Assoc. Prof. Dr. Supavadee Aramvith
1. Title: AI based Video Analytics
Abstract: In this talk, we will present and discuss the current trends and researches in video analytics. As surveillance cameras have been widely installed worldwide, although the main purpose of those cameras is for monitoring, but the most important task is to be able to analyze video contents and extract useful information. Artificial Intelligence (AI) and Deep learning-based computer vision techniques utilizing multi-layer neural network is drastically improving the performance of video analytics to a certain extent. Several on-going researches on deep learning-based video analytics such as image super resolution, real-time multiple face recognition system, video anomaly detection and several implementations of embedded video analytic system on FPGA and Single Board Computers will be discussed. Some use cases of utilizing video analytics will also be mentioned.
2. Title: Video Anomaly Detection for Intelligent Surveillance System
Abstract: Video anomaly detection has widely gained popularity for intelligent surveillance systems in recent years. Most works have struggled with challenging tasks such as detecting and localizing objects in complex and crowded scenes, especially with the object localization in a pixel-level evaluation. In fact, they can achieve either frame-level anomaly detection or pixel-level anomaly localization in some complex scenes. In this talk, we will present and discuss our proposed framework based on Deep Spatiotemporal Translation Network (DSTN), novel unsupervised anomaly detection and localization method based on Generative Adversarial Network (GAN) and Edge Wrapping (EW ). Our DSTN has been tested on publicly available anomaly datasets, including UCSD pedestrian, UMN, and CUHK Avenue. The results show that it outperforms other state-of-the-art algorithms with respect to the frame-level evaluation, the pixel-level evaluation, and the time complexity for abnormal object detection and localization tasks.
3. Title: Super Resolution Techniques for Video Analytic Applications
Abstract: In this talk, we will present and discuss video analytics researches focused on super resolution techniques conducted at video technology research group, Chulalongkorn University. Super-resolution (SR) is an image, video, and computer vision task that reconstruct the high quality or high-resolution (HR) image with large texture detail information from a single or multiple low quality or low-resolution (LR) image, under the limited conditional environment and low-cost imaging system. Despite its difficulty and limitations, SR could be applied in real world applications, such as security and surveillance imaging systems, face recognition, and medical and satellite imaging systems. This talk will discuss the evolution of SR from interpolation techniques to deep convolutional neural networks (CNNs) and will present some of the recent research works on multi-scale inception-based super-resolution (SR).
Prof. Wen-Huang Cheng
1. Title: Practical Deep Learning for Consumer Electronics Applications
The application of artificial intelligence (AI) into the consumer electronics (CE) industry has opened up new opportunities and growth avenues. Deep learning is now the dominant school in the study of AI, given the impact of deep learning on applications that learn from data rather than through explicit programming. However, deep learning in its current state has significant limits for developing CE applications. For example, deep learning easily suffers from imbalanced training data, i.e. there is a disproportionate ratio of observations in each data class. Most collectable data in CE domains tend to have imbalanced distributions because certain phenomena seldom happen, such as defect data from a manufacturing process (i.e. today’s defect rate of the CE production is relatively low). Other disadvantages of deep learning for developing CE applications include reliance on labeled data, reliance on large volumes of data, lack of explainability and lack of methods for integrating prior knowledge. Therefore, this talk aims to give an overview of these challenges, along with promising methods and best practices to address them. Examples of these methods include using self-supervision for utilizing unlabeled data; using weak supervision for learning with cheap and noisy data labels; using re-balancing strategies for learning from highly imbalanced data. Meanwhile, we will discuss some open problems and research directions which holds great potential for driving the future of CE technology.
2. Title: Intelligent Retail Meets Computer Vision
Intelligent retail has become one of the largest segments in the consumer electronics industry. For example, the global fashion apparel market alone has surpassed 3 trillion US dollars today, and accounts for nearly 2 percent of the world’s Gross Domestic Product (GDP). The augmented shopping experience, mainly conveyed by vision as a way for integrating the multichannel experience, has thus attracted much attention from computer vision researchers in recent years. However, many technical challenges remain to be addressed. For example, virtual try-on of clothes in intelligent retail is a fashionable technology for the consumer to virtually try a desired outfit but clothing is difficult to render with visually realistic results due to the nature it deforms and reflects light in folds and crevices. Given the rapid development, this talk aims to provide a comprehensive overview covering four main aspects for enabling intelligent retail with a focus on augmented shopping experience: (1) Consumer fashion detection includes landmark detection, fashion parsing, and item retrieval, (2) Consumer fashion analysis contains attribute recognition, style learning, and popularity prediction, (3) Consumer fashion synthesis involves style transfer, pose transformation, and physical simulation, and (4) Consumer fashion recommendation comprises fashion compatibility, outfit matching, and hairstyle suggestion. Meanwhile, this talk shares our experiences in working on innovative artificial intelligence solutions to overcome key technical challenges and turn the innovations into practical consumer electronics applications for intelligent retail. Also, promising directions for future research will be highlighted.
3. Title: Towards Facial Behavior Understanding for Smarter Consumer Technology
Facial behavior (e.g., facial expressions) is a type of nonverbal communication, which is a result of conscious suppression (intentional) or unconscious repression (unintentional) and can be viewed as a “leakage” of people’s true emotional states or feelings. In recent years, facial behavior understanding has drawn much attention because it can benefit a wide range of ‘smart’ applications through insightful understanding of the user, e.g., clinical diagnosis, depression analysis, customer understanding, and even business negotiation and police interrogation. In facial behaviors, micro-expressions are challenging to be analyzed due to the short span of time (0.5 seconds or shorter) and the fine-grained changes (low-intensity movements of the face). It is also difficult to collect and label micro-expression data. Even for human experts, it takes 2 hours to label a 1-minute video of facial micro-expressions on average. Therefore, this talk aims to present state-of-the-art methods in facial behavior understanding, with a focus on the micro-expression aspect. A systematic review is arranged to go through the typical algorithmic pipeline of a facial behavior understanding system, from the basic technical components of facial data acquisition, facial data preprocessing, facial feature extraction, behavior learning methods and applications. Finally, we will discuss some open problems in facial behavior understanding and the emerging trends to drive next level of innovation in consumer technology through facial behavior understanding capabilities.
4. Title: Autonomous Driving: Human-Like Understanding for Traffic Scenes
Autonomous driving has become one of top technology trends in consumer electronics. Particularly, autonomous vehicles are cars that have the capability to perceive the environment, locate its position and safely drive to the destination without any human intervention. This field has amazing improvement because of the advanced technologies and progress of artificial intelligence (AI) fields. This talk aims to present a novel system flow for empowering autonomous vehicles to understand the traffic scene and summarizes state-of-the-art research, e.g., common sense reasoning. Meanwhile, this walk will share our own experience in developing human-like understanding technologies for self-driving cars. For example, trajectory prediction is the capability of forecasting motion intent and future behavior of various on-road agents such as cars, buses, pedestrians, and animals. It can efficiently facilitate self-driving applications, e.g., an accurate trajectory prediction guarantees safe navigation for autonomous driving. However, many factors should be considered in heterogeneous traffic environments, i.e., consisting of multiple types of on-road agents, social interactions and terrains. Therefore, predicting accurate trajectories in the heterogeneous environment remains challenging. We will introduce the newly proposed system for tackling the trajectory prediction problems, called Attended Ecology Embedding-based Generative Adversarial Networks (AEE-GAN), where two enforced attention modules are developed to socially and visually attend the important information from ecology. Lastly, we will discuss some open problems and how the research community will drive the next revolution in consumer technology.
Dr Brij B GUPTA
1. Title: Social media security and trustworthiness
Abstract: The emerging social media with inherent capabilities seems to be gaining edge over comprehensiveness, diversity and wisdom, nevertheless its security and trustworthiness issues have also become increasingly serious, which need to be addressed urgently. Many researchers mainly aim at both social media content and user security, including model, protocol, mechanism and algorithm. Unfortunately, there is a lack of investigating on effective and efficient evaluations and measurements for security and trustworthiness of various social media tools, platforms and applications, thus has effect on their further improvement and evolution. To address the challenge, a closure on the state-of-the-art of social media networks security and trustworthiness particularly for the increasingly growing sophistication and variety of attacks as well as related intelligence applications are required and then a new direction on evaluating and measuring those fundamental and underlying platforms for crowd evaluations based on signaling theory and crowd computing, which is essential for social media ecosystem.
2. Title: Efficient IoT-based sensor BIG Data collection, processing and analysis
Abstract: Internet of Things (IoT) provides to everyone new types of services in order to improve everyday life. Along with the IoT, the other developed technologies such as Big Data, Cloud Computing, and Monitoring could take part in order to find out their common operations, and combine their functionality, in order to have beneficial scenarios of their use. For the implementation of smart cities applications, investigation of new systems for collecting and managing sensors’ data in a smart building which operates in IoT environment is required. As a bases technology for the sensor management system, a cloud server would be used, collecting the data that produced from each sensor in the smart building. These data are easy to be managed and controlled from distance, by a remote (mobile) device operating on a network set up in IoT technology. As a result, the solutions for collecting and managing sensors’ data in a smart building could lead us in an energy efficient smart building, and thus in a Green Smart Building.
3. Title: Surveillance System in IoT Smart City Framework
Abstract: Internet of Things (IoT) is the new technological revolution that aspires to connect all the everyday physical objects to the Internet, making a huge global network of uniquely things which can share information amongst each other and complete scheduled tasks, bringing significant benefits to users and companies of a Smart City . A Smart City represents a new future framework, which integrates multiple information and communication technology and Internet of Things (IoT) solutions, so as to improve the quality life of its citizens. However, there are many security and privacy issues which must be taken into account before the official launching of this new technological concept. The IoT network architecture and its security challenges and analyze the most important researches on media security and privacy in wireless sensor networks.
4. Title: Network Forensics and Analysis Techniques
Abstract: Cyber forensics, also known as computer forensics, which is a subdivision of digital forensic science, relating to evidence detection in computers and digital storage media. The purpose of cyber forensics is the forensically-sound investigation of digital media with the intent to: identify, preserve, recover, analyze, present facts, and opinions; concerning the digital information. Even though it is generally allied with the analysis of cyber-based crimes, computer forensics may also be used in civil proceedings. Evidence composed from cyber forensic analysis is typically subjected to similar procedures and performs as supplementary digital evidence. With these advancements, it was desired that cyber forensics be to protect users and remain citizen-centric.
Dr. Fumitaka Ono
1. Title: History of Facsimile in Office and Home
Abstract: The history of office facsimile and home facsimile started when G3 standard was defined. The lecture introduces how the technical issues were solved in office-use Facsimile and what kind of efforts were done to prevail facsimile into home.
2. Title: History of Still Image Coding
Abstract: Still image coding is composed of bi-level image coding and color image coding. In the lecture, the overviews of these coding schemes will be given together with the international standards being developed.
3. Title: Arithmetic coding : The essence and its design parameters
Abstract: The arithmetic coding has been known as a theoretical entropy coding method. Its practical importance was found first in the application to Markov-model bi-level image coding. The essence of the arithmetic coding will be made clear why it is now used even in video coding.
1.Title: Technologies for Electronics Size Reduction
Abstract: As the size of electronic products gets smaller, non-conventional approaches are required to reduce size. These new approaches include bare die assembly, device stacking, embedded components and digital implementations of traditional analog functions.
2. Title: Dynamic Voltage and Frequency Scaling (DVFS)
Abstract: DVFS was first introduced as a means of idle power reduction for battery powered computers. DVFS is now required for all cellphones to have reasonable battery life. As functional complexities increase in mobile devices DVFS becomes more complex to implement.
3. Title: DDR/GDDR/HBM/NOR/NAND/SRAM Memory Alphabet Soup
Abstract: There are so many different memory technologies, which ones are intended for what applications?
1. Title: Quantum Computing: What, Why, Who.
Quantum Computing (QC) has been a theoretical promise since the beginning of 1990`s. A lot of research effort has been invested, especially in two areas. First, on the mathematics, logics and algorithms area. Second, quantum physicist and materials experts have been working on how to implement such a machine.
Now, there are a few quantum computers available online through different providers. Although present machines still have low computing power, the industry is very optimistic about increasing it in a sustained rate during the following years. So, the promise of real software applications solving daily problems are close to come. That is why the field now is attractive for software companies and startups.
There is a lot of public activities, either academic, commercial, and governmental, concerning QC, and the field is gaining much interest and investments.
The lecture provides an introduction about QC, presents the basic aspects of the logic that sustains QC, enumerates the most known quantum algorithms, and shows the fields of application. Finally, presents the status of the industry, reviewing the players that are involved.
2. Title: Quality of Experience: Subjective and objective video quality assessment
Video is a fundamental component of the multimedia content that is delivered to thousands of millions of users worldwide through many different networks and operators, such as broadcasters, ISPs or mobile operators.
It is vital for communication services to assess the quality of the service provided to the user. Quality of Service (QoS) is a widely extended concept that looks after the importance of registering key parameters of the service to assure its compliance. Beyond QoS, Quality of Experience (QoE) focuses on the user experience of the service and on modelling the impact that service parameters have on it. When video is involved in a service, perceived video quality is a major component of the whole QoE.
The video quality perceived by the users is affected by two processes: video coding and video transmission. These processes are independent and degrade the original quality of a video clip in different ways.
In this context, it is important to develop models for the assessment of the video quality perceived by the consumer.
The lecture analyzes the background of the procedures to assess video quality. It outlines the methodology to perform valid subjective tests, as well as the basic aspects of objective methods.
2021 Distinguished Speakers
Dr. Haruhiko Okumura
Prof. Muhammad Khurram Khan
Dr. Sreeram (Ram) Dhurjaty
Prof. Roberto Minerva