuing to adapt to significant cultural and environmental changes. In January, I got a chance to visit Sanikiluaq, Nunavut, a small community in Hudson Bay with a predominantly Inuit population of about 900. In this “show and tell” I’ll share some of my experiences and the surrounding context from this trip, the people, the landscape, and a glimpse of how digital technology is being used to complement long-established ways of going about things. Bundle up!

March 22, 2023, 11:30 am, EC4-2101A

Janet Bannister has extensive connections in the tech industry, starting from working at Ebay to then founding Kijiji, as well as stints at P&G (operations) and McKinsey (consulting).  Since then, she has been on boards of many early-stage companies, worked with top tech founders in Canada, and is now the Founder & Managing Partner of Staircase Ventures, a firm that provides supports for founders to enable and enhance their efforts. Janet is now co-chair of C100, on the LEAP board, mentors with C100, and on the Ivey Business School Advisory Board.

*** 11:30am Presenter:  Navid Shahsavari ***

Title: Enhancing Thermocular’s Performance for Clinical Applications

In my upcoming seminar, I will outline the primary goals and recent progress of the “ThermOcular” project. This innovative effort is focused on a dual visual-thermal camera system designed to non-invasively measure and track temperatures across the ocular surface through thermal imaging. My work aims to boost the ThermOcular system’s performance by improving its ability to segment ocular components accurately, enhancing the precision and efficiency of image registration processes, and creating a software system designed for clinicians. This software enables the examination of patients and analysis of ocular surface temperatures, ensuring that the ThermOcular device can be effectively integrated into real-world clinical settings.

*** 12:00pm Presenter:  Rishav Bhardwaj ***

Title: New method to improve the diagnostic utility of OCTA images in retinal disease

There has been considerable progress in implicit neural representation to upscale an image to any arbitrary resolution. However, existing methods are based on defining a function to predict the Red, Green and Blue (RGB) value from just four specific loci. Relying on just four loci is insufficient as it leads to losing fine details from the neighboring region(s). We show that by taking into account the semi-local region leads to an improvement in performance. In this paper, we propose applying a new technique called Overlapping Windows on Semi-Local Region (OW-SLR) to an image to obtain any arbitrary resolution by taking the coordinates of the semi-local region around a point in the latent space. This extracted detail is used to predict the RGB value of a point. We illustrate the technique by applying the algorithm to the Optical Coherence Tomography-Angiography (OCT-A) images and show that it can upscale them to  random resolution. This technique outperforms the existing state-of-the-art methods when applied to the OCT500 dataset. OW-SLR provides better results for classifying healthy and diseased retinal images such as diabetic retinopathy and normals from the given set of OCT-A images.

*** 11:30am Presenter:  Muhammed Patel ***

Title: Detecting whales from aerial imagery

Remote sensing technologies have significantly advanced ecological monitoring, offering unprecedented capabilities for species detection and conservation efforts. However, the manual annotation of extensive aerial surveys remains a daunting task. In collaboration with the Department of Fisheries and Oceans (DFO), we introduce a novel semi-automated deep learning-based whale detection pipeline. This innovative approach aims to enhance the efficiency and scalability of annotating large-scale surveys by addressing challenges such as weak whale signatures, the presence of lookalikes, and heterogeneity. The seminar will focus on the intricacies of training a small object detection model tailored to these challenges, exploring solutions, and discussing aspects of online learning and domain adaptation for ever-changing environmental conditions.

*** 12:00pm Presenter:  Javier Noa Turnes ***

Title: Long-range spatial dependencies for sea ice semantic segmentation.

Studying sea ice presence in the Arctic is pivotal in the sustainable development of society in northern communities. For ice agencies, synthetic aperture radar (SAR) images are the most prevalent data source to generate ice charts that offer a coarse description of ice concentration and prevailing ice types. While such a product aids analysts, it lacks relevant pixel-level information to locate and describe the ice formation. Therefore, there is a significant effort to obtain automatic methods to generate detailed sea ice maps. The nature of SAR images produces significant class variability that leads to spatial non-stationary statistics, posing challenges for sea ice mapping when using current deep learning architectures that ignore large-scale context. Towards enhancing automated sea ice classification, I am presenting a methodology that makes the most of local and global information and gives treatment to boundaries on semantic segmentation of ice and water. Three blocks are utilized, a convolutional neural network (CNN), unsupervised segmentation, and self-attention from transformer networks. Finally, there are two prediction heads: i) pixel-based, on the CNN output, and ii) region-based, on the transformer output. In this manner, the model learns to capture large spatial feature interactions to tackle non-stationary statistics. The main function of region-based head is to promote region-consistent feature maps for pixel-based prediction.

*** 11:30am Presenter:  Fernando Cantu ***

Title: Incidence Angle Dependence of Texture Features from Dual Polarization RADARSAT-2

Summary: This presentation discusses the impact of synthetic aperture radar (SAR) incidence angle on  gray-level co-occurrence matrix (GLCM) texture features for sea ice classification. The session will cover the analysis of GLCM features’ dependency on incidence angles and their significance in distinguishing sea ice types,  while also studying the sensitivity to GLCM hyper-parameters. 

*** 12:00pm Presenter:  Andrew Hryniowski ***

Title: Representational Response Analysis (RRA)

Summary: Designing a CNN is not a straightforward process. Model architecture design, learning strategies, and data selection and processing must all be precisely tuned for a researcher to produce even a non-random preforming model. When building a new model, researchers will rely on quantitative metrics to guide the development process. Typically, these metrics revolve around model performance characteristics constraints (e.g., accuracy, recall, precision, robustness) and computational (e.g., number of parameters, number of FLOPs), while the learned internal data processing behaviour of a CNN is largely ignored. In this work we propose a novel analytic framework that offers a range of complementary metrics that can be used by a researcher to study the internal behaviour of a CNN. We call the proposed framework Representational Response Analysis (RRA). The RRA framework is built around a common computational kNN based model of the latent embeddings of a dataset at each layer in a CNN.  Using RRA we study the impact of specific CNN design choices. Specifically, we use RRA to investigate the consequences on a CNN’s latent representation when training with and without data augmentations, and to understand the latent embedding symmetries across different pooled spatial resolutions.  Using the insights from the pooled spatial resolution experiments we propose a novel CNN attention-based building block that is designed to take advantage of key latent properties of a ResNet.

Feb 16, 2023, 11:30 am, EC4-2101A

Zown is a pioneering real estate technology company aimed at revolutionizing the home buying and selling experience. Our platform employs advanced technology to streamline and demystify the real estate transaction process, making it more efficient, transparent, and affordable for individuals looking to buy or sell homes. By automating traditional real estate agent tasks, Zown significantly reduces commission fees for sellers, offering substantial savings that make selling homes more cost-effective. For buyers, Zown introduces an innovative down payment rewards program, enabling them to earn substantial contributions towards their down payments through various activities, thus making homeownership more accessible and less financially burdensome. At the heart of Zown’s approach is a powerful matching algorithm designed to connect sellers with the ideal buyers quickly and effectively, ensuring that each transaction is not just faster but also perfectly aligned with the needs and preferences of both parties. Zown’s mission is to democratize the real estate market, providing a fairer, more accessible platform that benefits a broader audience and transforms the way people engage with real estate.

Feb9 2023, 11:30 am, EC4-2101A

*** 11:30am Presenter:  Frank Mokadem ***

Summary: Please join me Friday as I present my research on reduce the memory footprint of CNNs through tensor representation and factorization. I combine approximation theory from linear algebra and grid search to find lighter representations of networks. This method, already proven to reduce memory footprint with at least 1 order of magnitude, can be improved to minimize loss in predictive power of the network. My hope is to reduce big models into deployable lighter models able to function on edge devices, furthermore, I hope to automate this process and expand it to more CNN architectures. 

Feb 2, 2023, 11:30 am, EC4-2101A

James Lowman is a co-founder of Cauchy Analytics.  From LinkedIn “James Lowman, a tenacious entrepreneur and problem solver, holds a BMath in Applied Mathematics and a MASc in Chemical Engineering from the University of Waterloo. As CEO of Cauchy Analytics, James combines his expertise in electrical engineering, applied math, and strong interpersonal skills to develop innovative, non-invasive cardiac monitoring solutions. Adept at fostering collaborative, trust-based working environments, James has led Cauchy to secure a spot at the prestigious Velocity incubator and remains dedicated to building a bright future for the company with relentless enthusiasm and commitment.”