Location

The lectures and poster sessions will happen in Graduate House at the University of Melbourne. Check the table below for the details.

Address: 220 Leicester St, Carlton VIC 3053

Abstracts:  

Matthieu Bloch

Georgia Institute of Technology

Title: An information-theoretic perspective on covert communications 

Abstract: In our connected world where ubiquitous networks underpin critical infrastructure and services, securing information is always a critical concern. Beyond classical objectives such as confidentiality, message authentication, and data integrity, an intriguing objective is the ability to hide the very existence of a message, which we call covert communication.

This tutorial explores covert communication through the framework of information theory, offering a unified perspective that connects diverse disciplines including signal processing, communication theory, and cryptography. Covert communication appears in many real-world systems, such as frequency-hopping spread spectrum, code-division multiple access (CDMA), and steganographic methods. What unites these methods is the need to operate below the detection threshold of a warden or adversary. We will investigate how classical and modern information-theoretic tools reveal the limits and capabilities of covert communication. In particular, we will:

1. Examine key metrics like relative entropy and total variation distance, exploring their nuanced differences in measuring statistical distinguishability.

2. Introduce random coding arguments and their role in proof-of-concept coding theorems, which are central to analyzing covert channels.

3. Explore hypothesis testing as a core methodology for modeling the adversary’s ability to detect transmissions.

4. Highlight supporting concepts such as concentration of measure phenomena, converse arguments, and achievability bounds that are critical in information-theoretic reasoning.

Mark  Wilde [PDF]

Cornell University

Title: Quantum Information Theory

Abstract: What are the ultimate limits that nature imposes on communication and what are effective procedures for achieving these limits? In order to answer these questions convincingly, we must reassess the theory of information under a quantum lens. That is, since quantum mechanics represents our best understanding of microscopic physical phenomena and since information is ultimately encoded into a phys-

ical system of some form, it is necessary for us to revise the laws of information established many years ago by Shannon. This is not merely an academic exercise, but instead represents one of the most exciting new frontiers for physics, mathematics, computer science, and engineering. Entanglement, superposition, and interference are all aspects of quantum theory that were once regarded as strange and in some cases,

nuisances. However, nowadays, we understand these phenomena to be features that are the enabling fuel for a new quantum theory of information, in which seemingly magical possibilities such as teleportation are becoming reality. Two other notable examples are increased communication capacities of noisy communication channels and secure encryption based on physical principles. Concepts developed in the context of quantum information theory are now influencing other areas of physics as well, such as quantum gravity, condensed matter, and thermodynamics. Furthermore, quantum information theory has given us a greater understanding of the foundations of quantum mechanics and might eventually lead to a simpler set of postulates for quantum mechanics.

This tutorial will review the basics of quantum information. An outline is as follows:

1. background on quantum information and connections to classical information, including density op-

erators, channels, measurements, purification, isometric extension, coherent measurement;

2. noiseless protocols of entanglement distribution, teleportation, superdense coding;

3. distance measures including trace distance and fidelity. Uhlmann’s theorem and gentle measurement;

4. quantum entropy and entropy inequalities;

5. protocols including Schumacher compression, classical communication, entanglement-assisted classi-

cal communication, quantum communication. Nonadditivity of capacities and superactivation.

Peter Bartlett [PDF]

University of California, Berkeley

Title: Topics in statistical learning theory 

Abstract: These lectures will review tools for the analysis of the generalization performance of prediction methods on classification and regression problems: uniform convergence properties and associated notions of complexity, such as Rademacher averages and combinatorial dimensions, uniform convergence results for neural networks, and benign overfitting and performance guarantees for interpolating prediction rules.

Yuting Wei  [PDF]

University of Pennsylvania

Title: Mathematical Foundations of Reinforcement Learning 

Abstract: As a paradigm for sequential decision making in unknown environments, reinforcement learning (RL) has received a flurry of attention in recent years. However, the explosion of model complexity in emerging applications and the presence of nonconvexity exacerbate the challenge of achieving efficient RL in sample-starved situations, where data collection is expensive, time-consuming, or even high-stakes (e.g., in clinical trials, autonomous systems, and online advertising). How to understand and enhance the sample and computational efficacies of RL algorithms is thus of great interest and in imminent need. In this tutorial, we aim to present a coherent framework that covers important algorithmic and theoretical developments in RL, highlighting the connections between new ideas and classical topics. Employing Markov Decision Processes as the central mathematical model, we introduce several distinctive RL scenarios (i.e., RL with a simulator, online RL, offline RL, robust RL, and RL with human feedback), and present several mainstream RL approaches (i.e., model-based approach, value-based approach, and policy optimization). Our discussions gravitate around the issues of sample complexity, computational efficiency, as well as algorithm-dependent and information-theoretic lower bounds in the non-asymptotic regime.

Emanuele Viterbo, Brendon McBain

Monash University

Title: From Molecules to Zettabytes: Information Storage in Synthetic DNA

Abstract: DNA has emerged as a promising medium for digital information storage because it is compact, durable, and capable of preserving data for centuries. As the global volume of digital information continues to expand, DNA offers a long-term solution to safeguarding the knowledge of today for future generations. 

This tutorial introduces the fundamental principles of storing and retrieving information in synthetic DNA, bridging concepts from molecular biology, computer science, and information theory. We begin with an overview of how digital data can be encoded into DNA molecules and then recovered through sequencing technologies. Particular emphasis is placed on nanopore sequencing, a rapidly evolving platform that reads DNA directly from ionic current signals and presents rich information-theoretic challenges. 

Participants will gain an intuitive understanding of the complete DNA storage pipeline, from synthesis to sequencing, and learn how coding enhances its reliability and how information theory describes its fundamental limits. The session will also highlight ongoing research directions and open problems in this field, including channel modelling, inference, and coding.

Jean Honorio

The University of Melbourne 

Title:  Limits and Possibilities of Machine Learning and Optimisation: Two Stories of Information (1-hour talk)

Abstract: Information theory plays an increasingly important role in machine learning (ML) and optimisation, from the study of fundamental limits of learning and convergence, to the understanding of generalisation of particular algorithms, thus settling important questions of statistical and computational complexity of various problems. In this talk, I provide two stories of information.

In the first story, I discuss how information characterises fundamental statistical and computational limits, with two different mathematical objects largely studied in ML and optimisation: graphs (either recovering properties of graphs or learning graphs) and functions (optimisation, integration and gradient estimation). Recovering properties of graphs is relevant in social network analysis, network science and inference in probabilistic graphical models (PGMs), while learning graphs relates to the problem of structure learning of PGMs. Here information allows to characterise whether current computationally efficient algorithms are statistically optimal. While integration is ubiquitous in ML, gradient estimation is important for zero-order (function access only) optimisation. The analysis of optimisation allows to show that first-order (gradient access only) methods require exponential time to converge to global minimisers of non-convex functions. Information also allows to show that first-order methods have a logarithmic dependence on dimension to avoid saddle points in non-convex optimisation.

In the second story, I discuss the role of information in generalisation bounds, through various change of measure inequalities. This view allows to generalise the PAC-Bayes framework to different divergences and assumptions on losses. Information also allows to find generalisation bounds by using Karush-Kuhn-Tucker conditions, leading to tighter generalisation bounds for deep neural networks.

Wanchun Liu 

University of Sydney

Title: When Control Meets the Airwaves: Stability of Networked Systems over Fading Channels (1-hour talk)

Abstract: In modern control systems, sensing and control signals are increasingly transmitted over wireless networks rather than fixed cables. While this flexibility enables large-scale and distributed systems such as autonomous vehicles and industrial automation, it also introduces a key challenge: wireless channels fluctuate over time, and these variations can fundamentally affect the stability of the control loop. This talk examines the theoretical foundation of stability in remote state estimation when both the plant dynamics and the wireless channel evolve simultaneously. The channel is modelled as a Markov fading process, capturing how previous channel conditions influence future communication reliability. By analysing the joint evolution of the system and channel states, the work derives closed-form conditions that determine whether estimation errors remain bounded or diverge. These results provide a rigorous characterisation of the interplay between control dynamics and stochastic communication behaviour, offering deeper insights into the stability limits of networked control systems operating under uncertain and time-varying wireless environments.

James Saunderson 

Monash University

Title: Optimising quantum relative entropies (1-hour talk)

Abstract: Relative entropy-like quantities, such as the KL divergence and more general f-divergences, appear naturally in (classical) information theory. Various quantum analogues of these quantities play an important role in quantum information theory. In this talk I'll discuss convexity properties of some of these functions, as well as discuss certain key ideas behind numerical algorithms for solving optimisation problems involving classical and quantum relative entropies.