Tutorials
Speaker: Dr. Yi Wang, The Hong Kong Polytechnic University, Hong Kong SAR.
Title: Emerging AI-driven Image/Video Restoration and Its Applications in Compressed-domain Environments
Venue & Time: Room Y411, Building Y, 9:00am-12:00pm, Saturday 1 November
Abstract: Emerging AI-driven image and video restoration in compressed-domain environments is transforming how we handle real-world data degradation and digital forensics, as demonstrated by recent advances in bitstream-corrupted recovery research. This tutorial will explore innovative methodologies that address the challenges posed by unpredictable bit errors and corruption patterns inherent in compressed multimedia streams, such as those encountered in JPEG and H264 bitstreams. Attendees will gain insights into state-of-the-art techniques, including robust decoders with error-resilient mechanisms, two-stage compensation and alignment frameworks that rectify color casts and block shifts, and advanced video inpainting methods tailored for realistic bitstream corruption. These approaches not only enhance restoration quality by leveraging deep learning architectures and novel loss functions but also pave the way for practical applications in areas like telepresence, live streaming, and multimedia forensics. By integrating theoretical underpinnings with practical case studies and benchmark datasets, the tutorial aims to equip researchers and practitioners with the manners necessary to improve image/video quality in compressed-domain applications, ultimately bridging the gap between academic research and real application in a rapidly evolving digital landscape.
Biography: Dr. Yi Wang (Member, IEEE) is a Research Assistant Professor in the Department of Electrical and Electronic Engineering, The Hong Kong Polytechnic University, Hong Kong SAR. He received BEng degree in electronic information engineering and MEng degree in information and signal processing from the School of Electronics and Information, Northwestern Polytechnical University, Xi’an, China, in 2013 and 2016, respectively. He earned PhD in the School of Electrical and Electronic Engineering from Nanyang Technological University, Singapore, in 2021. Dr. Wang worked as a Research Fellow in Nanyang Technological University, Singapore from 2021 to 2023. His research interest includes Image/Video Processing, Computer Vision, Machine Learning, and Digital Forensics. He served as Associate Editor of The Visual Computer. He has served as session chairs/co-chair in international conferences IEEE ICASSP 2024, IEEE ISCAS 2024.
Speaker: Prof. Nianxiong Tan, Zhejiang University, China
Title: Agile Signal Processing Hardware Design and Verification
Venue & Time: Room Y412, Building Y, 9:00am-12:00pm, Saturday 1 November
Abstract: Modern high-performance, low-power applications demand specialized signal processing hardware. Yet designing such hardware using traditional sequential workflows is slow and costly. The conventional flow—starting from algorithm modeling (with floating-point to fixed-point conversion), then manual RTL implementation, and finally verification—makes it increasingly expensive to deliver an optimized design. High-Level Synthesis (HLS) tools (and environments like Simulink) can generate RTL from high level algorithmic descriptions to accelerate development. However, these automated flows give designers litle control over the microarchitecture, which limits opportunities for architectural innovation. They also lack fine-grained optimizations for power and area, otien producing rigid function modules that require a full redesign when requirements change. In recent years, emerging hardware construction languages (HCLs) such as SpinalHDL and Chisel have begun to address these limitations. HCLs allow developers to design at a higher level of abstraction while still preserving low-level control over the implementation. They make it straightiorward to create parameterized hardware generators—flexible designs that can be reconfigured on the fly. By simply adjusting parameters, a single generator can produce many hardware variants, enabling rapid adaptation to evolving requirements without a complete redesign. This generator-based agile design approach dramatically improves development efficiency and encourages architectural exploration. Verification remains an equally significant botleneck. Traditional verification emphasizes matching every bit of hardware output to a golden model, otien using cycle-accurate C/C++ reference models. This approach is time-consuming and focuses on low-level functional correctness rather than the system’s end-to-end performance in real-world conditions. In practice, what ultimately maters is how the signal processing hardware performs on system level metrics under realistic operating conditions. For example, metrics like bit-error rate (BER), signal-to-noise ratio (SNR), and error vector magnitude (EVM) determine success in noisy, interference-prone environments. To that end, we advocate a holistic verification methodology that directly validates these system-level performance targets. Instead of exhaustively comparing every output bit, we simulate real operating scenarios by injecting white noise, interference, and other impairments into the testbench. We then compare the hardware’s performance against a high-level system model (e.g. a MATLAB algorithm) on those key metrics. In this approach, the verification “scoreboard” checks whether the hardware meets the target BER, SNR, EVM, etc., not just whether it produces the exact same bits as the reference. This holistic verification method concurrently ensures functional correctness and that performance goals are met, eliminating redundant low-level comparisons. By focusing on end results, it significantly improves verification efficiency while guaranteeing that the hardware meets its specifications in a realistic environment. These insights are not just theoretical. This tutorial draws on five years of research at Zhejiang University and several successful industrial designs at the State Grid Corporation of China. The agile design and verification techniques presented here have been validated in practice, with multiple projects already in production. In this tutorial, we systematically present the generator based agile hardware design methodology alongside the holistic, performance-centric verification approach. We illustrate their effectiveness with real-world case studies in electrical utility applications, demonstrating how these methods shorten development cycles and ensure robust system performance.
Biography: Founder and CEO of Vango Technologies, Inc. (a State Grid Corporation of China subsidiary), and Professor at the College of Integrated Circuits, Zhejiang University. He received his B.S. (1988) and M.S. (1991) degrees in Electronic Engineering from Tsinghua University, and his Ph.D. (1994) in Electrical Engineering from Linköping University, Sweden. He has held design engineering positions in Sweden, the USA, and China. Notably, he was the design director and lead designer of digital subscriber line (DSL) chips at Globespan Inc., a Bell Labs spin-off that became the world’s top DSL chipset supplier. He also founded AnaLutions, Inc. in the USA and Vango Technologies in China (later acquired by State Grid Corporation). Dr. Tan has received numerous honors, including Chinese national patent awards and provincial Science and Technology Progress Awards. He has authored or edited three books, published over 100 papers, and holds 33 granted U.S. patents (along with many Chinese patents). His current research interests include agile and AI-assisted digital design and verification, as well as mixedsignal chip design for the Internet of Energy.
Speaker: Prof. Jiliang Zhang, Northeastern University, China
Title: Fundamental Wireless Performance of a Building
Venue & Time: Room Y411, Building Y, 2:00pm-5:00pm, Saturday 1 November
Abstract: Over 80 percent of wireless traffic already takes place in buildings. Like water, gas, and electricity, wireless communication is becoming one of the most fundamental utilities of a building. It is well known that building structures have a significant impact on in-building wireless networks. If we seek to achieve the optimal network performance indoors, buildings should be designed with the objective of maximizing wireless performance. So far, wireless performance has not yet been considered when designing a building. I will introduce a novel and interdisciplinary concept of building wireless performance (BWP) to a wide audience in both wireless communications and building design, emphasizing its broad impacts on wireless network development and deployment, and on building layout/material design. I will first give an overview of the BWP evaluation framework proposed in our state-of-the-art works and explain their interconnections. Then, I will outline the potential research directions in this exciting research area to encourage further interdisciplinary research.
Biography: Jiliang Zhang (M’15, SM’19) is currently a full Professor at College of Information Science and Engineering, Northeastern University, Shenyang, China. He is the Head of the Department of Electronics Science & Technology, Northeastern University. His research interests include, but are not limited to wireless channel modeling, modulation system, relay system, vehicular communications, ultra-dense small cell networks, and smart environment modeling. He has pioneered systematic building wireless performance evaluation, modeling, and optimization, with the key concepts summarized in Fundamental Wireless Performance of a Building, IEEE Wireless Communications, 29(1), 2022. He is also the Academic Editor for Wireless Communications and Mobile Computing since 2019. He was the recipient of the IEEE Wireless Communications Letters Exemplary Reviewer Award in both 2021 and 2022.
Speaker: Dr. Huck Yang, NVIDIA Research, USA
Title: Reasoning and Language Modeling System for Applied Signal Processing
Venue & Time: Room Y412, Building Y, 2:00pm-5:00pm, Saturday 1 November
Abstract: In this tutorial, we explore the emerging synergy of “temporal information reasoning” between large language models (LLMs) and applied signal processing, and show how it can enhance the analysis, interpretation, and generation of complex signal-based data. We begin by surveying recent breakthroughs in audio-centric LLM research, which includes emotion-aware prompting, ASR error correction, and descriptive image-quality evaluation. We identify key architectural and training-regime innovations that have driven performance across voice, bio-signal, and image benchmarks. Building on these insights, we introduce several new modular pipelines and test-time adaptation algorithms that integrate pre-trained transformer-based language models with specialized signal encoders and decoders, enabling more robust, real-world information modeling.
Biography: Ph.D.Huck Yang is a Sr. Research Scientist at NV Research, mainly based in Taipei and occasionally in Sunnyvale, and his current focus covering applied signals processing with language modeling and multilingual modeling. He obtained his Ph.D. and M.Sc. from Georgia Institute of Technology, USA with Wallace H. Coulter fellowship.