THE TALKS

Speaker: Prof. Trần Chiêu Minh 

Title: Growth in groups: direct and inverse problem

Abstract: Let G be a group equipped with a notion of size (cardinality, Haar measure, etc), and let  A be a subset of G. We are interested in the following problems: 

(i)                  Find inequalities relating the size of the product set A^k ={a_1  a_k: a_1, a_2 \in A} and the size of A. 

(ii)                 Understand when the size of  A^k is not too large compared to that of A. 

These are called the direct and inverse problems for growth in groups. In this talk, I will explain how these problems arise in analysis, group theory and additive combinatorics. I will also discuss a number of recent progresses in addressing them including our solution of a conjecture by Emmanuel Breuillard and Ben Green.

Speaker: Dr. Phạm Minh Phương

Title: Quantum Attacks Against Symmetric-key Cryptosystems

Abstract: Contemporary cryptography plays an essential role in the security of digital communication, internet, and data storage. The level of security of a cryptosystem is measured by the computational resources needed to break it. Traditional cryptanalysis focused on analyzing and breaking cryptographic systems, with the aim of uncovering their weaknesses and vulnerabilities using classical computers. Contemporary cryptography plays an essential role in the security of digital communication, internet, and data storage. The level of security of a cryptosystem is measured by the computational resources needed to break it. Traditional cryptanalysis focused on analyzing and breaking cryptographic systems, with the aim of uncovering their weaknesses and vulnerabilities using classical computers. However, the emergence of quantum computing introduces a need to revise these security levels, considering a completely new set of algorithms.

The main objective of this talk is to explore the security aspects of block ciphers and hash functions, considering both traditional and quantum computing environments. Through the utilization of automated tools and quantum techniques, various attacks, such as distinguishers and collision attacks, were employed to evaluate the security levels of these fundamental cryptographic components.

Speaker: Prof. Mai Anh Tiến

Title: Imitating Opponent to Win: Imitation Learning Strategies for Success in in Two-player Competitive Games

Abstract:  This talk is about the problem of training an AI agent to win in a two-player competitive game against a well-trained opponent, which has important implications in enhancing decision-making and game theory. I will present our innovative idea of using imitation learning to mimic the opponent, providing useful information to achieve higher winning rates. Specifically, the core idea of our new algorithm is to create a new imitator — the imitator will learn to imitate the opponent’s policy while the training policy will be updated not only based on interactions with the opponent agent but also based on feedback from the imitator to forecast the opponent’s intention. By doing so, we can leverage the capability of imitation learning to well capture underlying characteristics of the opponent's policy based solely on sample trajectories of the opponent. Our extensive experiments using some benchmark competitive game environments show that our proposed policy learning algorithm outperforms state-of-the-art algorithms.

Speaker: Prof. Nguyễn Minh Tân

Title: Transformers Meet Image Denoising: Mitigating Over-smoothing in Transformers via Regularized Nonlocal Functionals

Abstract: Transformers have achieved remarkable success in a wide range of natural language processing and computer vision applications. However, the representation capacity of a deep transformer model is degraded due to the over-smoothing issue in which the token representations become identical when the model’s depth grows. In this work, we show that self-attention layers in transformers minimize a functional which promotes smoothness, thereby causing token uniformity. We then propose a novel regularizer that penalizes the norm of the difference between the smooth output tokens from self-attention and the input tokens to preserve the fidelity of the tokens. Minimizing the resulting regularized energy functional, we derive the Neural Transformer with a Regularized Nonlocal Functional (NeuTRENO), a novel class of transformer models that can mitigate the over-smoothing issue. We empirically demonstrate the advantages of NeuTRENO over the baseline transformers and state-of-the-art methods in reducing the over-smoothing of token representations on various practical tasks, including object classification, image segmentation, and language modeling.