PhD Defence: “Large-scale Automatic Learning of Autonomous Agent Behavior with Structured Deep Reinforcement Learning”, Edward Beeching, amphi, Chappe Building, 3rd of May 2022 at 10:00 AM

 

The defense willtake place on Tuesday, May 3 at 10:00 am, in the amphitheatre of the Telecommunications Department (Claude Chappe building), INSA Lyon, Villeurbanne.

The presentation will be available on Youtube at the following link: https://youtu.be/k7dh-thqbSk

 

Title

Large-scale Automatic Learning of Autonomous Agent Behavior with Structured Deep Reinforcement Learning

 

 

Abstract

Autonomous robotic agents have begun to impact many aspects of our society,with application in automated logistics, autonomous hospital porters, manufacturing and household assistants. The objective of this thesis is to explore Deep Reinforcement Learning approaches to planning and navigation in large and unknown 3D environments. In particular, we focus on tasks that require exploration and memory in simulated environments. An additional requirement is that learned policies should generalize to unseen map instances. Our long-term objective is the transfer of a learned policy to a real-world robotic system. Reinforcement learning algorithms learn by interaction. By acting with the objective of accumulating a task-based reward, an Embodied AI agent must learn to discover relevant semantic cues such as object recognition and obstacle avoidance, if these skills are pertinent to the task at hand. This thesis introduces the field of Structured Deep Reinforcement Learning and then describes 5 contributions that were published during the PhD.

We start by creating a set of challenging memory-based tasks whose performance is benchmarked with an unstructured memory-based agent. We then demonstrate how the incorporation of structure in the form of a learned metric map, differentiable inverse projective geometry and self-attention mechanisms; augments the unstructured agent, improving its performance and allowing us to interpret the agent’s reasoning process.

We then move from complex tasks in visually simple environments, to more challenging environments with photo-realistic observations, extracted from scans of real-world buildings. In this work we demonstrate that augmenting such an agent with a topological map can improve its navigation performance. We achieve this by learning a neural approximation of a classical path planning algorithm, which can be utilized on graphs with uncertain connectivity.

From work undertaken over the course of a 4-month internship at the research and development department of Ubisoft, we demonstrate that structured methods can also be used for navigation and planning in challenging video game environments. Where we couple a lower level neural policy with a classical planning algorithm to improve long-distance planning and navigation performance in vast environments of 1km×1km. We release an open-source version of the environment as a benchmark for navigation in large-scale environments.

Finally, we develop an open-source Deep Reinforcement Learning interface for the Godot Game Engine. Allowing for the construction of complex virtual worlds and the learning of agent behaviors with a suite of state-of-the-art algorithms. We release the tool with a permissive open-source (MIT) license, to aid researchers in their pursuit of complex embodied AI agents.

 

 

Jury

    • Mme. Elisa Fromont – Université de Rennes 1 – Rapporteur
    • M. David Filliat – ENSTA Paris – Rapporteur
    • M. Cédric Démonceaux – Université de Bourgogne – Examinateur
    • M. Karteek Alahari – INRIA Grenoble – Examinateur
    • Mme. Christine Solnon – INSA-Lyon – Examinateur
    • M. Olivier Simonin – INSA-Lyon – Directeur de thèse
    • M. Jilles Dibangoye – INSA-Lyon – Co-encadrant de thèse
    • M. Christian Wolf – Naver Labs Europe – Co-directeur de thèse

PhD Defence: “Applications of Deep Learning to the Design of Enhanced Wireless Communication Systems”, Mathieu Goutay, Showcase room, Chappe Building, 28th of January 2022 at 2:00 PM

This thesis has been done within Nokia Bell Labs France and the Maracas team of the CITI laboratory.

The defense willtake place on Friday, January 28 at 2:00 pm, in the showcase room of the Telecommunications Department (first floor), INSA Lyon, Villeurbanne.

The presentation will be available on Zoom at the following link: https://insa-lyon-fr.zoom.us/j/96199554997

 

Title

Applications of Deep Learning to the Design of Enhanced Wireless Communication Systems

 

 

Abstract

Innovation in the physical layer of communication systems has traditionally been achieved

by breaking down the transceivers into sets of processing blocks, each optimized independently based on mathematical models. This approach is now challenged by the ever-growing demand for wireless connectivity and the increasingly diverse set of devices and use-cases. Conversely, deep learning (DL)-based systems are able to handle increasingly complex tasks for which no tractable models are available. By learning from the data, these systems could be trained to embrace the undesired effects of practical hardware and channels instead of trying to cancel them. This thesis aims at comparing different approaches to unlock the full potential of DL in the physical layer.

First, we describe a neural network (NN)-based block strategy, where an NN is optimized to replace one or multiple block(s) in a communication system. We apply this strategy to introduce a multi-user multiple-input multiple-output (MU-MIMO) detector that builds on top of an existing DL-based architecture. The key motivation is to replace the need for retraining on each new channel realization by a hypernetwork that generates optimized sets of parameters for the underlying DL detector. Second, we detail an end-to-end strategy, in which the transmitter and receiver are modeled as NNs that are jointly trained to maximize an achievable information rate. This approach allows for deeper optimizations, as illustrated with the design of waveforms that achieve high throughputs while satisfying peak-to-average power ratio (PAPR) and adjacent channel leakage ratio (ACLR) constraints. Lastly, we propose a hybrid strategy, where multiple DL components are inserted into a traditional architecture but trained to optimize the end-to-end performance. To demonstrate its benefits, we propose a DL-enhanced MU-MIMO receiver that both enable lower bit error rates (BERs) compared to a conventional receiver and remains scalable to any number of users.

Each approach has its own strengths and shortcomings. While the first one is the easiest to implement, its individual block optimization does not ensure the overall system optimality. On the other hand, systems designed with the second approach are computationally complex and do not comply with current standards, but allow the emergence of new opportunities such as high-dimensional constellations and pilotless transmissions. Finally, even if the block-based architecture of the third approach prevents deeper optimizations, the combined flexibility and end-to-end performance gains motivate its use for short-term practical implementations.

 

 

Jury

    • Reviewer: Didier LE RUYET, Professor, CNAM, Paris, France
    • Reviewer: Charlotte LANGLAIS, Permanent Research Staff, IMT Atlantique, Brest, France
    • Examiner: Inbar FIJALKOW, Professor, ENSEA, Cergy, France
    • Examiner: Stephan TEN BRINK, Professor, University of Stuttgart, Stuttgart, Allemagne
    • Thesis supervisor: Jean-Marie GORCE, Professor, INSA Lyon, Villeurbanne, France
    • Thesis co-supervisor: Jakob HOYDIS, Principal Research Scientist, Nvidia, France*
    • Thesis co-supervisor Fayçal AIT AOUDIA, Senior Research Scientist, Nvidia, France*

* Previously at Nokia Bell Labs France


CITI seminar – Frédéric Prost (UGA) – 27/01 at 14:00

Title: Projet Samildanach – Communication Scientifique dans une Culture Digitale

Date and Place: January 27th 14h, Amphi Chappe

Lien visio : https://insa-lyon-fr.zoom.us/j/97446195543

Speaker: Dr. Frédéric Prost (UGA)

 

Abstract: 

L’objectif de ce séminaire est de présenter un projet relatif à la communication scientifique. L’idée est d’utiliser plusieurs technologies récentes comme la blockchain, les tables de hachage décentralisées ainsi que les calculs de réputation issus des réseaux sociaux pour les appliquer au domaine de la publication scientifique. La blockchain est utilisée pour certifier,présenter une résistance à la censure, assurer la non répudiation et une structure d’incitations pour le  développement du réseau (rémunération des acteurs qui aident le réseau).   Ce projet est intrinsèquement multi-disciplinaire et au croisement de nombreuses technologies et domaines de l’informatique.

Biographie:

Frédéric Prost est MdC à l’université Grenoble Alpes et au laboratoire LIG. Il a  principalement travaillé dans la théorie des langages de programmation (réécriture de graphes, sémantique des langages d’interrogation des BD graphes) et les problématiques de confidentialité (analyse de non interférence, anonymisation de bases de données graphe).

 

 

 


PhD Defence: “On the Performance of Spatial Modulation and Full Duplex Radio Architectures”, Yanni Zhou, Amphitheater, CITI, 10th of December 2021 at 10:00 PM

The defense will be held in the Amphitheater, Claude Chappe, and will be streamed live here.

 

Title

On the Performance of Spatial Modulation and Full Duplex Radio Architectures

 

 

Abstract

Index modulation techniques have exhibited great potential in the scenarios foreseen in next-generation wireless networks. Applying in the spatial domain, spatial modulation (SM) as a single radio-frequency (RF) multiple-input–multiple-output (MIMO) solution has attracted wide attention. The SM system has only one transmitting antenna activated for each time slot which results in low system complexity and cost. It exploits the index of the transmitting antennas to convey additional information bits.

To analyze the SM performance, a simulated framework over the time-varying Rician fading channel is built with ADS and Matlab software and a channel state information (CSI) detector is highlighted. The simulation results are verified by the experimental implementation based on the National Instruments (NI) PXI chassis hardware and LabVIEW programming environment. In the practical analysis, two models of the propagation environments are considered, where a channel sounding method is employed in order to extract the channel coefficients.

Despite issues on system complexity and cost, a shortage of spectrum resources can also restrict the development of mobile communications technology. Full duplex (FD) communications have been developed to double the radio link data rate and spectral efficiency through simultaneous and bidirectional communication. The main challenge of FD systems is self-interference (SI), which is caused by the coupling of the transmitting antenna with the receiving one. The combination of FD and SM will not only maintain spectral efficiency but also decrease the complexity of the self-interference cancellation (SIC) because of the single RF chain.

Based on these, a full duplex spatial modulation (FDSM) system is proposed as well as the SIC method. Moreover, the impact of SIC accuracy on the system performance is studied. We focus on the FDSM system imperfections including IQ imbalance, phase noise, power amplifier (PA) nonlinearities and RF switch nonidealities. The bit error rate (BER) performance under different scenarios with these imperfections is analyzed, along with the estimation and cancellation method.

 

 

Jury

  • Marco Di Renzo, Research Director at CNRS, Reviewer
  • Matthieu Crussière, Professor at INSA-Rennes, Reviewer
  • Christelle Aupetit-Berthelemot, Professor at Université de Limoges, Examiner
  • Taneli Riihonen, Associate Professor at Tampere University, Examiner
  • Jean-Marie Gorce, Professor at INSA-Lyon, Examiner
  • Dinh-Thuy Phan-Huy, Engineer at Orange Lab, Examiner
  • Guillaume Villemaud, Associate Professeur at INSA-Lyon, Thesis director
  • Florin-Doru Hutu, Associate Professor at INSA-Lyon, Co-director

CITI seminar – Miriam Kolar (Stanford University) – 16/11 at 14:00

Title: Human Centered Archaeoacoustics

Date and Place: November 16th 14h00-16h00, Amphi Chappe (CITI Lab, INSA-Lyon, Batiment Claude Chappe), 6 avenue des arts, 69621 Villeurbanne

Speaker: Dr. Miriam Kolar (Adjunct professor at Stanford University)

 

Abstract: 

Despite over 125 years of modern room acoustics, spatial acoustics has only recently been applied in archaeological research. Likewise, musical instrument acoustics remains a growing but infrequent archaeometric approach. Auditory science is even less frequently incorporated. Since 2008, Miriam Kolar has led archaeoacoustics fieldwork and instrument performance studies at the 3,000-year-old UNESCO World Heritage Centre archaeological complex Chavín de Huántar, Perú, with a second Andean project about sound as an Inca administrative tool. In this presentation, Dr. Kolar will share case-study examples from her work in developing methodologies for “human-centered” archaeometric research, relating acoustics to human experience and social behavior in ancient contexts. Acoustical and psychoacoustical experiments in archaeological settings and with artifact sound-producing instruments enable data-driven reconstructions of heritage sites and instruments in use. Physics-based evaluations of human sensory perspectives support the ecological validity of heritage acoustics, opening a new technological frontier for cultural heritage research, preservation, and knowle.

 

Biography:

Miriam A. Kolar, M.F.A., Ph.D., is an Adjunct Professor at the Center for Computer Research in Music and Acoustics (CCRMA) at Stanford University and a Visiting Professor at Amherst College (USA). She studies human-sonic interrelationships across time and geography, applying acoustical and auditory perceptual science methodologies within an anthropological framework. Principal investigator of the integrative archaeoacoustics project at the 3,000-year-old Andean ceremonial site and UNESCO World Heritage Centre Chavín de Huántar, Peru, Dr. Kolar collaborates on novel applications of digital technologies for cultural heritage research and engagement. Her cultural acoustics research (http://culturalacoustics.org) leverages cross-disciplinary theories and tools to understand sonic experiential aspects of past and present life. In current work, and as co-organizer of the NEH-supported Digital Aural Heritage project (http://auralheritage.org), she explores the potential of auralizations for scholarship and public interfacing. Topics of interest include contextual knowledge representation, information ethics, and ecological validity.

This talk is organized in the context of the creation of the Emeraude team which is a collaboration between the Grame institute and Citi-lab at Insa-Lyon (https://team.inria.fr/emeraude/).

 


Health and Privacy-Preserving Machine Learning Workshop, 28th of October 2021 at 2pm, CITI, Campus de la Doua, Lyon

Health and Privacy-Preserving Machine Learning Workshop 2021

La nature sensible des informations de santé est un aspect critique pour le déploiement de l’Intelligence Artificielle (IA) en médecine. L’apprentissage automatique préservant la confidentialité et l’apprentissage fédéré deviennent des paradigmes d’apprentissage centraux pour garantir la confidentialité, la propriété et la sécurité des données médicales sensibles dans les applications d’IA. Malgré les récentes avancées techniques et méthodologiques, l’adoption efficace de paradigmes d’apprentissage sécurisés dans des applications de soins de santé présente encore des défis non résolus et pose des questions juridiques et éthiques. Ce workshop traitera de ces problématiques à travers plusieurs présentations d’expert du domaine.

Inscription sur le site web du workshop.

Retransmission du workshop ici : https://bbb.insa-lyon.fr/b/bou-dzs-2hg-otl

 


CITI seminar – El Hourcine Bergou (INRAE) – 09/04 at 15:00

Title:  Stochastic Three Points Method For Unconstrained Smooth Minimization

Date and Place: 9th April 2021 15:00 – link

Speaker: Dr El Hourcine Bergou (INRAE)

 

Abstract: 

In this work, we consider the unconstrained minimization problem of a smooth function in a setting where only function evaluations are possible. We design a novel randomized derivative-free algorithm—the stochastic three points (STP) method—and analyze its iteration complexity. At each iteration, STP generates a random search direction according to a certain fixed probability law. Our assumptions on this law are very mild: roughly speaking, all laws which do not concentrate all measure on any halfspace passing through the origin will work. Although, our approach is designed to not explicitly use derivatives, it covers some first order methods. For instance, if the probability law is chosen to be the Dirac distribution concentrated on the sign of the gradient, then STP recovers the Signed Gradient Descent method. If the probability law is the uniform distribution on the coordinates of the gradient, then STP recovers the Randomized Coordinate Descent Method.
The complexity of STP depends on the probability law via a simple characteristic closely related to the cosine measure which is used in the analysis of deterministic direct search (DDS) methods. Unlike in DDS, where $O(n)$ ($n$ is the dimension of the problem) function evaluations must be performed in each iteration in the worst case, our method only requires two new function evaluations per iteration. Consequently, while the complexity of DDS depends quadratically on $n$, our method depends linearly on $n$.

 

Biography:

Dr l Hourcine Bergou is a research scientist at INRAE. My research interests are in all areas that intersect with optimization, including algorithms, machine learning, statistics, and operations research. I am particularly interested in algorithms for large scale optimization including randomized and distributed optimization methods.

 


CITI seminar – Michael Barros (University of Essex, UK) – 01/04 at 14:00

Title:  Molecular Communications using Astrocytes for Boolean logic gates implementation in mammalian cells

Date and Place: 1st April 2021 14:00 – link

Speaker: Dr Michael Barros (University of Essex, UK)

 

Abstract: 

In this talk we will show the use of astrocytes to realize Boolean logic gates, through manipulation of the threshold of Ca2+ ion fows between the cells based on the input signals. Through wet-lab experiments that engineer the astrocytes cells with pcDNA3.1-hGPR17 genes as well as chemical compounds, we show that both AND and OR gates can be implemented by controlling Ca2+ signals that fow through the population. A reinforced learning platform is also presented in the paper to optimize the Ca2+ activated level and time slot of input signals Tb into the gate. This design platform caters for any size and connectivity of the cell population, by taking into consideration the delay and noise produced from the signalling between the cells. To validate the efectiveness of the reinforced learning platform, a Ca2+ signalling simulator was used to simulate the signalling between the astrocyte cells. The results from the simulation show that an optimum value for both the Ca2+ activated level and time slot of input signals Tb is required to achieve up to 90% accuracy for both the AND and OR gates. Our method can be used as the basis for future Neural–Molecular Computing chips, constructed from engineered astrocyte cells, which can form the basis for a new generation of brain implants.

 

Biography:

Dr Barros is an Assistant Professor (Lecturer) since June 2020 in the School of Computer Science and Electronic Engineering at the University of Essex, UK. He is also a MSCA-IF Research Fellow (part-time) at the Tampere University, Finland. He received the PhD in Computer Science at the Waterford Institute of Technology in 2016. He previously held multiple academic positions as a Research Fellow in the Waterford Institute of Technology, Ireland.
He has over 70 research peer-reviewed scientific publications in top journals and conferences such as Nature Scientific Reports, IEEE Transactions on Communications, IEEE Transactions on Vehicular Technology, in the areas of molecular and unconventional communications, biomedical engineering, bionano science and 6G. Since 2020, he is a review editor for the Frontiers in Communications and Networks journal in the area of unconventional communications. He also served as guest editor for the IEEE Transactions on Molecular, Biological and Multi-Scale Communications and Digital Communications Networks journals. He received the CONNECT Prof. Tom Brazil Excellence in Research Award in 2020. Dr Barros was awarded also the Irish Research Council’s (IRC) Government of Ireland Post-doc Fellow from 2016-2018 and the Enterprise Ireland’s (EI) Commercialization Funding from 2018-2019.

 


CITI seminar – Somantika Datta (Univ. Idaho, USA) – 18/02 at 14:00

Title: Construction and properties of certain real multi-angle tight frames

Date and Place: 18th February 2021 14:00 – link

Speaker: Prof. Somantika Datta (Univ. Idaho, USA)

 

Abstract: 

Frames are now standard tools in signal processing, and have applications ranging from compressed sensing, to communication systems and quantum sensing. Designing frames with some special structure such as equiangularity and tightness is highly desirable in applications. However, constructing equiangular tight frames (ETFs) with a given size in a specific dimension can be difficult or impossible in some cases. This leads one to consider the construction of frames with few distinct angles among pairs of frame vectors. In the special case of d+1 vectors in a d-dimensional space, it is well known that the vertices of a regular simplex will give an ETF. Using this, we will show a specific construction which, for a given dimension d and integer 1 < k ≤ d, gives a real unit norm tight frame such that the number of distinct angles among the vectors is bounded above by k. We will present several properties of this multi-angle tight frame. We also show how one can strategically choose subsets of such a multi-angle tight frame that will be equiangular or orthogonal. This property is meaningful in the context of erasures. We will also discuss a connection between certain unit norm tight frames with three angles and adjacency matrices of regular graphs.

 

Biography:

Somantika Datta is an associate professor of mathematics at the University of Idaho. She received a Ph.D. in mathematics from the University of Maryland, College Park. This was followed by postdoctoral positions at Arizona State University and Princeton University. Her research interests lie in the area of applied harmonic analysis with focus on frame theory and applications in signal processing.

 


CITI seminar – Daryus Chandra (University of Southampton, UK) – 11/02 at 14:00

Title: Quantum Communications over Noisy Entanglement

Date and Place: 11th February 2021 14:00 – link

Speaker: Dr Daryus Chandra (University of Southampton, UK)

 

Abstract: 

Within the Quantum Internet framework, multiple quantum devices are interconnected via pre-shared maximally-entangled quantum states for enabling various applications, including the on-demand classical and quantum communication. Hence, the pre-shared entanglement, which is constituted by the EPR pair, can be viewed as the primary consumable resources within the Quantum Internet. However, the generation and the distribution of the EPR pairs are subject to quantum decoherence imposed by the quantum channels, which will manifest as quantum errors. Similar to the classical domain, the quantum errors imposed by the quantum channels can be mitigated using quantum error-correction codes. In this talk, we will explore two approaches for achieving reliable quantum communication over noisy entanglement by incorporating quantum error-correction codes. More specifically, the first approach is constituted by the consecutive steps of quantum entanglement distillation followed by quantum teleportation, while the second approach can be viewed as the direct quantum communication over noisy entanglement. We will also discuss the pros and the cons of each approach while examining their compatibilities for a broader range of applications for the Quantum Internet framework.

 

Biography:

Daryus Chandra is a research fellow at the Next-Generation Wireless Research Group, School of Electronics and Computer Science, University of Southampton, UK. He received the B.Eng. and M.Eng. degree from the Department of Electrical Engineering and Information Technology, Universitas Gadjah Mada, Indonesia, in 2013 and 2014, respectively. He obtained his PhD with the Next-Generation Wireless Research Group, School of Electronics and Computer Science, University of Southampton, UK, in 2020. He returned to Southampton in 2021 after completing a one-year postdoctoral research fellowship at Quantum Internet Research Group, University of Naples Federico II, Italy.