Jan 7, 2025

Find it here

December 6th, 2024 – 11am-12pm, EC4-2101A

Deep learning has often been criticized as a black box, whose predictions are superb and yet opaque. Lots of efforts have been made to explain/interpret the predictions of machine learning. In this talk I will first give a selected (and biased) overview of some existing approaches to feature and data attribution. Then, I will discuss faster approximation algorithms for computing the probabilistic value (of which the celebrated Shapley value is a special case). Lastly, I will present some limitations of current approaches.

November 29th, 2024 – 11am-12pm, EC4-2101A

Metal additive manufacturing processes have advanced form prototyping techniques, to full production-compatible systems. The first half of the presentation will introduce a brief history of additive manufacturing and a dive into metal additive manufacturing techniques with demonstrators. The second half of the presentation will introduce the Data Workflows and untapped opportunities, along with brief case studies on how machine vision can be used to optimize processes.

**11:00 am Presenter: Chang Liu**

Semantic segmentation tasks require expensive and time-consuming pixel-level annotations. Unsupervised domain adaptation (UDA) aims to transfer knowledge from a label-rich source domain to a target domain with no labels. Recently, vision-language models (VLMs) have shown promise for domain-adaptive classification, but remain under-explored for domain-adaptive semantic segmentation (DASS). Existing language-guided DASS methods align pixel-level features with generic class-wise prompts, which require target-domain knowledge, and do not leverage the intricate spatial relationships and object context endowed by language priors. In this work, we propose LangDA, the first domain-agnostic approach to explicitly induce context-awareness in language-driven DASS. In LangDA, we align image features with VLM-generated context-aware scene descriptions via a consistency objective. LangDA achieves state-of-the-art results on three adaptation benchmarks, outperforming existing methods by 3.9%, 2.6%, and 1.4%.

November 14th, 2024, 11am-12pm, EC4-2101A

Sequential decision making in the real-world involves reasoning about and responding to multiple interacting agents in a dynamic environment. Multi-agent reinforcement learning (MARL) is an emerging field of machine learning that aims to learn policies for such environments and has seen much success in the past decade. However, MARL is yet to find wide application in large-scale real-world problems due to two important reasons. First, MARL algorithms have poor sample efficiency, where many data samples need to be obtained to learn meaningful policies, even in small environments. Second, MARL algorithms are not scalable to environments with many agents since, typically, these algorithms are exponential in the number of agents in the environment. In this talk, I will describe critical aspects of our research that addresses both MARL challenges. Towards improving sample efficiency, an important observation is that many real-world environments already deploy sub-optimal or heuristic approaches for generating policies. To this end, we propose a principled framework for accelerating MARL training using such pre-existing solutions and show its effectiveness both theoretically and empirically. Towards scaling MARL, we explore the use of mean field theory, which abstracts other agents in the environment by a single virtual agent. Subsequently, we combine our work in mean field learning and learning from pre-existing knowledge to show that we can achieve powerful MARL algorithms that are more suitable for large real-world environments as compared to prior approaches. In this talk, I will describe real-world applications of our work in domains spanning autonomous driving, robotics, fighting wildland fires and ride-pool matching problems in addition to classic video games.

November 8th, 2024 -11am-12pm, EC4-2101A

A quick overview on how to approach the visual presentation (plots, graphs, tables etc) of concepts and data.

***11:00 am Presenter: Yan Song (Kevin) Hu***

As robotics and AI systems become more sophisticated, the demand for real-time, high-quality spatial data has increased in importance. A promising approach for generating such data comes from emerging radiance field techniques, particularly 3D Gaussian Splatting, that can quickly generate dense maps of environments. My research focuses on generating dense volumetric maps in real-time by combining 3D Gaussian Splatting with Direct Sparse Odometry (DSO), a state-of-the-art robotic navigation system. I have found that certain characteristics of DSO, especially its pixel-based tracking method, enable 3D Gaussian Splatting to produce maps more quickly and with greater quality.

***11:30 am Presenter: Yizie Liu***

We introduce a hybrid approach to solve two popular industrial tasks: Image-based Scene Change Detection (SCD) and Pose-agnostic Object Anomaly Detection (PAD). Our method maintains both a learning-based model Gaussian Splatting for Novel View Synthesis and a Structure-from-Motion model (Hierarchical Localization) for localization, which takes the advantage of the fast training and inference of 3D Gaussian Splatting and the fast localization of Hierarchical Localization. We also explored the possibility of SAM2 Based Mask Refinement on the SCD task, where the change is object-level.

Oct 21, 2024

Find it here

October 11th, 2024,11am-12pm, EC4-2101A

This talk will provide an overview of the current state of autonomous vehicles, exploring both active deployments and emerging research directions. I will briefly examine the designs powering existing and next-generation autonomous systems, including classical architectures, end-to-end neural networks, and dual-processing systems. Drawing on the growing body of experience from real-world deployments, I will highlight key challenges such as new categories of driving errors, safety assurance hurdles, and the high development costs of systems intended for deployment on public road. These insights will lead into a discussion of future research needs in the field. Additionally, I will offer a brief overview of ongoing research at the WISE Lab.

Krzysztof Czarnecki is a Professor of Electrical and Computer Engineering and a University Research Chair at the University of Waterloo, where he leads the Waterloo Intelligent Systems Engineering (WISE) Laboratory. His research focuses on ensuring the safety of AI systems and driving behavior. In 2018, he co-led the development of the first autonomous vehicle tested on public roads in Canada. He has made significant contributions to automotive AI and software safety standards, including SAE J3164 and ISO 8800. Before joining the University of Waterloo, he worked at DaimlerChrysler Research in Germany (1995-2002), where he advanced software development practices and technologies for enterprise, automotive, and aerospace sectors. His work has been recognized with numerous awards, including the Premier’s Research Excellence Award (2004) and the British Computing Society’s Upper Canada Award for Outstanding Contributions to the IT Industry (2008). He has also received twelve Best Paper Awards, two ACM Distinguished Paper Awards, and five Most Influential Paper Awards.

Sept 27th, 2024, 11am-12pm, EC4-2101A

Summary: Microfluidics exploits fluids and their physical and chemical properties at the microscale, enabling miniaturized platforms that offer lower cost, faster pace, higher performance, and increased portability than their macroscale counterparts. This talk will briefly discuss three research themes including droplet microfluidics, microwave sensing and soft robotic wearable systems. 

·      Two-phase droplet microfluidics employs monodispersed water-oil emulsions as mobilized test tubes to perform high throughput analysis (HTA). Despite numerous novel technologies reported, the adoption of droplet microfluidics as an HTA tool by non-microfluidics experts has not been seen. Modular-based droplet microfluidics, enabling easy assembly of application-specific systems, presents tremendous potential to break this barrier. This talk will introduce our work towards this goal including, a suite of physical models that can serve as design tools for passive-based droplet modules such as droplet generators, mergers, sorters and heaters, and a unique active droplet microfluidics method that relies on visual feedback of droplet position to actuate a pressure source to actively control individual droplets realizing functional modules. 

·      Simultaneous sensing and heating of individual droplets are critical but very challenging. Microwave resonators present tremendous potential to meet this need which will be the second part of this talk. Microwave sensing finds various applications beyond droplet microfluidics. Its applications for the detection of virus such as SARS-CoV-2 and E. coli as well as metal ions will be discussed.

·      Soft robotic wearable systems offer hope to improve the quality of life for those in need because of their compliance nature. Most existing systems are expensive, power intensive and tethered to external power sources, limiting user mobility. Microfluidics enables miniaturization of the system including its front end (e.g. wearable sleeves) and back end (control unit), translating to low cost, tetherless and energy-efficient operation. This talk will present wearable sleeves for treating lymphedema, arthritis, and pressure ulcers due to the ill fit of prosthetic sockets.