Please note that the time for lunch break is 12:30PM-1:30PM on January 5th ; The ICCE 2024 conference does not provide free lunches for the participators of tutorials, and the participators could find restaurants inside the hotel of the conference venue.
SRI International
This hands-on tutorial will touch upon a few different aspects of training and deploying deep learning-based models to edge or consumer devices. After reviewing basic and contemporary concepts such as self-supervised pre-training, we will deploy a simple app. We will gradually cover concepts that enable us to adapt the model to the users’ needs, making the model better, smaller, and faster on the edge device. Topics covered will include Masked Modelling, Hyperdimensional Computing, and Federated Learning. A basic ASIC understanding of machine learning terminology is useful but not required.
Aswin Raghavan, PhD, is a Senior Computer Scientist at SRI International's Center for Vision Technologies. Aswin received his PhD in Computer Science from Oregon State University in 2017, during which he studied probabilistic planning and deep learning. Since joining SRI, he has contributed to the field of Edge Intelligence and created Deep Learning solutions that can operate at the edge with low Size, Weight, and Power (SWaP), spanning topics such as quantization learning, data thinning, Hyperdimensional Computing, adversarial robustness, and few-shot domain adaptation at the edge. Currently, he leads fundamental research projects as a Principal Investigator (PI) and leads multi-institution teams.
STMicroelectronicsIEEE Fellow in Computer Society
Minimize memory footprint, maximize computational efficiency and accuracy are great challenges when trying to use low bit-depth neural network for microcontroller deployment. Deeply Quantized Neural Networks (DQNNs) offer the most interesting approach to these. The design and the training of DQNN also is not a trivial task. Unfortunately, current off-the-shelf microcontrollers are not yet able to exploit their potentialities. Furthermore, a ML practitioner shall learn how to design DQNN using an experimental deep learning tool such as QKeras to achieve interesting accuracies compared to more traditional design approaches. In this talk all those aspects will be discussed with reference to latest efforts of ST including a) learning QKeras and b) tools for efficient DQNN deployment on micro controller. Part of the talk will include associated demo and code inspection.
Danilo PAU (h-index 26, i10-index 69) is a ST Fellow in STMicroelectronics, Agrate Brianza Italy. Graduated in 1992 at Politecnico di Milano. He worked on memory reduced HDMAC HW design, MPEG2 video memory reduction, video coding, transcoding, embedded 2/3D graphics, and computer vision. Currently, his work focuses on developing solutions for tiny machine learning tools. Since 2019 Danilo is an IEEE Fellow and AAIA Fellow on 2023; with 78 and 68 respectively European and US application patents, 166 publications, 113 ISO/IEC/MPEG authored documents and 79 invited talks/seminars at various Universities and Conferences, Danilo's favorite activity remains supervising undergraduate students, MSc engineers and PhDs.
MathWorks (MATLAB)
Crafting and deploying accurate but low complexity deep learning networks on microcontrollers requires an easy, systematic, and very productive end-to-end workflow. In this tutorial, we present a comprehensive approach that combines the power of advanced optimization techniques and cloud-based deployment tools to address the challenges of model selection and performance evaluation on tiny devices. The MATLAB workflow leverages Bayesian optimization, a powerful complexity bounded technique for hyperparameter tuning, to guide the selection of the most suitable deep learning model architecture. IoT practitioners usually design and train a range of models, exploring different topologies, layers, and hyperparameters. We then introduce a cutting-edge platform for deploying and optimizing deep learning models on STM32 microcontrollers. The practitioners will gain insights into the trained models by evaluating performance benchmarks against the stringent requirements of tiny devices at the edge. Throughout the tutorial, the practitioners will engage in practical exercises, applying the end-to-end workflow to real-world examples. By the end, they will possess the skills to navigate the complexities of designing tiny deep learning models and performance evaluation to achieve deployment goals.
(See A2 for Dr. Pau.) Dr. Brenda Zhuang is a consulting engineer and development manager at MathWorks, where she leads a team responsible for software tools for automatic deployment of embedded applications, such as motor controls and deep learning, to microprocessors and FPGAs. Brenda joined MathWorks in 2007. Her contributions include key features in the Simulink product family. She received her Ph.D. from Boston University in systems engineering, developing optimal scheduling approaches in smart sensor networks. She is strongly involved with the international research community of numerical simulation and design automation. She also serves on the technical program committee for many international conferences in control theory, modeling, and simulation. She created hands-on workshops based on emerging technologies, such as drones, robots and hyperspectral cameras and presented in conferences such as Grace Hopper Celebration and WiDS. Jack Ferrari is a product manager for MATLAB Coder at MathWorks, focused on C/C++ code generation for deep learning models. He is also the product manager for Deep Learning Toolbox Model Quantization Library, enabling MATLAB users to compress and deploy AI models to edge devices and embedded systems. He primarily works with embedded AI developers from the automotive and aero-def industries. Jack holds a B.S. in Mechanical Engineering from Boston University
UCSDXNIT InstituteUniversity of Novi Sad
In the last two years, we have witnessed the increased push to legislators to approve the robotaxi vehicles in several cities in the USA for commercial use, with two permits issued, and one revoked. The endeavor to proliferate self-driving vehicles proves to be very hard, since the technology stack required, dominated by software, presents extreme challenges in safety critical design and certification according to practices laid out in ISO 26262 and ISO 21448 SOTIF. In this tutorial we would contrast the due dilligence in automotive functional safety with the real-world design challenges and what has been deployed on roads. We would discuss legal debates around the technology shortcomings following the first lawsuits regarding the casualties caused by the car autopilots. Finally, we would work out in a hands-on example a process of designing a safe autonomous vehicle computer for a traffic jam pilot function, witnessing all the challenges first hand.
Dr. Bjelica is the Associate Professor at the FTN Uni Novi Sad, a Functional Safety Instructor at the University of California San Diego, and the CEO of NIT Institute. He actively participates in research and innovation activities for various computer-engineering sectors, focusing on consultancy and training in the fields of system safety, functional safety, automotive engineering, and consumer electronics. During his career, he consulted companies in the automotive industry (ZF Germany, TTTech Austria, Qualcomm Automotive USA, Daimler Germany) and also other companies in consumer electronics, industrial machinery, and computing domains. He is a frequent participant and a speaker at major industry events worldwide. Dr. Bjelica holds a Ph.D. degree in computer engineering from the University of Novi Sad, Serbia, as well as an Academic Safety Engineer degree from the FH CampusWien - Vienna Institute for Safety and Systems Engineering (Austria). His professional and research focus is on complex System and software architectures with specific interactions and virtualization. He authored over 100 publications across major journals and scientific conferences and holds 30 patents.
University of Strathclyde
Recently, semantic communication has attracted huge attention from researchers and industries due to its capabilities to use in future media and mobile systems. However, implementation of semantic communications for the general communications system is a key challenge in the journey towards 6G. While promising significant improvements in channel bandwidth efficiency to go beyond what is generally considered as Shannon theory-based communications, semantic communications create new opportunities with modern high bandwidth applications. The need for a highly efficient communication model has become of paramount importance due to the exponential growth in data traffic on image transfers, streaming, video conferencing, and online learning, which consumes a significant amount of system resources and necessitates longer transmission and processing times. This tutorial focuses on introducing the theory of semantic communication in relation to multimedia content in a mobile communications system.
Anil received his BSc. (Hons.) degree in Electronics and Telecommunications Engineering from the University of Moratuwa, Sri Lanka in 1995and his MSc degree (with Distinction) in Telecommunications from the Asian Institute of Technology (AIT), Thailand in 1997. He completed his PhD in Video Coding and Communications from the Department of Electrical and Electronic Engineering, University of Bristol, UK, in 2001. He is currently a Professor at the Department of Computer and Information Sciences, University of Strathclyde, Glasgow, UK and is leading the multimedia communications research group. His research interests include video processing/coding and communications, Artificial Intelligence and Machine Learning, resource optimization, Quality of Experience, intelligent video encoding for wireless systems, and video communication in 5G/6G. He has published over 450 peer reviewed articles in international journals and conference proceedings on these domains and graduated over 100 PhD students. Anil led the video communications research group at the University Surrey between 2007 and 2021. Due to his contributions towards video communications, his work has been recognized by international research community though the IEEE International Shall award and NAB 2020 award.
Consumer Reports
The Internet of Things has brought great benefits to the consumer, and connected products are becoming prevalent throughout the home. The proliferation of IoT products raises issues regarding the collection and use of consumer data, as well as digital privacy and security of the devices. This tutorial will discuss consumer attitudes about digital privacy, examine concepts related to privacy and security of Internet of Things (IoT) products, and explore consumer-focused guidelines for improving these products.
As the Senior Director of Product Testing at Consumer Reports, Maria Rerecich leads teams who evaluate and rate the performance of consumer products. Maria is involved in CR’s initiatives to tackle privacy, security, and data concerns, focusing in particular on the testing of Internet of Things (IoT) devices. She led a pilot test of several mobile applications, which resulted in an app developer making immediate improvements to protect consumers’ data and privacy. Prior to joining Consumer Reports, she worked for 29 years for Standard Microsystems Corporation in the semiconductor industry and was responsible for integrated circuit design, validation, and product engineering of silicon chips used in PCs. Maria is on the National Institute of Standards and Technology IoT Advisory Board. She holds bachelor’s and master’s degrees in electrical engineering from the Massachusetts Institute of Technology.
IP Action Partners, Inc.IEEE Fellow
The speaker will present frameworks and systematic approaches to plan an R&D project portfolio that aligns with the overall business strategy of your organization. Because R&D funds and engineers are limited resources, ensuring you are making well founded resource allocations to the right projects will contribute to the success of your organizations.
Stu Lipoff is an IEEE Life Fellow and past president of IEEE CTSoc, past VP of publications, and VP of Industry and Standards Activities for CTSoc. He is currently serving as The Historian for CTSoc. He has also been active in the organizing and conduct of The IEEE International Conference on Consumer Electronics (ICCE). In his day job, he is president of IP Action Partners Inc which provides contract engineering, technology assessment, and consulting services to stakeholders in the Telecom, Info Tech, Media, and Electronics (TIME) industries. Among his major contributions has been leading the effort that developed the first DOCSIS cable modem specifications and developing adopted recommendations for hybrid fiber optics cable TV networks. In work performed under contract to MSTV and NAB, his analysis supported recommendations adopted by the FCC for a time-table for the rollout of digital TV. In projects performed over nearly 50 years he has worked in nearly every industry from consumer to military and with a wide range of technologies. His expertise includes RF and communications systems. Previous employment was with Motorola, Bell & Howell, and Arthur D Little, Inc. Mr. Lipoff has earned BA, BSEE, BS Eng Physics, MSEE, and MBA degrees. He is a licensed professional engineer in Massachusetts and Nevada as well as holding a Certificate in Data Processing (CDP).
CTSoc Administrator Charlotte Kobert charlotte.kobert@ieee.org
