Distinguished Lecturer Program Series

• 

  60minsElementary YouTube

  Deep Unfolding — A Data Driven Approach for Signal Processing —

  Tadashi Wadayama (Nagoya Institute of Technology)

  Deep learning is not only useful for training deep neural networks, but it can also be applied to optimize the internal parameters of "differentiable iterative algorithms" involving input and output. Based on well-known iterative algorithms, it is possible to construct flexible algorithms with data-based learnability by embedding learnable parameters within them. This approach is known as deep unfolding. This talk provides an introduction to and an overview of deep unfolding, focusing on sparse signal recovery algorithms, which represent one of the linear inverse problems. In the first half of the talk, we will present the basics of deep unfolding by showing several examples, including gradient descent, projected gradient descent, and the proximal gradient method. In the second half, we will discuss a sparse signal recovery problem and present a sparse signal recovery algorithm derived through deep unfolding, explaining the observed convergence acceleration in algorithms constructed based on deep unfolding.

  Check

• 

  60minsElementary Livestream: Monday, March 18, 15:00 PM JST/KST (GMT+9)YouTube

  Swarm Intelligence Optimization and Machine Learning

  Kenya JIN’NO (Department of Intelligence Systems, Tokyo City University)

  In recent years, deep learning has exhibited remarkable performance, leading to the proposal of numerous applications characterized by high accuracy and efficiency. Deep learning systems learn their parameters through gradient descent methods, which rely on gradient information of the loss function to minimize the output loss. Given the vast number of parameters within deep learning systems, the associated loss functions are anticipated to possess numerous local minima. Consequently, the learning process can become ensnared in these local solutions, potentially halting progress and preventing the achievement of desired performance levels. To circumvent these issues, the adoption of swarm intelligence optimization methods, which do not depend on gradient information for optimization, has been proposed. These methods offer an alternative means of training deep learning models by leveraging the collective behavior and intelligence of groups. This presentation will explore the outcomes and challenges associated with employing swarm intelligence optimization for deep learning. By examining these aspects, we aim to shed light on the potential of these optimization techniques to enhance the learning process and overcome the limitations posed by conventional gradient-based methods.

  Check

• 

  90minsElementaryIntermediate Livestream: Wednesday, February 28, 10:30 AM JST/KST (GMT+9)YouTube

  Reading of "Reading" for "Actuating": Augmenting Human Reading and Learning by Sensing and Actuating Technologies

  Koichi Kise (Osaka Metropolitan University)

  Reading constitutes a pivotal element in the architecture of our knowledge acquisition processes. Analogous to the adage, "You are what you eat," it is equally pertinent to assert that our intellectual fabric is woven from the threads of our reading material; a substantial portion of our knowledge is derived from this fundamental activity. This presentation delves into the concept of "meta-reading" — the analysis of reading behavior — as a window into the reader's interests, cognitive level, and psychological state. This talk will unveil the findings of our recent research, which navigates through the intricate landscape of reading and learning behaviors. Specifically, it focuses on the deployment of various feedback mechanisms, or 'actuators', tailored to enhance the reader's engagement and comprehension. The efficacy of these actuators is contingent on their alignment with the individual characteristics of the reader. In addition to presenting these insights, the talk will showcase practical applications, demonstrating how the nuanced understanding of reading behavior can be leveraged to optimize learning processes. Case studies and examples of effective actuation strategies, grounded in our behavioral analysis, will be highlighted. Furthermore, the talk will commence with an introductory overview of the German Research Center for Artificial Intelligence (DFKI) and its Japan Laboratory. This is to provide context, as much of the research discussed is affiliated with the pioneering efforts undertaken at DFKI.

  Check

• 

  60minsElementaryLivestream: Wednesday, December 27, 10:00 AM JST/KST (GMT+9)YouTube

  From QoS to QoE: Review of Changes in Service Quality Research

  Tatsuya Yamazaki (Niigata University)

  The recent focus in networking has shifted from Quality of Service (QoS) to Quality of Experience (QoE). QoS traditionally dealt with mechanisms and technologies for controlling network traffic and managing network resources, often through differentiated handling and resource allocation for high-priority traffic. Although QoS evaluation involves parameters like bandwidth, throughput, latency, delay jitter, and packet loss for network systems, human intervention is rarely considered in this context. In contrast, QoE has become a key metric for assessing user satisfaction with service quality. QoE is subjective and linked to user perceptions and expectations, leading to ongoing research challenges. These challenges include how to measure subjective QoE, how to connect QoE with QoS parameters related to system performance, and how to integrate QoE into multimedia and communication services. Researchers have made significant contributions in addressing these issues. QoE's definition by ITU-T as "the overall acceptability of an application or service" is somewhat ambiguous, so it is more practical to focus on specific applications to discuss QoE. Currently, video streaming and Web applications are prominent areas of QoE research due to their high user demand. In this lecture, the historical evolution of QoS and QoE is discussed, with a particular emphasis on current QoE research topics, especially in the context of video streaming. Researchers are making strides in understanding and improving the user experience in these application domains.

  Check

• 

  90minsElementaryLivestream: Tuesday, November 14, 3:00 PM JST/KST (GMT+9)YouTube

  AI and ethics

  Prof. Hiroshi Nakagawa (RIKEN Center for Advanced Intelligence Project)

  The following three themes of AI ethics will be presented. 1. Problems of generative AI: Text-generating AI such as ChatGPT and image-generating AI such as Midjouorney have started to be widely used, and their limitations and abuses have been discussed. This paper presents an analysis of these issues from ethical and other perspectives. 2. Avatars are being used more and more recently. when they become autonomous, a generative AI is expected to be utilized for the conversational ability of these avatars. In such a situation, issues such as trust of autonomous avatars will be discussed. 3. AI ethics has been discussed in the EU for recent years, and has been proceeding toward legislation in the form of the Trustworthy AI Guidelines, the AI White Paper, and the AI act. We briefly explain this process of the current status of AI ethics in the EU, including technical aspects.

  Check

• 

  60minsElementaryLivestream: Tuesday, October 26, 3:00 PM JST/KST (GMT+9)YouTube

  Explore the Latest Fundamentals of AI-based Computer-Aided Diagnosis (AI-CAD) for Medical Imaging

  Prof. Hiroshi FUJITA (Gifu University)

  With AI "deep learning" technology, which is a type of "machine learning" (learning functions and rules) in which computers learn by themselves, the accuracy of image recognition has reached a level that exceeds that of humans. Computer-aided diagnosis of medical images, so-called CAD, has rapidly entering the mainstream of practical medicine. Especially in the detection of breast cancer by mammography (breast imaging), it is a part of daily clinical work. In this case, the computer output is used as a "second opinion" to help the doctor interpret the image. However, recent powerful AI technologies, including deep learning, have taken CAD development and performance to the next level, traditional CAD has diversified, and even autonomous diagnostic AI is emerging. This is sometimes called AI-CAD and is gradually shifting from a mere R & D level to a commercialization level, verification at the actual clinical stage, and insurance reimbursement stage. In this talk, we would like to examine and discuss the current state of AI-CAD, and the problems that need to be solved in order to make AI-CAD more practical in clinical practice. Through this presentation, we hope that many audiences will become more interested in AI-CAD for medical imaging and will engage in research and development in this field. References: 1) H.Fujita, “AI-based computer-aided diagnosis (AI-CAD): The latest review to read first,” Radiological Physics and Technology (RPT), vol.13, no.1, pp.6-19, 2020, received Most Citation Award (2023.4) from RPT journal. 2) G.Lee and H.Fujita (Eds.), “Deep Learning in Medical Image Analysis: Challenges and Applications,” Springer, 2020.

  Check

• 

  65minsElementaryYouTube Premiere Session: Saturday, September 30, 6:30 PM JST (GMT+9)YouTube

  Introduction to optical circuit

  Prof. Hiroshi Takahashi (Sophia University)

  In this introductory talk, I will review the fundamentals of optical circuit devices, which are essential components in optical communication systems. An optical circuit consists of optical waveguides with a width of several microns formed on a substrate. It is highly compact and can be mass-produced at a low cost. First, I will derive the wave equation and represent the electric field for optical waveguides as mathematical notation to describe light within optical circuits. Then, I introduce the definition and usage of the transfer function to explain the operating principles of the following optical circuit devices. The optical switch is based on the Mach-Zehnder interferometer (MZI) optical circuit with an integrated electric micro heater. It is employed for altering the light path within optical fiber communication networks. The transmission characteristics of asymmetric MZI circuits exhibit periodic frequency dependence and can be utilized as a frequency filter and frequency discriminator. The arrayed-waveguide grating (AWG) consists of hundreds of waveguides and functions similarly to a diffraction grating. It is used as a wavelength demultiplexer in wavelength division multiplexing optical communication systems. The optical circuit for a coherent receiver is constructed using two 90-degree hybrid circuits and a polarization splitter circuit. It is used in conjunction with photodiodes and various electronic components to extract the phase of the modulated optical signal. Phase modulation is an essential technique for achieving ultra-high-speed communication exceeding 100 Gbit/s, making this device of utmost importance.

  Check

• 

  60minsIntermediateAdvancedLivestream: Tuesday, August 29, 16:00 PM JST/KST (GMT+9)YouTube

  Two-dimensional rectangular-coordinate orthogonal multiplexing antenna system for non-far field region communication

  Prof. Jiro Hirokawa (Tokyo Institute of Technology)

  This talk presents an orthogonal multiplexing antenna system for non-far field region communication, which is based on two-dimensional polarity in the rectangular coordinate system. In other words, it is based on two-fold symmetry and anti-symmetry between the transmission modes. The two-dimensional polarity comes from the structural symmetry of magic-T components used in the corporate-feed circuit in the antenna in the millimeter-wave band. The structural symmetry gives wideband operation. The feasibility is shown by a system where a corporate-feed waveguide slot array antenna and a monopulse circuit are combined. We experimentally confirm 4-multiplex 40cm CW transmission using an 80GHz-band 16x16-slot waveguide array, 8-multiplex 20cm CW transmission using a 60GHz-band 16x16-slot dual-polarization waveguide slot array, and 2-multiplex 40cm QPSK transmission using a 60GHz- band 16x16-slot waveguide array.

  Check

• 

  90minsElementaryIntermediateLivestream: Tuesday, July 25, 4:00 PM JST (GMT+9)YouTube

  Analysis of Complex Dynamical Behavior as a Temporal Network

  Prof. Tohru Ikeguchi (Tokyo University of Science)

  In this webinar, we will discuss the analysis of time-varying complex phenomena by considering measured contact data as a temporal network. Firstly, we will introduce some of the contact data currently recorded. Then, as an elemental technique for analyzing these contact data as temporal networks, we explain the analysis method for static networks. Secondly, we explain the importance of analyzing such contact data as temporal networks. We also explain how to transform contact data into temporal networks. Thirdly, we explain the distance measure between temporal networks in order to detect and quantify system dynamics from the transformed temporal networks. Furthermore, we explain how to analyze the dynamics of the changes in the contact data by converting the temporal changes in the distance into time series signals using the classical multidimensional scaling method. Finally, we conclude the methods for analyzing contact data as a temporal networks,and discuss a future direction of network analysis.

  Check

• 

  90minsIntermediateYouTube Premiere Session: Friday, June 30, 6:30 PM JST (GMT+9)YouTube

  Invitation to Vision and Language

  Prof. Yoshitaka Ushiku (OMRON SINIC X Corp.)

  This talk aims to introduce the technical area called Vision and Language to those with experience in research and development applying deep learning to images and texts. Until the end of the 2000s, machine learning for images and text was independently studied in computer vision and natural language processing, creating distinct technical systems. As a result, the mutual barriers to entry for technologies that handle both images and text were a challenge. When deep learning enabled images and text to be analyzed with similar neural network architectures, this mutual barrier to entry was significantly mitigated. Thus, an integrated understanding of images and text, which had begun to be studied, blossomed all at once. In this presentation, I will introduce such research fields of Vision and Language, broadly classifying them according to the modality of input/output (i.e., image or text). Vision and Language can also be considered as the study of multimodal/cross-modal machine learning; there are some points to be aware of. I will discuss some of those points in the last part of this talk. Nowadays, generative models, which generate images from texts, have attracted a great deal of attention, and I expect that many people are interested in this field. Through this presentation, I hope that such people will become more interested in Vision and Language and will engage in research and development in this field.

  Check

• 

  110minsElementaryIntermediateYouTube Premiere Session: Wednesday, May 31, 6:30 PM JST (GMT+9)YouTube

  Basics and recent progress in lightwave modulators

  Prof. Tetsuya Kawanishi (Waseda University)

  The purpose of this webinar is to provide basics of electro-optical modulation which plays an important role in photonic networks such as fiber optic links, visible light communications, and fiber wireless, as well as in high-performance sensing systems. This webinar will offer as an overview of various types of optical modulation devices, focusing on modulators based on the electro-optic effect (Pockels effect), called electro-optic (EO) modulators, which induce optical phase retardation by an electric field applied thoroughly through an electrode. By using optical interferometers based on the EO effect, a variety of functions can be achieved, including phase modulation, intensity modulation, and vector modulation. Mach-Zehnder interferometer modulators (MZMs) are very useful for the most common modulation schemes, such as on-off keying (OOK) and binary phase shift keying (BPSK). In addition, MZMs would be elements for more complex modulators, high-speed optical switches, vector modulators, etc.

  Check

• 

  45minsIntermediateYouTube Premiere Session: Friday, April 28, 6:30 PM JST (GMT+9)YouTube

  Introduction of Electromagnetic Information Security

  Prof. Yuichi Hayashi (Nara Institute of Science and Technology)

  With the rising importance of information security, the necessity of implementing better security measures in the physical layer as well as the upper layers is becoming increasing apparent. Given the develop-ment of more accurate and less expensive measurement devices, high-performance computers, and larger storage devices, the threat of ad-vanced attacks at the physical level has expanded from the military and governmental spheres to commercial products. In this paper, we review the issue of information security degradation through electromagnetic (EM)-based compromising of security measures in the physical layer (i.e., EM information security). Owing to the invisibility of EM radia-tion, such attacks can be serious threats. We first introduce the mecha-nism of information leakage through EM radiation and interference and then present possible countermeasures. Finally, we explain the latest research and standardization trends related to EM information security.

  Check

• 

  90minsElementaryIntermediateLivestream: Wednesday, March 15, 10:30 AM JST/KST (GMT+9)YouTube

  Locating and Controlling Chaotic Saddles

  Prof. Tetsushi Ueta (Tokushima University)

  Some planar discrete nonlinear systems may have a horseshoe structure with stable and unstable manifolds associated with a saddle fixed point. Under this situation, as is well known, there are infinitely many unstable periodic points around the saddle. By making correspondence of the cross sections of manifolds and symbolic dynamics, one can obtain a “candidate area” containing an unstable periodic point whose period is given by the gray code. These unstable periodic points can be stabilized by applying any chaos control technique into that candidate area, for which the external force control method exhibits a very robust control performance. In this talk, a horseshoe structure derived by the Poincaré mapping for a two-dimensional nonautonomous hybrid system is analyzed as an example. As an example, locating scheme for unstable periodic points is demonstrated, and their control results are presented.

  Check

• 

  90minsBeginnerIntermediateYouTube

  Augmenting Humans — From Physiology Signals towards Digitalized Emotion

  Prof. Kai Kunze (Keio University)

  In my research, I combine design and technology to augment human senses, looking for novel interaction paradigms. Engineering and computing have often been used to mimic or surpass some human abilities (for example autonomous driving, playing Go). Such efforts appear to put humans and computers in a competitive relationship, as emphasized in AI vs. Human game competitions. Once fantastic fear of AIs “replacing” human workers is now taken much more seriously and discussed in the public sphere. My research proposes a different approach to the human-computer relationship by applying a cooperative and empowering framework, using wearable computing to actively augment human capabilities. At the beginning of my talk, I will discuss our efforts to augment audiences with physiological sensors to understand more about audience/performer entrainment and create feedback spirals. I explore the concepts of entrainment and what it means to be live in person at a meeting/concert or other gathering. We discuss application scenarios of capturing inter-personal synchrony between audience and performers as well as potential body schema extension technologies.

  Check

• 

  60minsIntermediate

  Study of Propagation Characteristics of Terahertz Waves in Beyond 5G

  Prof.Akihiko Hirata (Chiba Institute of Technology)

  The use of 300 GHz band wireless links is currently being considered for various use cases such as backhaul/fronthaul lines, mobile access, wireless local area networks (LANs), and body area networks (BANs) as a way of achieving >100-Gbit/s data rate. In order to make these use cases practical, it is necessary to evaluate the radio propagation characteristics in the 300 GHz band and build a radio propagation model at this frequency. To build a radio propagation model for each of the above use cases, we conduct various radio propagation experiments and simulations. In this lecture, I will present the results of these experiments and simulations of radio propagation in the 300 GHz band.

  Check

• 

  90mins

  Past, present, and future of Internet congestion control

  Prof.Go Hasegawa (Tohoku University)

  Internet Congestion control is essential not only from the performance perspective but also for its stability and robustness. Congestion control has been the subject of continuous research and development since the birth of the Internet.
  In this lecture, we review the Internet congestion control mechanisms that have been proposed, studied, and implemented so far. First, we summarize the existing congestion control mechanisms from the viewpoint of metrics for the degree of network congestion, that is, loss-based, delay-based, bandwidth-based mechanisms, and their hybrid mechanisms.Next, we introduce the history of the congestion control deployed for specialized network environments such as wireless networks, high-speed and long-delay networks, and datacenter networks.
  We further introduce the congestion control mechanisms which utilize some information obtained from the network, such as Explicit Congestion Notification (NCP) and eXplicit Control Protocol (XCP). Some existing mechanisms which introduce in-network control, such as Performance Enhancing Proxy (PEP), are also explained.
  We then explain recent research trends in congestion control mechanisms, including the congestion control plane that implements congestion control mechanisms in user space, SDN-based congestion control, and machine learning-based congestion control.
  Finally, the lecturer introduces his recent research on in-Network Congestion Control architecture (NCC), which breaks away from the end-to-end principle on which the existing Internet congestion control relies, and divides the network between the transmitting and receiving terminals into multiple sub-networks, and applies congestion control algorithms algorithm for each sub-network separately.

  Check

• 

  90minsElementaryIntermediate

  Brainmorphic Computing Paradigm and Hardware

  Prof.Yoshihiko Horio (Tohoku University)

  In order to realize brain-specific computational schemes such as consciousness/subconsciousness, prediction, inference, memory, attention, self, etc., novel computational principle, architecture, and devices are required, which would be very different from those in von Neumann-type digital computers. This talk introduces the brainmorphic computing paradigm and its hardware architecture. In the brainmorphic computing paradigm, information processing is performed through imitating the anatomical and physiological structure and dynamics of the brain, which have been evolutionally developed. The latest knowledge from brain science, especially, on high-order brain specific functions coemerged from high-dimensional complex structure and neuro-dynamics, should be reflected in the brainmorphic computing paradigm. In addition, the bodily and environmental constraints are considered and utilized as embodiment in this computational paradigm. In order to implement the brainmorphic computing paradigm as hardware, physical and dynamical characteristics of the novel brain-type constituent devices, especially nonlinear analog circuits, and nanometer-scale materials and devices, should be naturally utilized. Our efforts on physical implementation of the brainmorphic computing hardware are illustrated; for example, spin-orbit torque devices for naturally realize neuronal and synaptic dynamics directly, and a chaotic neural network reservoir for computation through high-dimensional complex dynamics.

  Check

• 

  60minsElementaryIntermediate

  Introduction to Bioelectromagnetics – Fundamentals and Applications

  Prof.Akimasa Hirata (Nagoya Institute of Technology)

  The human body whose electrical behavior depend on the frequency should be considered in the modern design of wireless devices which are used in proximity of the human body. However, the complexity of the modeling may depend not only on the frequency but also the individual variabilities, environmental factors, size of the devices and use scenarios. In this lecture, first, a discussion on modeling of the human in the electromagnetics for different cases is provided. Then, the effect of the human presence on the wireless devices including wireless communication terminals and wireless power transfer systems is explained considering typical exposures scenarios. The electromagnetic fields emitted from the electrostatic discharge is also discussed.

  Check

• 

  40minsIntermediateSampling theoremAntenna arrayAntenna diversity

  MIMO and Smart Antenna with Single RF-Baseband Chain and Switching Antenna Array

  Prof.Makoto Taromaru (Fukuoka University)

  MIMO, diversity, or adaptive array antenna, so called “smart antenna,” is an effective technique by space-domain processing to enhance channel capacity or to mitigate multipath fading and interference. However, it is difficult to reduce its cost, circuit size, and power consumption because it needs one by one RF-to-baseband chain for each array element; large scale RF/baseband hardware is necessary especially for large number of antenna elements i.e., massive MIMO and so on. To solve this problem, fast periodical antenna switching or beam scanning techniques have been proposed, where we call them “switching antenna array”. Although these are very attractive since MIMO or digital beamforming can be done with a single analog RF/baseband chain, they cannot perfectly have the same properties as the conventional multi receiver/transmitter architecture due to antenna switching or sampling. In this talk, spurious response/emission, and SNR (signal-to-noise ratio) property of the switching antenna array, are theoretically analyzed, and BER: bit error rate performance is quantitatively evaluated.

  Check

• 

  90minsIntermediate

  Technologies and Commercial Applications of Far Field Wireless Power Transfer

  Naoki Shinohara (Kyoto University)

  A far field Wireless Power Transfer (WPT) is one of innovative radio wave applications and is recently considered as one of game changing technologies based on radio wave. It had been studied historically with microwaves and its application has been narrow beam WPT in which the wireless power is transmitted from one transmitter to one receiver with high beam efficiency. Recently, the WPT is applied as wide beam WPT in which the wireless power is transmitted from one/multi transmitter(s) to one/multi receiver(s) with various radio waves like a wireless communication system. In this talk, I introduce basic theory and technology, recent innovative technologies, and commercial applications of the far field WPT in the world. It involves an antenna / beam forming technology and a circuit technology in a receiver. Additionally, I also introduce discussion status of new radio regulation for the far field WPT in Japan and in the world.

  Check

• 

  90minsIntermediateArtificial intelligence (AI)Neural network hardwareEdge computingPhotonicsSpinPattern information representation and processing

  Physical Reservoir Computing: Its Advantages and Significance

  Prof.Akira Hirose (The University of Tokyo)

  This talk focuses on reservoir computing and its physical realization in the AI & sensor-network era. First, we have a glance of the hardware history in artificial intelligence (AI) including logic architecture and neural networks. The key point in modern AI is emphasized, namely, pattern information representation and pattern information processing. We also present the basic idea and dynamics of reservoir computing by referring to deep learning and time-serial information processing. Then we look through various physical reservoir-computing ideas briefly to catch the variety of physics used in this area and also the key properties. In particular, we go into the details of reservoirs based on photonics and spin waves. The former has been studied for long, resulting in an accumulation of experiences. The latter is new and rich in nonlinearity, hysteresis, anisotropy and dispersion, which enhance the processing ability. Lastly, we discuss the prospects of reservoir computing in the near future society in the context of edge-computing as well as server use in the sensor-network society.

  Check

• 

  40minsAdvancedwireless communicationterahertzCMOS6Gbeyond 5Gtransceiver300 GHz

  Sub-Terahertz Transceivers in Silicon

  Prof.Minoru Fujishima (Hiroshima University)

  The 300GHz band is a frequency band recently identified for wireless communication and is attracting attention towards 6G.However, it is generally believed that as the frequency increases, the communication distance becomes shorter and the applications are limited. In this talk, the relationship between frequency and communication distance will be explained, and the possibility of a 299GHz transceiver using CMOS integrated circuits will be discussed.

  Check