Bounding the Figures of Merit of an Adversary in the Presence of Side Channel Leakage

We show how to leverage Gibbs' inequality to obtain tight bounds on the figures of merit of a side channel adversary in tems of the evaluated side channel leakage (as measured by mutual information or, more generally, by Sibson's alpha information) and of the number of measurements given to the adversary.

A Geometric Treatment of Lossy Compression of Log-Concave Random Variables

We give a geometric proof for the rate-distortion function of log-concave random variables with symmetric continuous log-concave densities with their matched distortion measures. This generalizes the well-known geometric proof for the Gaussian-quadratic rate-distortion function.

Universal Decoding with the KT Estimator

We study the problem of universal decoding over memoryless channels with a decoder based on the Krichevsky-Trofimov estimator. We show that, despite being ignorant of the channel in use, this decoder asymptotically achieves the same random-coding error exponent as the optimal maximum-likelihood decoder for that channel. Numerical results illustrate its performance with practical linear and convolutional codes.

Information-Theoretical Analysis and Practical Coding Scheme for Distributed Regression under Communication Constraints

This talk will take place exceptionally at 10:00 in room 3C02!

Goal-oriented communication is an emerging field at the intersection of information theory and machine learning, focusing on addressing specific tasks under communication constraints. This work extends classical rate-distortion theory to address regression problems in a distributed setup, where side information is available only at the decoder. Specifically, we generalize the Wyner-Ziv framework to analyze parametric and non-parametric regression under joint constraints on communication rate and reconstruction accuracy.

We introduce the generalization error as a key performance metric; derive achievable rate-generalization error regions; and examine the fundamental trade-offs between reconstruction and regression in both asymptotic and non-asymptotic regimes. Finite block-length analysis, usually used for the rate-distortion problem, is pushed further to derive non-asymptotic bounds and assess the trade-off between data reconstruction and regression error; our theoretical findings are supported by numerical simulations.

Our work provides a theoretical foundation for designing communication-efficient learning systems, with practical implications for regression in distributed setups. The results demonstrate that carefully designed coding schemes can achieve both efficient data compression and optimal learning performance, offering insights for future research into complex distributed systems.

Distributed Hypothesis Testing: Improved Random-Binning Exponent

This talk will take place exceptionally at 10:30 in room 3C02!

Consider the problem of distributed binary hypothesis testing with two terminals, where the guess is made at one of them (the “receiver”). We study the exponent in the error probability of the second type. Previously, an achievable exponent was derived by Shimokawa, Han, and Amari using a “quantization and binning” scheme. We propose a simple modification on the receiver’s guessing rule in this scheme to attain a better exponent.

Towards Robust and Resource-Efficient Quantum Communication over Bosonic Channels

The emerging Quantum Internet, envisioned as an entanglement sharing network, aims to enable distant computers, transceivers, and sensors to process information in a coordinated manner. Recent theoretical and experimental findings support the feasibility of these networks and underscore the advantages of quantum operations, ranging from enhanced security to improved sensing capabilities. However, entanglement is expected to remain a high-cost resource for the foreseeable future, necessitating the development of efficient schemes and protocols for entanglement resource allocation.

In this talk, we will present two theoretical findings that contribute to the design and development of a robust and efficient Quantum Internet. Firstly, we will re-examine the issue of entanglement-assisted communications over bosonic channels and present a bound for the performance of Phase Shift Keying (PSK)-modulated Two-Mode Squeezed Vacuum (TMSV) communication. This bound, which is non-asymptotic in the physical parameters of the bosonic channel, offers insights into the threshold where quantum benefits begin to manifest. Secondly, we will revisit the issue of entanglement-assisted covert communication over bosonic channels as an application. We will demonstrate that two-layer coding allows for the advantageous use of entanglement with significantly fewer TMSV pairs than previously reported.