Special Session 1 – Brain-Inspired Hyperdimensional Computing for Ultra-Efficient Edge AI

Organizers:

  • Mohsen Imani (University of California, Irvine)
  • Xun Jiao (Villanova University)
  • Hussam Amrouch (University of Stuttgart, Germany)
  • Yiannis Aloimonos (University of Maryland, College Park).

Abstract: Hyperdimensional Computing (HDC) is rapidly emerging as an attractive alternative to traditional deep learning algorithms. Despite the profound success of DNNs in many domains, the amount of computational power and storage that they demand during training makes deploying them in edge devices very challenging if not infeasible. This, in turn, inevitably necessitates steaming the data from the edge to the cloud which raises serious concerns when it comes to scalability, security, and privacy. Further, the nature of data that edge devices often receive from sensors is inherently noisy. However, Deep Neural Network (DNN) algorithms are very sensitive to noise, which makes accomplishing the required learning tasks with high accuracy immensely difficult. In this special session, we aim at providing a comprehensive overview of the latest advances in HDC. HDC aims at realizing real-time performance and robustness through using strategies that more closely model the human brain. HDC is, in fact, motivated by the observation that the human brain operates on high-dimensional data representations. In HDC, objects are thereby encoded with high-dimensional vectors which have thousands of elements. In this special session, we will discuss the promising robustness of HDC algorithms against noise along with the ability to learn from little data. Further, we will present the outstanding synergy between HDC and beyond von Neumann architectures and how HDC opens doors for efficient learning at the edge due to the ultra-lightweight implementation that it needs, contrary to traditional DNNs.

Talk 1: Hyperdimensional Computing Platform for Robust and Efficient Perception and Decision Making at the Edge


Speaker: Mohsen Imani, UC Irvine

Abstract: HDC has several advantages over competing learning solutions towards efficient and robust computation. We exploit HDC to develop a hardware computing platform that dynamically learns in the field and at the edge. Our general platform does not rely on task-specific accelerators or large-scale data centers. Instead, it can enable online learning from limited data, handling noise and uncertainty, and intelligent decision-making at the edge. Our HDC-based algorithm is implemented with large vectors, simplifying the “compute” in data-centric applications from linear algebra to simple, low precision, and lower power array operations that can be highly parallelized. Our edge-based hardware platform should support iso-accuracy with state-of-the-art machine ML/AI solutions while enabling (1) at least 100× faster and 1000× energy efficiency than DNNs for online perception at the edge and (2) significant robustness to lack of data or noise for intelligent edge-based decision making.

Bio: Mohsen Imani is an Assistant Professor in the Department of Computer Science at UC Irvine. He is also a director of Bio-Inspired Architecture and Systems Laboratory. He is working on a wide range of practical problems in the area of brain-inspired computing, ML, computer architecture, and embedded systems. His research goal is to design real-time, robust, and programmable computing platforms that can natively support a wide range of learning and cognitive tasks on edge devices. He received his Ph.D. from UC San Diego, and he has a stellar record of publication with over 120 papers in top conferences/journals. His contribution has led to a new direction on brain-inspired HDC that enables ultra-efficient and real-time learning and cognitive support. Dr. Imani research has been recognized with several awards, including the Bernard and Sophia Gordon Engineering Leadership Award, the Outstanding Researcher Award, and the Powell Fellowship Award. He also received the Best Doctorate Research from UCSD, the best paper award in DATE’22, and several best paper nomination awards at multiple top conferences including DAC’19, DAC’20, DATE’20, and ICCAD’20.

Talk 2: Hyper-dimensional Active Perception


Speaker: Yiannis Aloimonos, University of Maryland, College Park

Abstract: Action and perception are often kept in separated spaces, which is a consequence of traditional vision being frame-based and only existing in the moment and motion being a continuous entity. This bridge is crossed by the dynamic vision sensor (DVS), a neuromorphic camera that can see motion. We propose a method of encoding actions and perceptions together into a single space that is meaningful, semantically informed, and consistent by using hyper-dimensional binary vectors (HBVs). We show that the visual component can be bound with the system velocity to enable dynamic world perception, which creates an opportunity for real-time navigation and obstacle avoidance with active perception. Actions performed by an agent are directly bound to the perceptions experienced to form its own “memory.” Furthermore, because HBVs can encode entire histories of actions and perceptions— from atomic to arbitrary sequences—as constant-sized vectors, auto-associative memory was combined with deep learning paradigms for controls. We demonstrate these properties on a quadcopter drone ego-motion inference task and the MVSEC (multivehicle stereo event camera) dataset.

Bio: Yiannis Aloimonos is a Professor of Computational Vision and Intelligence at the Department of Computer Science, University of Maryland, College Park, and the Director of the Computer Vision Laboratory at the Institute for Advanced Computer Studies (UMIACS). He is also affiliated with the Institute for Systems Research and the Neural and Cognitive Science Program. He was born in Sparta, Greece, and studied Mathematics in Athens and Computer Science at the University of Rochester, NY (Ph.D. 1990). He is interested in Active Perception and the modeling of vision as an active, dynamic process for real-time robotic systems. For the past five years, he has been working on bridging signals and symbols, specifically on the relationship of vision to reasoning, action, and language using Hyper-dimensional Computing.

Talk 3: Robust Hyperdimensional Computing Against Hardware Errors and Cyber Attacks


Speaker: Xun Jiao, Villanova University

Abstract: Hyperdimensional computing (HDC), also known as vector-symbolic architectures (VSA), was recently introduced as an emerging AI method mimicking the “human brain” at the functionality level. Currently, HDC research largely has two focuses: applications of HDC to different domains and hardware acceleration of HDC. However, increasing deployment of AI methods in safety-critical systems, such as healthcare and robotics, means it is not only important to strive for high accuracy of AI models, but also to ensure its robustness under even highly uncertain and adversarial environments. In this talk, we will present our recent work in developing robust hyperdimensional computing against uncertainties in both hardware and data. Specifically, we will introduce two studies, one focuses on enhancing the robustness of HDC to voltage-induced hardware errors, while the other focuses on the adversarial testing of HDC.

Bio: Xun Jiao has been an assistant professor in the ECE department of Villanova University since 2018. He obtained his Ph.D. degree from the University of California, San Diego in 2018. His research interests lie in the broad areas of embedded systems, design automation, bio-inspired computing, and machine learning. He received 6 best paper awards/nominations in international conferences such as DATE, EMSOFT, DSD, and CPSCOM. He is an associate editor of IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (IEEE TCAD), Program Committee member of DAC, GLSVLSI, LCTES, ICESS, and COINS.

Talk 4: HW/SW Codesign for Efficient Brain-inspired Hyperdimensional In-Memory Computing


Speaker: Hussam Amrouch, University of Stuttgart, Germany

Abstract: In this part, we will discuss how specialized hardware accelerators beyond von-Neumann architectures, that offer processing capability in where the data resides without moving it, become more and more indispensable to implement efficient edge AI. This is, in fact, indispensable to overcoming the famous von-Neumann bottleneck. We will demonstrate how the emerging Ferroelectric transistor (FeFET) technology has a great potential to realize novel beyond von-Neumann architectures that outstandingly synergize with hyperdimensional computing. We will focus on demonstrating how HW/SW codesign is a key to build reliable HDC on unreliable beyond-CMOS transistors and how abstracted, yet accurate reliability models can be developed in a cross-layer manner bridging the gap between the underlying technology and the HDC algorithms running on top of such novel architectures.

Bio: Hussam Amrouch is a Jun.-Professor heading the Chair of Semiconductor Test and Reliability (STAR) at the University of Stuttgart, Germany. He received his Ph.D. degree with the highest distinction (Summa cum laude) from KIT, Germany in 2015. He serves as an Editor in Nature Scientific Reports. His main research interests are brain-inspired computing, AI processor design, the impact of emerging technologies on future computing, and ML for CAD. He holds eight HiPEAC Paper Awards and three best paper nominations at top EDA conferences: DAC’16, DAC’17, and DATE’17. He has 160+ publications including 65 journals in multidisciplinary research areas across the computing stack, starting from semiconductor physics to circuit design all the way up to computer architectures. He has given so far 25+ invited talks at many international conferences and leading companies (e.g., Synopsys, Advantest, Silvaco) as well as 10+ tutorials in many top EDA conferences like DAC, ICCAD, ICCAD, DATE, and others.