Thursday Morning, 25 MAY 2023

• T1: Moving Beyond Traditional Anomaly Detection (9:00-12:30)
• T2: A Gentle Introduction to Technologies Behind Language Models and Recent Achievement in ChatGPT (9:00–11:00)
• T5: Multi-Aspect Learning – Issues, Algorithms and Applications (9:00–12:30)

Thursday Afternoon, 25 MAY 2023

• T3: Match Video Annotation for Tactical Analytics of Badminton (13:30–16:30)
• T4: Group Anomaly Detection (13:30–16:30) Canceled

T1: Moving Beyond Traditional Anomaly Detection (9:00–12:30)

Speakers:

Ye Zhu (Deakin University), Gang Li (Deakin University), Chen Li (Nagoya University), Yang Cao (Deakin University)

Abstract:

Anomaly detection is a significant task of data mining, and also a hot research topic in various fields of artificial intelligence in recent years. It has a wide range of applications, such as extreme climate event detection, mechanical fault detection, terrorist detection, fraud detection, malicious URL detection etc. This tutorial aims to present a comprehensive review of both shallow and deep learning-based anomaly detection and explanation. We first introduce the key intuitions, objective functions, underlying assumptions, and advantages and disadvantages of state-of-the-art anomaly detection methods. We also introduce several principled approaches used to provide anomaly explanations for deep detection models. Furthermore, we will discuss the connections between classic shallow and novel deep methods and provide a practical guide on how to select an outlier detector in different applications.

Tutorial Notes:

PDF

T2: A Gentle Introduction to Technologies Behind Language Models and Recent Achievement in ChatGPT (9:00–11:00)

Speakers:

Jun Suzuki (Tohoku University), Naoaki Okazaki (Tokyo Institute of Technology), Kyosuke Nishida (NTT Corporation)

Abstract:

Language models (LMs) have a long history in natural language processing (NLP) research. Their usage was mainly a text generation module in machine translation and speech recognition systems, used together with translation models or acoustic models. After the current neural era, LMs take a more essential role in the NLP field. In fact, LMs are integrated into any models/systems to tackle almost all the NLP tasks and provide state-of-the-art performance on conventional benchmarks. The usage of LMs is considered to be shifting to more like a world model of languages or a general-purpose feature generator of any language-related tasks. More recently, the public sometimes treats LMs like ChatGPT, and its successor GPT-4, as general-purpose AI after starting an online service in the public domain. This tutorial will first introduce some introductory topics we should know when discussing the recent advances in LMs like ChatGPT. We will then briefly introduce the technologies behind ChatGPT-like LMs. Additionally, we also provide ChatGPT’s social impacts discussed recently in public.

Tutorial Notes:

Part 1 & 2

Part 3 & 4

Part 5

T5: Multi-Aspect Learning – Issues, Algorithms and Applications (9:00-12:30)

Speakers:

Richi Nayak (Queensland University of Technology), Md Abul Bashar (Queensland University of Technology), Duoyi Zhang (Queensland University of Technology)

Abstract:

Multi-aspect data, which consists of information from multiple perspectives or modalities, has become increasingly prevalent and significant in various domains. The main advantage of multi-aspect data is its ability to capture rich and diverse information that can facilitate machine learning algorithms to discover multiple patterns and types inherent in the data and produce informative outcomes for different learning tasks, whether supervised or unsupervised. However, multi-aspect data also poses several challenges to machine learning due to its natural complexity and heterogeneity. Therefore, it is essential to pay meticulous attention to the characteristics and properties of multi-aspect data and develop appropriate methods and techniques for effective machine learning on this type of data. This tutorial aims to provide a comprehensive overview of an emerging research topic in machine learning, namely multi-aspect learning. We first introduce the different types and characteristics of multi-aspect data and discuss the challenges and opportunities that they present for machine learning tasks. We then review the state-of-the-art machine learning methods that can effectively exploit the underlying structures and relationships in multi-aspect data and achieve improved performance over traditional methods that ignore the multi-aspect nature of the data. Finally, we highlight current research gaps and open questions in this field and suggest possible directions for future research.

Tutorial Notes:

PDF

T3: Match Video Annotation for Tactical Analytics of Badminton (13:30-16:30)

Speakers:

Tsì-Uí İk (National Yang Ming Chiao Tung University), Yung-Chang Huang (National Yang Ming Chiao Tung University)

Abstract:

This proposal is to apply to organize a half-day 3-hours tutorial on video annotation of badminton match shot-by-shot data collection at PAKDD 2023, Osaka, Japan. This tutorial is composed of three parts, including camera geometry, object detection and tracking, and shot-by-shot labeling. The technology introduced in this tutorial can help researchers understand the imaging principles in computer vision, widely used object detection, and brand-new microscopic skill data labeling system for badminton. This tutorial not only helps people enter the study of badminton tactical analysis but is also a good class for sports science and sports image analysis.

Tutorial Notes:

PDF