Tutorial 1: Towards Communication-Efficient and Heterogeneity-Robust Federated Learning

Time: Nov 28th, 14:00 - 16:00
Presenter: Guodong Long and Yue Tan

Federated learning has emerged as a promising approach to machine learning where data is kept in a decentralized manner, thereby minimizing privacy and security risks. However, the decentralized nature of federated learning introduces two major challenges: ensuring communication efficiency and robustness to data heterogeneity across different data silos or devices. In response to these challenges, we introduce 'Towards Communication-Efficient and Heterogeneity-Robust Federated Learning', a comprehensive guide that uncovers the intricacies of these challenges and provides actionable insights on how to address them. This tutorial aims to review the recent advancement in federated learning with an emphasis on establishing communication-efficient and heterogeneity-robust federated learning paradigms. The content can be summarized into the following five parts. Firstly, the introduction and background are delivered to provide a comprehensive review of federated learning. Secondly, we discuss the emerging trends that aim to improve communication efficiency in FL. Thirdly, we present existing efforts in FL combating data heterogeneity problems. This part can be further categorized into three aspects, i.e., clustering-based FL, prototype-based FL, and graph-structured FL. Fourthly, diverse applications of FL with non-IID data are introduced. Lastly, we finalize the tutorial with discussions on promising future directions as well as the conclusion.

Tutorial 2: Reinforcement Learning for Automated Negotiation Supply Chain Management League as an Example

Time: Nov 28th, 14:00 - 16:00
Presenter: Yasser F. O. Mohammad

Automated negotiation between intelligent agents is attracting more attention from the research community especially with the wider market penetration of intelligent agents and the need to coordinate their behavior. The International Automated Negotiating Agents Competition (ANAC) provided stimulation for this research since its introduction in 2010. Since 2019, a new league was added to ANAC focusing on application of automated negotiation in a realistic business-like Supply Chain Management scenario (SCML). In SCML 2024, a new RL based version of the problem is to be tackled. This tutorial will introduce the audience to SCML, and its RL formulation and walk them through the development of an agent for the competition highlighting the research challenges.

Why is it interesting? Reinforcement learning is shown to be effective in solving several types of games. In this tutorial, we show that automated negotiation is a game with general-sum incomplete information which has the special feature that the uncertainty is in the payoffs themselves which makes it an appropriate next step for RL and MARL technologies to target (e.f. after Stratego [4]) SCML provides a nice introduction into this problem.

What will the audience walk away with? The audience will walk with a firm grasp on the research problems involved in SCM league and will have hands-on experience in developing a basic agent for it. Moreover, they will be introduced to the RL and MARL formulations of the problem and learn about the general framework for developing solutions within the NegMAS library used in the competition.

Tutorial 3: Decoding the grammar of DNA using Natural Language Processing

Time: Nov 28th, 14:00 - 16:00
Presenter: Tyrone Chen, Sonika Tyagi

DNA is the blueprint defining all living organisms. Therefore, understanding the nature and function of DNA is at the core of all biological studies. Rapid advances in DNA sequencing and computing technologies over the past few decades resulted in large quantities of DNA generated for diverse experiments, exceeding the growth of all major social media platforms and astronomy data combined. However, biological data is both complex and high-dimensional, and is difficult to analyse with conventional methods.

Machine learning is naturally well suited to problems with a large volume of data and complexity. In particular, applying Natural Language Processing to the genome is intuitive, since DNA is a natural language. Unique challenges exist in Genome-NLP over natural languages, including the difficulty of word segmentation or corpus comparison.

To tackle these challenges, we developed the first automated and open-source genomeNLP workflow that enables efficient and accurate knowledge extraction on biological data [1], automating and abstracting preprocessing steps unique to biology. This lowers the barrier to perform knowledge extraction by both machine learning practitioners and computational biologists. In this tutorial, we will demonstrate how our workflow can be used to address the above challenges, with implications in fields such as personalised medicine.

AJCAI 2023 - Australasian Joint Conference on Artificial Intelligence 2023, Brisbane, Australia