Hassan Talaeian
Title: Improving power efficiency of wireline clock generation using open loop fractional dividers
Abstract: Improving power efficiency of wireline clock generation using open loop fractional dividers.
Bio: Hassan Talaeian studied his BCs in electrical engineering at Sharif University of Technology in Iran. He worked mainly on hardware design and validation before joining UBC for his Master’s in which right now his studies are focused on clock generation and synthesis.
Yuxin Jiang
Title: sparse Gaussian gradient code
Abstract: The talk is titled as “sparse Gaussian gradient code”. It will start with a brief introduction of gradient coding and how it’s applied in a machine learning task. Then the talk will propose a new grading code which is called sparse gaussian gradient code.
Bio: Yuxin Jiang is a 1st year PhD student at UBC ECE. She just obtained B.Sc. of Statistics from UBC in May 2023. She is very passionate about research and especially interested in coding theory. She loves bouldering and hiking.
Nikhil Pratap Ghanathe
Title: T-RECX: Tiny-Resource Efficient Convolutional neural networks with early-eXit
Abstract: T-RECX is an early-exit network architecture optimized for tinyML models running on ultra-low-power devices. T-RECX achieves an average of 30% reduction in execution time by skipping several layers of computations for easy-to-classify inputs.
Bio: Nikhil P Ghanathe is a 5th year PhD student, working with Prof Steve Wilton. He completed his Masters from University of Florida in 2016 and went on to work at CERN and Microsoft Research before his PhD. His current interests involve optimizing/deploying/monitoring tiny ML algorithms on ultra-low-power devices.
Mani Sadati
Title: A Precise and Efficient Fault Injection Framework for Deep Learning Accelerators
Abstract: Deep Neural Networks (DNNs), pivotal in autonomous vehicles and medical diagnosis, are increasingly dependent on specialized hardware accelerators. These accelerators, however, are susceptible to permanent hardware faults, notably stuck-at faults, leading to silent data corruptions (SDCs) which silently alter machine learning (ML) model predictions. Traditional Fault Injection (FI) frameworks are divided into less accurate software-based and precise but cumbersome hardware-based approaches. Our research introduces an innovative FI framework utilizing LLVM’s Intermediate Representation (IR) to merge the speed and user-friendliness of software FI with the precision of hardware FI. This novel approach involves translating DNNs to IR level, followed by a one-off hardware FI experiment to map IR instructions to hardware components. This enables accurate hardware fault injections at the IR level across various ML models and accelerator designs. This groundbreaking work aims to enhance the resilience analysis of ML models and accelerator hardware, marking a significant advancement in deep learning accelerator reliability analysis.
Bio: Mani Sadati is a first year master’s student at the Electrical and Computer Engineering Department of the University of British Columbia, where he is actively engaged in research at the Dependable Systems Lab. His research interests span a broad spectrum, including deep learning, reliability, FPGA, and CAD tools. Mani completed his bachelor’s degree in computer engineering just last year, and during his undergraduate studies, he also served as a research assistant for two years, gaining valuable experience in his field. He also won the CAD-contest at ICCAD 2020 the top computer aided design conference for GPU acceleration of logic simulation.
Joel Nider
Title: The end of the CPU era
Abstract: Soon, CPUs will no longer be a central component in computer systems, since most processing will be handled by other more efficient hardware such as GPUs, TPUs and FPGAs. Our operating systems are designed for CPU-centric systems and rely on it for coordination. But we must ask: is this really the best way to build the systems of the future?
Bio: Joel works on operating systems and related technologies. He has worked on projects in processing-in-memory, virtualization, file systems and drivers. Before coming to UBC, he works at IBM Research in Haifa, Israel. At UBC, he is contemplating what is required to build an operating system from autonomous devices that do not require a CPU.
Victor Sira
Title: Multi-Phase Ring Oscillators and Their Applications.
Abstract: Multi-Phase Ring Oscillators and Their Applications. Reliable and low phase noise ring oscillators are critical for reducing valuable area in many high-speed communication ICs. As CMOS technologies scale standard ring-oscillator frequencies do not improve. Multi-phase ring oscillators are a way to improve phase noise and frequency without the standard number of stages vs frequency trade-off.
Bio: Victor Sira is a MASc student in the SoC lab under the supervision of Dr. Sudip Shekhar. He is interested in mixed/signal, analog and RF IC design.
Ruixuan(Matt) Xie
Title: Multi-Rate DSP for Sub-mm Astronomy: Polyphase Synthesis Filter Bank on FPGA for Enhanced KID Readout
Abstract: The Prime-Cam instrument on the Fred Young Submillimeter Telescope employs a novel polyphase synthesis filter bank for AC biasing of detectors, enhancing bandwidth and dynamic range through real-time, dynamic signal generation. This advanced digital signal processing design shows promising preliminary results and potential for future submillimeter wave astronomy KID readout systems.
Bio: Mr. Xie is a master’s student in the Department of Electrical and Computer Engineering at the University of British Columbia. His academic journey has been marked by a consistent focus on digital design, culminating in his current involvement with the CCAT Collaboration, contributing to the development of the Prime-cam instrument’s 850GHz module for the Fred Young Sub-Millimeter. Having obtained his Bachelor of Applied Science in the same field from the University of British Columbia, his expertise extends to circuit theory, VLSI design, digital design, and machine learning. Mr. Xie is currently seeking employment opportunities after the completion of his M.A.S.c program.
Abraham Chan
Title: Reliable ML Systems against Faulty Training Data
Abstract: Abraham Chan will present the problem behind faulty training data, and their impact on ML resilience. He will start with existing work that I published in the area, before highlighting some prospective future work directions.
Bio: Abraham Chan is a fourth-year PhD candidate, working in the area of dependable machine learning. His research has explored both application level and hardware faults on ML systems. His advisors are Professors Karthik and Sathish. He is always on the lookout for research internships and industry R&D positions in machine learning and compiler engineering.
Farhad Abbasi
Title: Interleaved CLLC Converters with Dual Phase-Shift Modulation
Abstract: Component mismatch in interleaved CLLC converters causes uneven current sharing and thermal distribution, where phase shift modulation is presented to balance currents in interleaved cells.
Bio: Farhad Abbasi was born in Tabriz, Iran. He received the B.Sc. degree in electrical engineering from Amirkabir University of Technology, Tehran, Iran, in 2013, and the M.Sc. degree in electrical engineering from University of Tabriz, Tabriz, Iran, in 2016. He is currently working toward the Ph.D. degree in electrical engineering with the University of British Columbia, Vancouver, BC, Canada.
He has been also a Research Scholar with Delta-Q Technologies, Burnaby, BC, Canada, since 2020. His current research interests include high-power bidirectional DC–DC resonant converters for battery chargers and renewable energy applications. He is seeking for internship and employment opportunities.
Dias Azhigulov
Title: Electro-optic co-simulation as the enabler of the next generation computing innovation
Abstract: Have you ever wondered how transistors took over the world so quickly? Part of the answer to that question is the invention of BSIM models and its standardization among EDA tools. Nowadays there’s a new kind of technology that aims to solve the next generation of problems – Silicon Photonics. However, to make Silicon Photonics as successful as its electronic predecessors, we need to develop solid models of photonic devices that are reliable and versatile. And this has been my master’s work for the last 3 years, which I’m excited to share with everyone.
Bio: Dias Azhigulov is a 3rd year master student supervised by Prof. Sudip Shekhar and Lukas Chrostowski. Before coming to UBC, he worked as a Data Scientist and gathered extensive coding experience. At UBC, he has been involved in electronic and photonic IC design, modelling of photonic devices. In summer of 2022 he interned at Marvell Semiconductor Inc. where he developed a coherent communications link simulation test bench using his photonic device models and 45 nm transistors. His interests include system level co-integration of photonic and electronic ICs, programming, and Machine Learning.
Phil Kirwin
Title: Integrated Quantum Transducers
Abstract: Useful quantum computers may be built in modules, with optical interconnects linking collections of qubits together. To facilitate such modular design in superconducting and spin-based quantum computing architectures, a coherent microwave-to-optical link is required. These so-called quantum transducers aim to coherently transfer the quantum state of photons between the microwave and optical domains. In this talk, I will give an overview of integrated quantum transducers. I will discuss the key figures of merit, engineering challenges, and leading physical implementations.
Bio: Phillip Suwan Kirwin is a master’s student in the Stewart Blusson Quantum Matter Institute at the University of British Columbia, advised by Lukas Chrostowski, Jeff Young, and Joseph Salfi. His research interests are in integrated quantum photonics and quantum transducers. Phillip holds a BSc in Electrical Engineering from the University of Alberta. He is an alumnus of the Canada-Japan Co-op Program, where he interned in the quantum optical physics group at Nippon Telegraph and Telephone.
Shaurya Patel
Title: CHERI-picking: Leveraging capability hardware for prefetching.
Abstract: DRAM now accounts for over 30% of overall datacenter expense, due to its increasing cost and decreasing scaling. As applications demand more memory, operators look for cost-effective solutions to handle these increasing requirements. One way to address the problem is to use disaggregated or far memory. Far memory solutions have an access latency approximately an order of magnitude slower than DRAM. We introduce a new generalized kernel pointer prefetcher using CHERI: Capability Hardware Enhanced RISC Instructions. Our approach, called CHERI-picking, leverages CHERI pointer capabilities to identify locations that contain pointers and prefetch the pages those pointers reference, subject to a policy.
Bio: Shaurya Patel is a 3rd year Ph.D. student advised by Alexandra (Sasha) Fedorova and Margo Seltzer. His research interests lie in operating systems and memory management.
Andrew Musgrave
Title: Enabling Peer-to-peer Transactions in Measurement-based Distribution System Market
Abstract: This talk presents a measurement-based electricity market structure to establish peer-to-peer (P2P) transactions along with imports from or exports to the upstream network. A key benefit of the proposed P2P market is that participants therein can fully express their proclivities by setting their individual preferences for buying and selling partners independently. Moreover, a linear sensitivity model estimated from online measurements ensures that the resulting P2P transactions satisfy power flow constraints of the underlying distribution system without needing an offline network model. The optimal solution of the OPF problem comprises the P2P transactions (specifying partners, quantity, and price for each trade) and optimal dispatch. The effectiveness of the proposed method is demonstrated via numerical simulations involving a 22-bus system.
Bio: Andrew Musgrave is a M.A.Sc student in the department of Electrical and Computer Engineering at UBC, returning to studies following two years of industry experience as a hardware designer. He completed his undergraduate degree in Engineering Physics (Minor in Mathematics) at Carleton University. He has broad interests in power systems and clean energy and is researching optimization of distributed energy resources and electricity markets. Andrew is seeking opportunities related to power systems following expected graduation in April 2024.
Hanying Liang
Title: Intelligent Robotic Ultrasound: Integrating Multidimensional Information for Enhanced Imaging and Clinical Workflow
Abstract: Ultrasound, as a radiation-free, cost-effective, real-time, and convenient medical imaging modality, plays a crucial role in clinical screening, disease diagnosis, and treatment. Despite its advantages, traditional ultrasound imaging encounters unique challenges that hinder its broader application in clinical practice. The emergence of Robotic Ultrasound (RUS) presents a promising solution with significant potential to enhance the clinical state and workflow of ultrasound. Our research focuses on perceiving and integrating multi-dimensional and multi-source information into RUS, aiming for an autonomous ultrasound scanning procedure with standardized and repeatable imaging results. The primary research goal is to promote the clinical capabilities and applications of ultrasound modalities.
Bio: Hanying Liang is a one-year visiting PhD student in UBC ECE. She received her B.Eng. in electronic information engineering from Tianjin University, Tianjin, China, in 2020. She is currently pursuing a doctorate in biomedical engineering with the Tsinghua University, Beijing, China.Her research interests include medical image processing, medical robotics, and robotic learning.
Hooman Vaseli
Title: ProtoASNet: Dynamic Prototypes for Inherently Interpretable and Uncertainty-Aware Aortic Stenosis Classification in Echocardiography
Abstract: The talk is about a novel classification model for ultrasound video data, that is inherently interpretable by design, thus more trustworthy, and can flag a case if the input data is noisy and can lead to uncertain prediction. It describes the cardiac problem of interest, its challenges in AI-based diagnosis, and my approach on how I use deep learning in an inherently interpretable way (specifically using prototypical neural networks) to classify for the disease and be robust against noisy input data.
Bio: [Seeking Internship] Hooman Vaseli is a PhD candidate at ECE department, in Robotics and Control Lab (RCL), researching deep learning in computer vision and natural language processing for cardiac disease diagnosis. His focus is mostly towards developing models that are interpretable by design. He has industry research experience as an intern at Hitachi Energy. He got his Bachelor’s from UBC ECE as well, in electrical engineering with biomedical specialization. He also enjoys athletic activities, like skiing in winters, and soccer or cycling in summer.