The research of the Information Theory and Applications (TIA) team is primarily concerned in developing new information theory and statistical models applied in various scientific and technological domains:
Wireless communications: Channel estimation theories enable high bandwidth an increased spectral efficiency at the physical layer in mobile communications using models based on channel state information (CSI) and multiple-input-multiple-output (MIMO) over time-frequency wireless channels used for 5G and the internet of things (IOT).
Optical fiber telecommunications: Fiber optics high bandwidth and low attenuation make it ideal for multi-gigabit telecommunications. Future optical networks seek to process coherent information directly in the optical domain for optical amplification wavelength multiplexing/demultiplexing (WDM), filtering, switching, and correlation processing.
Multi-user multi-space optical channels: There is growing interest in optical wireless communications (OWC) such as light fidelity (LiFi), visible-light communications (VLC) used e.g. for autonomous vehicle patrols, laser free-space optics (FSO), underwater optical communications and satellite optical communications.
Channel modeling and characterization of fundamental limits: Important research directions propose new fundamental results in communication theory that address the interaction among multiple services having the potential to change communication system architectures as well as the regulation of the spectrum and communication services.
Coding and signal processing for telecommunications: Error correcting codes greatly improve the bandwidth-power efficiency and reliability of communication systems. LDPC codes and polar codes are adopted in 5G. Research in signal processing for communications focuses the on analysis and design multi-rate systems, beam-forming, array processing and smart-antennas.
Machine learning methods for communications: Applications of machine learning (ML) are currently used in various aspects of communications and networking such as network planning and performance prediction, cross-layer network optimizations for software-defined networks, and autonomous and reliable network operations.
Bridges between machine learning and information theory in deep learning neural networks: Information theory has made significant contribution to deep learning and artificial intelligence (AI). The derived tools are for example cross-entropy loss function, maximum information gain decision trees, Viterbi algorithm widely used in natural language processing (NLP) and speech recognition.
Web data analysis: Data analytics has emerged as a critical technology to deal with explosively growing amounts of data, Problems in data analysis are concerned with statistical learning and use information-theoretic tools based on linear algebra and graph theory.
Anonymization and reliability assessment: Information about a user’s sensitive attributes and user web-connection information must be protected. Classical anonymization techniques are subject to attacks exploiting structural information on anonymized graphs. Graph edge modification techniques can provide structural similarity improving anoymization.
Quantum information approaches for semantic web and robotics: Models are investigated for Information Retrieval (IR) and for data analysis using vector and operator representations using quantum interference and entanglement. Derived quantum probabilistic decision making models applied to the concept of quantum robot agents show non-classical emergent behaviors.
Information models for statistical mean-field analysis, learning and inference: Models based on mean field Ising approximations are tools used for inference in high-dimensional stochastic systems. They are particularly adapted for the study of perceptrons representing systems such as: neural networks, deep learning machines, constraint satisfaction problems, spin glasses…
Dynamic Markov models: Dynamic Markov models generalize random walk models by representing the states of a system by a set of variables, depend on actual and previous states. These methods have applications in web mining, computational biology, speech recognition, NLP, robotics, and fault diagnosis.
Analysis of spectral graphs and information decoding: Spectral graph theories use probability and Random Matrix theory to tackle key problems in complex networks and big data. These approaches seek to learn representations that encode structural information about the graph using summary graph statistics and kernel functions.
New logical and information approaches with applications to quantum computing: Inspired by recent quantum learning models using density matrix operators, logic is expressed in an operator matrix language. A central role is played by truth-table logical semantics as a major tool for adapting logic to physics permitting to interpret quantum computing logical gate universality.