Fully funded PhD position available at CRIL CNRS Université d’Artois (Lens, France) Starting date: Fall 2023

Duration: 3 years

Funded by: CNRS - University of Arizona joint PhD programme

Brief project description: Some grand challenges facing humanity include climate change and health. Unfortunately, the implementation of effective solutions for these challenges is counteracted by misinformation, disinformation, and malinformation (MDM), e.g., climate-change denying or anti-vaccine propaganda.

To address this issue, project SURGING (Using argumentation for fack-checking) proposes robust and holistic fact verification methods. The proposed methods can reduce data bias, aggregate information across multiple statements, and yield global conclusions.

While humans can utilize background and domain knowledge to argue about the veracity of a fact, computers do not normally have access to such information. To capture this real-world background knowledge, we aim to mine arguments from the web and construct domain-agnostic fact graphs that indicate if facts attack/support each other.

We then propose to develop argumentation theory-based graph algorithms to aggregate and argue over this knowledge. Based on these steps, we will arrive at the truthfulness value of a given argument that considers all the background knowledge available.

https://www.fainstitute.arizona.edu/using-argumentation-fact-checking-surfing

The PhD student will be hosted by the French team, and will primarily be working in the area of computational argumentation.

Supervisor: Dr Srdjan Vesic (CNRS)

Leader of the USA team: Dr Mihai Surdeanu (University of Arizona)

For more information and to apply, send a CV to [email protected]

As part of a development project on the theme of computer-assisted language learning, the SyNaLP team at Loria (CNRS UMR 7503 located in Nancy) is looking for an engineer in full-stack application development (M/F) for a period of 18 months.

Details of the offer are available on the CNRS portal at the next address :

https://emploi.cnrs.fr/Offres/CDD/UMR7503-YANPAR-001/Default.aspx

Host laboratory: Connaissance et Intelligence Artificielle Distribuées (CIAD) – http://www.ciad-lab.fr

Belfort, FRANCE

Specialty of PhD: Computer Science

Keywords: Artificial intelligence, Operations Research, Signaling systems, optimization of cyclist/vehicle interactions, virtual reality

Contacts : Arnaud Doniec [email protected] Guillaume Lozenguez [email protected] Flavien Lucas [email protected]

Topic: Distributed Vehicle Routing Problem in Large Transportation Networks

Home: CERI SN, IMT Nord Europe Keywords: Multi-Agent Modeling, Combinatorial Optimization, Transport Simulation, Distributed System, Coordination

We propose a funded thesis at the MICS laboratory at CentraleSupélec in computational social choice.

Topic: Explaining fairness in preference-based assignment problems

Keywords: Computational social choice, explainable AI, multi-agent systems, fair sharing, fair matching

Supervision: Anaëlle Wilczynski and Wassila Ouerdane

Environment: MICS laboratory at CentraleSupélec (Paris-Saclay University)

Funding: 3-year doctoral contract (ANR Apple-Pie project)

Start of thesis: October 2023

Profile sought: Students in Master 2 of Computer Science or Applied Mathematics (especially in the specialty Artificial Intelligence or Operational Research) or student-engineers in their final year (BAC+5) interested in collective decision-making and algorithms. Applications: Contact Anaëlle Wilczynski ( [email protected] ) and Wassila Ouerdane ( [email protected] ) by sending a CV, a cover letter as well as the available notes from the past and current year.

Subject: With the increasing autonomy of land, air or sea robots, robot fleets are now used for many types of mission. This is for example the case of the delivery of parcels, flying taxis, exploration of areas, rescue missions or intervention at disaster sites. Some of these applications require fleets of heterogeneous robots, ie which have different capacities, eg to detect, communicate, observe, move, etc.

In this thesis, we wish to address the constraints and criteria of performance and robustness from the configuration phase of the fleet, and this in cases where the configuration is complex. Specifically, given one or more mission types for the fleet with target performance and robustness and associated robot capabilities, a set of possible equipment with their compatibility constraints, and the relationships between equipment and robot capabilities, we seek to define the number of robots making up the fleet and the configuration of each robot so as to achieve the desired performance and robustness objectives. The problem then consists in exploring the space of configurations while being guided by the evaluation of performance and robustness. More details here: https://lnkd.in/ecFvhyPE

Management team: Stephanie Roussel , Gauthier Picard And Elise Vareilles

Possibility of M2 internship (6 months) upstream at ISAE SUPAERO

Details at: https://www.uu.se/en/about-uu/join-us/details/?positionId=601303

Application deadline: April 6, 2023.

The project A fully funded postdoctoral position is available for highly motivated, creative and responsible individuals, with experience and interest in machine and deep learning applied to cancer research. The project aims to develop computational analysis tools to analyze multidimensional images of cancer tissue, towards optimized immunotherapy for cancer patients. Immunotherapy has become a life-saving option for advanced cancer patients. However, only a minority of patients develop a durable response. Despite great efforts to explain the variable responses to immunotherapy and to optimize patient selection, the currently used clinical biomarkers demonstrate only modest predictive performance. Starting from a large collection of acquired multispectral images, and by developing advanced data driven approaches for image data analysis, we wish to increase understanding of the effects of immunotherapy, towards improved personalized cancer treatments.

Requirements PhD degree (or equivalent) in a field closely related to the position (e.g., computerized image analysis/processing, data science, computer science). Those who have obtained a PhD degree not more than three years prior to the application deadline are primarily considered for the employment. Documented experience in image analysis, machine and deep learning, including completed courses at the master and doctoral level of education, as well as first hand experience in method development, implementation, evaluation and publication (first author) of scientific articles in internationally recognized journals and conferences. Programming in Python, and experience of working with deep learning in the PyTorch environment.

Additional qualifications Meriting are: - Experience with graph-based methods, network analysis tools, and graph convolutional/neural networks; - Experience with explainable AI (XAI); - Experience of university level teaching and supervision; - Experience of programming in Matlab, JavaScript, C++, Java, software version control with Git, typesetting with LaTeX, use and administration of Linux computers, Bash scripting; - Interest in biomedical research and experience in application of image analysis in medicine.

Application The application should consist of:

1. A Curriculum Vitae (CV);

2. A copy of degree/diploma (translated into Swedish or English), with the list of relevant completed courses;

3. List of publications;

4. Up to five selected publications in electronic form, with stated own contributions;

5. PhD thesis in electronic form;

6. Description of your current and previous research (max 1 page) and suggestions for future research (max 1 page) along the project goals relevant for this position. The statement should explain how your profile fits the position;

7. Contact information for two references (name, e-mail, and phone number);

8. A personal letter in which you briefly justify why you are applying for this position and state the earliest possible starting date (max. 1 page).

For further information about the position, please contact: Nataša Sladoje, [email protected] Joakim Lindblad, [email protected]

#Location: Clermont-Ferrand, France.

#Host institutions: Institut Pascal.

#Starting Date: 01/10/2023.

#PhD Duration: 3 years.

#Supervisors: Dr. Mohammad Alkhatib, Dr. Erol Ozgur, Prof. Youcef Mezouar.

#Application Deadline: Open until filled.

#Context: Liver cancer is a leading cause of cancer death worldwide. An estimated 830,000 people around the world died from the disease in 2020. Liver resection is considered as one of the most effective treatments. In this respect, laparoscopic liver resection (LLR) comes up by reducing substantially patient trauma compared to open liver resection. The patient recovers faster which in return reduces healthcare costs.

However, the use of LLR remains limited. This is because of three challenges. First, controlling intraoperative bleeding using laparoscopic instruments requires advanced technical skills. Second, the surgeon cannot manually palpate the liver and thus cannot locate the tumors and their resection margins easily. Consequently, this raises a risk of inadequate resection on the patient’s liver such as the removal of too much healthy tissue and the leaving of some cancerogenic tissue behind. Third, laparoscopic ultrasonography (LUS), the only tool for intraoperative subsurface imaging which allows real-time tumor localization, has a long learning curve. This is because its design consists of a small transducer with a small field of view attached to the end of a long shaft with a pivoting mechanism.

In order to ease LLR, robotic solutions would provide great assistance by controlling the LUS. For that, the goal of this PhD is to develop and implement a solution for contact servoing of a flexible ultrasound probe on the deformable liver surface. The setup includes a robotic arm mounted with a flexible ultrasound probe, a camera and a phantom liver organ. Contact servoing will be based on the images of the laparoscope and the ultrasound probe.

#Research: We are looking for one highly motivated PhD student to study multimodal contact servoing on a deformable liver. The PhD student will focus on the following open problems: 1. Automatic liver segmentation in laparoscopy images; 2. Multimodal contact servoing: a. bring the ultrasound probe in contact with the liver; b. to move the probe on the liver surface.

The PhD will be in close collaboration with the scientists and surgeons and the successful outcome of this PhD will simplify mini-invasive liver surgery. It will shorten hospital stays, improve surgical safety and accuracy, and contribute to an overall better quality of patient life and reduction of healthcare costs.

#Requirements: Applicant must have: 1/ undergraduate and graduate degrees in Computer Science, Robotics or closely related fields; 2/ excellent programming skills in C++ and python; 3/ strong theoretical and applied background in computer vision, machine learning, visual servoing and robotics; 4/ proficiency in written and spoken English language.

#Application: Applicant must submit: 1/ a one-page cover letter, 2/ curriculum vitae with publications list and contacts of 2 references, 3/ a copy of academic transcripts (bachelor/master grades), 4/ availability (the earliest possible starting date).

All should be sent, ***in a single PDF document***, with the email subject [PhD application - liver] to:

[email protected]; [email protected]; [email protected]

Applicants must also be prepared to provide two reference letters upon request. Once we receive your application and it fits well for the position, you will be contacted within two weeks.

*Place of work*: Inria center in Paris area (Paris / Rocquencourt / Saclay)

*Duration*: 2 years

*Starting date*: as soon as possible

*Keywords*: artificial intelligence, natural language processing, information extraction, domain adaptation, chemistry, cybersecurity, geopolitics

*Context*

This post-doc position fits within the roadmap activities of Inria’s Defense & Security Department, which is devoted to applications-driven research.

Among the various fields of NLP, information extraction is a crossroad topic that, by focusing on how to turn raw documents into structured data models, echoes the practical needs of many end-users in a broad range of sectors. Information extraction components such as entity recognition or relation extraction are thus key to a number of industrial and general-public applications.

However, whereas information extraction has seen major progresses in the last few years on common language (Wikipedia, news, everyday language), it still lags behind on specialty language, which effectively affects a number of practical applications. Main challenges include unknown words and concepts, unusual phrasings, or differences in the nature of information that is interesting to extract.

The goal of this post-doc is to bridge that gap by developing new methods that enable to model and account for the specificities of a given domain with specialty language, while still benefitting from the models and capabilities developed for the common language.

The first specialty-language domain that has been identified as a test bed for the developed approaches is the scientific literature on chemistry (e.g. ChemRxiv papers). Other domains that are considered for experimentation throughout the period are cybersecurity (e.g. technical documentation) and geopolitics. Inspiration can be drawn from existing work on biomedical NLP, but that domain is not expected to be at the core of the work.

The post-doc will work under the supervision of Lauriane Aufrant (lead NLP researcher at Inria Defense & Security). Work can include direct collaborat ion with other academic or industrial partners of the department.

*Candidate profile*

Holding a PhD (or about to defend) in Natural Language Processing, Computational Linguistics or Computer Science with a specialization in Machine Learning

Theoretical and practical knowledge of deep learning, as well as traditional machine learning. Experience with knowledge-driven or hybrid AI would be appreciated.

Prior experience on at least one of the following topics: information extraction, semi-supervised learning, domain adaptation, low-resourced NLP

Strong programming skills (at least Python, git, Linux environment)

Fluency in English. Knowledge or interest for the French language. Knowle of a second foreign language would be appreciated.

*How to apply*

Send a CV and a cover letter to lauriane.aufrant and frederique.segond (both at inria.fr) Indications of referees or reference letters would be appreciated but are not mandatory.

*Work description*

The post-doc will focus on developing new algorithmic methods along the following research tracks:

- Automated terminology and concepts extraction

- Identification of new relations that are specific to a domain

- Adapting models (in particular embedding models) to account for extended vocabulary

- Semi-supervised learning to leverage a small amount of in-domain annotations

Special care will be given to the transferability of the methods to other specialty domains, rather than developing approaches that are tailored to one particular domain.

Angelo Fanelli and I are looking for candidates for a thesis in algorithmic game theory at LAMSADE (Paris Dauphine University). The profile sought and the details of the thesis are available at the bottom of the following page.

https://sites.google.com/view/angelofanelli?pli=1

Please contact us before mid-April.

Transform your future with a transnational joint PhD from two leading universities Attractive scholarships are available including funded travel to Melbourne, an Australian stipend, and overseas health cover

Join the unique collaboration between BITS (Birla Institute of Technology and Science) and RMIT (Royal Melbourne Institute of Technology) offering PhD program specifically designed for Indian students. Experience world-class education combining the research expertise of both institutions and gain exposure to cutting-edge research.

Title of the Project: 1. Car driving training support via augmented reality-based simulation 2. Rethinking Self-Supervised Learning for Low Data Scenario 3. Medical Diagnosis based on Fusion of Small Multimodal Data Highlights

Acquire an internationally recognized qualification from an Indian Institute of Eminence University and a top Australian university

Receive a full RMIT tuition fee scholarship for the duration of your enrolment

Receive a generous scholarship with nominal tuition fees from BITS

Benefit from the combined expertise of two leading universities and access world-class research facilities in India and in Melbourne

Travel to Australia for up to one year of candidature and be supported by an Australian stipend for the duration of your time in Melbourne

Candidates admitted to the program are jointly supervised by faculty from BITS and RMIT.

For further details see the attachment. For any queries feel free to contact undersign,

We are accepting applications for PhD fellowships at the CYENS Centre of Excellence, located in sunny Nicosia, Cyprus. The positions are full-time for 3 years starting from an MSc or equivalent.

This post concerns the thematic area of Continual Skill Learning.

Short description: The field of reinforcement learning (RL) has seen tremendous progress over the last few years. Despite their impressive (often superhuman) performance, current RL systems suffer from the following drawbacks: (1) they are “specialists” rather than “generalists”, i.e., they become highly specialized to their current task, often forgetting completely how to solve their previously learned tasks, and (2) they require enormous data and computational resources for training. In contrast to such systems, human learning is efficient and robust to changing environments. Our brains can construct abstract predictive models from their sensorimotor experience that permit efficient planning, rapidly assimilate new knowledge in memory, and flexibly use it to build complex skills that are reused in many tasks. This project aims to make a significant step towards the highly sought goal of “artificial general intelligence”, by developing software agents that continually acquire, reuse and improve a variety of skills, similarly to humans. The successful candidate will have the opportunity to build on top of the latest research on areas such as continual learning in deep neural networks, hierarchical reinforcement learning, memory-augmented neural networks, meta-learning, quality-diversity optimization or open-endedness, and create procedurally-generated environments in which the developed agents will be evaluated.

The candidate will be registered at the University of Cyprus, but based at CYENS at the "Learning Agents and Robots" research group (lear.cyens.org.cy). The candidate will have the opportunity to spend 15%-25% of their PhD time at IT University of Copenhagen (collaborating partner).

Questions about the position can be directed to Vassilis Vassiliades, CYENS, email: [email protected]

For more information and how to apply please visit: https://www.cyens.org.cy/en-gb/vacancies/job-listings/research-associates/phd-fellowships-cyens-doctoral-training-progra-1/

Deadline: 31 March 2023

We are accepting applications for 3-year PhD fellowships at the CYENS Centre of Excellence, located in sunny Nicosia, Cyprus. The positions are full-time for 3 years starting from an MSc or equivalent.

This post concerns the thematic area of Quadrupedal Robot Learning.

Short description: Quadruped (4-legged) robots are an emerging technology that has recently become commercially available. These robots have unique potential advantages over other types of robots, such as increased agility on unstructured terrain compared to wheeled or tracked systems, or increased payload carriage compared to flying systems. Deploying such robots in the real-world, however, presents several challenges related to perception, locomotion, navigation, as well as human safety. A common approach to address such challenges is to train robots in simulation, however, effective transfer from simulation-to-reality (sim2real) is still a subject of ongoing research. This project aims to investigate novel approaches for bridging the sim2real gap by developing (potentially photorealistic) simulation scenarios and machine learning algorithms for training quadruped robots towards their deployment in real-world settings. The successful candidate will have the opportunity to investigate approaches such as deep reinforcement learning for robust locomotion from vision, damage recovery, lifelong localization and mapping, as well as aspects of human-robot interaction and shared autonomy. The developed approaches will be evaluated both in simulated environments and on physical quadruped robots.

The candidate will be registered at the University of Cyprus, but based at CYENS at the "Learning Agents and Robots" research group (lear.cyens.org.cy). The candidate will have the opportunity to spend 15%-25% of their PhD time at University College London (collaborating partner).

Questions about the position can be directed to Vassilis Vassiliades, CYENS, email: [email protected]

For more information and how to apply please visit: https://www.cyens.org.cy/en-gb/vacancies/job-listings/research-associates/phd-fellowships-cyens-doctoral-training-progra-1/

Deadline: 31 March 2023

We are looking for postdoctoral researchers in machine learning and robotics at the CYENS Centre of Excellence in Cyprus. The postdoctoral researchers will join the Learning Agents and Robots research group, working on basic and applied research.

For more information and how to apply, please visit: https://www.cyens.org.cy/en-gb/vacancies/job-listings/research-associates/post-doctoral-researcher-positions-in-machine-lear/

Deadline: 30 March 2023

Keywords: video description, deep learning, convolutional neural network, coreference resolution, knowledge graph, multi-tasking, multimodality

Abstract: The topic of this PhD thesis is the automatic generation of video descriptions based on automatic natural language processing and deep learning. The goal is to overcome the limitations of existing databases in terms of encoding, multimodality, standardization, ground truth and contextualization to improve the performance of video description methods. To this end, we plan to apply convolutional neural networks on videos enriched with textual and semantic data, relying in particular on the knowledge graphs of the Web of Data. This PhD thesis work involves solving scientific challenges such as coreference resolution and multi-task and multi-modal processing for performance evaluation. In addition, the project will contribute to the development of large-scale standardized databases for performance evaluation of video description methods, which is essential for future research in this area.

Full subject: https://www.info.univ-tours.fr/~soulet/download/phdsubject_short_en.pdf

Laboratory: LIFAT Laboratory (BDTLN and RFAI teams), University of Tours (UT), Tours city, France Location: Blois and Tours cities, France

Duration: 3 years

Funding: A fully funded 3-years PhD position / salary of 1600€ a month (take-home pay)

Profile of the candidate: Master's degree in Computer Science, initiation to research (teaching, or project, or internship), motivation for NLP, Computer Vision and deep learning, some knowledge in French would be appreciated. Link for application: https://collegedoctoral-cvl.fr/as/ed/voirproposition.pl?site=CDCVL&matricule_prop=46957&langue=en

Deadline for application: 15/05/2023

Required documents: M1 and M2 report cards, letters of recommendation

Contact emails: [email protected] / , [email protected], [email protected], [email protected] Links: University of Tours https://international.univ-tours.fr/, LIFAT Lab https://lifat.univ-tours.fr/ , RFAI group https://www.rfai.lifat.univ-tours.fr/ , Home pages ( https://www.info.univ-tours.fr/~soulet/ , https://www.info.univ-tours.fr/~friburger/ , http://mathieu.delalandre.free.fr/ )

We are looking for an enthusiastic and proactive Ph.D. (or M.S.) student to join the Aerial Robotics Research Facility (https://knurobot.wixsite.com/arrf) at Kyungpook National University (KNU) in South Korea.

We’d love to hear from applicants for the following domains.

Field 1

- Aerial Robotics, Drone, Unmanned Aerial Vehicle (UAV), - Urban Air Mobility (UAM), Advanced Air Mobility (AAM) - Autonomous Flight, Autonomous Navigation, Path Planning, Obstacle Avoidance - Multiple Heterogeneous Unmanned Vehicles, Multi-agent System

Field 2

- Robot Vision, Computer Vision, Perception - Simultaneous Localization and Mapping (SLAM), Visual-Inertial Odometry (VIO) - Visual Servoing - Point Cloud Processing

To apply, send your resume / CV to [email protected]. Be sure to include Robotics Ph.D. (or MS) Student and your name in the subject line.

If you have any questions regarding this post, please feel free to contact Prof. Kyuman Lee [email protected].

https://www.info.univ-tours.fr/~soulet/download/phdsubject_short_fr.pdf

Abstract: The subject of the thesis concerns the automatic generation of video descriptions based on the automatic processing of the natural language and deep learning. The goal is to overcome the limitations of existing databases in terms of encoding, multimodality, standardization, ground truth and contextualization to improve the performance of methods of video description.

For this, we plan to apply convolutional neural networks on videos enriched with textual and semantic data based in particular on the web of data knowledge graphs. This thesis work involves solving scientific challenges such as solving coreferences and multi-task and multimodal processing for performance evaluation. In addition, the project will contribute to the constitution of standardized and large-scale databases for performance evaluation of video description methods, this which is essential for future research in this area.

Keywords: video description, deep learning, convolutional neural network, resolution of coreferences, knowledge graph, multitasking, multimodality

Laboratory: LIFAT ( https://lifat.univ-tours.fr/ ), BDTLN teams and RFAI, University of Tours

Location: Blois and Tours

Duration: 3 years

Funding: 3 years / salary: 1600€

Candidate profile: Master's degree in computer science, initiation to research (teaching followed, or project, or internship), motivation for TAL and deep learning imagery

Link to apply with file submission under ADUM: https://collegedoctoral-cvl.fr/as/ed/voirproposition.pl?site=CDCVL&matricule_prop=46957 Deadline for complete file in ADUM: 05/15/2023

Necessary documents: M1 and M2 report cards, letters of recommendation

Contacts for information: [email protected] , [email protected] , [email protected] , [email protected]

Title: Unveiling and Incorporating Knowledge in Physics-Guided Machine Learning Models

Keywords: Physics-guided models; Neural networks; Sparsity; Transfer learning; Optimization

Location: Laboratoire Hubert Curien, 18 rue Benoit Lauras, 42000 Saint-Etienne, France

Candidate profile: Master in computer science, applied mathematics, engineering school, physics with good computer science background. Background in machine learning, good programming skills in python.

Application deadline: May 1st, 2023

Application documents: CV, cover letter, master grades, recommendation letters

Contacts: [email protected] and [email protected]

Start date: from October 2023

Starting date: September 1st, 2023 (flexible)

Application deadline: July 10th, 2023

Interviews (tentative): July 15th, 2023

Salary: ~2000€ gross/month (social security included)

Mission: research oriented (teaching possible but not mandatory)

Keywords: speech processing, automatic speaker recognition, anti- spoofing, deep neural network

CONTEXT It is now widely accepted that automatic speaker recognition (ASV) systems are vulnerable not only to speech produced artificially by text-to-speech (TTS) [1], but also to other forms of attacks such as voice conversion (VC) and replay [2]. Voice conversion can be used to manipulate the voice identity of a speech signal, has progressed extremely rapidly in recent years [3], and has indeed become a serious threat.

The progress made in recent years in deep neural networks training has enabled spectacular advances in the fields of text-to-speech (TTS) and voice conversion (VC): DeepVoice, Tacotron 1 and 2 [4], Auto-VC [5,6]. Existing architectures now make possible producing synthesized or manipulated artificial voices with a realism close to or equal to that of human voices [4]. At the same time, voice conversion algorithms (from one speaker to another) have also made spectacular advances. It now becomes possible to clone a voice identity using a small amount of data. In the space of two years, extremely significant advances have been made [5,6,7]. The ability of these algorithms to forge voice identities capable of deceiving speaker recognition and counter-measure systems is an urgent topic of research.

Progress in terms of the fight against identity theft has been led by the initiative of the ASVspoof community, formed in 2013 and recognized as competent at the international level [8]. The most significant efforts have been made at the level of the acoustic parametrization (front-end) making it possible to better differentiate authentic (human) utterances from fraudulent utterances. The best performing system [9], which combines acoustic parameters based on Cepstrum-Mel, Cepstrum based on cochlear filters and instantaneous frequencies using a classifier based on a Gaussian mixture model, obtained the best performance.

For the past years, research efforts have focused on the back-end. As in speaker recognition research, the anti-spoofing community has embraced the power of deep learning and, unsurprisingly, the neural architectures used are almost the same. Advances in anti-spoofing have followed the rapid advances in TTS and VC. The best anti-spoofing system again used traditional acoustic parameters, with a classifier based on ResNet-18 [10].

SCIENTIFIC OBJECTIVES As part of this thesis, the robustness of existing countermeasures against new forms of adversarial attacks designed specifically to deceive them will be assessed. One of the advances expected in this thesis will focus on the design of new countermeasures to detect such emerging, increasinly adversarial attacks. To do this, two avenues will be explored. The first is to redesign front-end feature extraction to capture cues that characterize adversarial attacks, then use them with re-trained classifiers.

As it is not always easy to identify reliable characteristics, the second direction will aim at the adoption of end-to-end architectures able to learn characteristics automatically. Although these advances improve robustness to adversarial attacks, it will be important to ensure that the resulting countermeasures remain robust to previous attacks. This is known as the problem of the generalization. An effective anti-spoofing countermeasure must reliably detect any form of attack it encounters, not just the specific attacks it is trained to detect. Finally, improving adversarial attack detection performance should not come at the cost of increased false positives (genuine speech labeled as spoofed speech), which can hurt usability and convenience. The progress and results targeted in this thesis will therefore be countermeasures capable of defending speaker recognition systems against adversarial and non-adversarial attacks.

In parallel to this competition between research teams specializing in attacks and research teams specializing in counter-attacks, the speaker recognition community is focused on the creation and design of high-performance systems that are robust to acoustic variability. Recognition systems are trained to recognize speakers in increasingly difficult conditions (presence of several types of noise: additive, reverb, etc.).

This robustness against difficult acoustic conditions can lead to weakness against recordings of attacks that were not taken into account during training. Of course this vulnerability can be reduced by using countermeasures (CM) systems. This approach can impact the usability of ASV systems since the countermeasures can also reject genuine clients (authentic users). This thesis will therefore go beyond the state of the art by optimizing both the ASV and the CM system, so that they work together to achieve the best possible compromise between security and usability/convenience.

REQUIRED SKILLS - Master 2 in speech processing, computer science or data science

- Good mastering of Python programming and deep learning framework

- Previous experience in bias in machine learning would be a plus

- Good communication skills in English

- Good command of French would be a plus but is not mandatory

LAB AND SUPERVISION The PhD position will be co-supervised by Nicholas Evans from EURECOM and Driss Matrouf from LIA-Avignon. Joint meetings are planned on a regular basis and the student is expected to spend time in LIA-Avignon. The students, a long with the partners (IRCAM specialized in attack generation and EURECOM specialized in countermeasures) will closely collaborate.

INSTRUCTIONS FOR APPLYING Applications must contain: CV + letter/message of motivation + master notes + be ready to provide letter(s) of recommendation; and be addressed to Driss Matrouf ([email protected]), Mickael Rouvier ([email protected]) and Nicholas Evans ([email protected]).

Application deadline: March 24, 2023

Desired start: from May 2023

Duration: 12 months

Funding: DOING regional project (APR-IA, Center Val de Loire region) Research laboratory: LLL-CNRS ( https://lll.cnrs.fr/ ), University from Orleans, France

Collaboration with the following laboratories: ---- LIFAT ( https://lifat.univ-tours.fr/ ) and ---- LIFO ( https://www.univ-orleans.fr/lifo/?lang=en ) Salary: 2580€ gross monthly

# CONTEXT The DOING regional project aims to develop methods and tools to first extract information from data text by structuring them in a graph database, then to intelligently manipulate this knowledge graph. THE field of application chosen is the field of health, with primarily the use of freely available data (such as clinical cases). DOING aims to design a new form of declarative queries, which can integrate analyses, which will guide the healthcare professionals in their decision-making. DOING is designed on a real interdisciplinary collaboration (Treatment Automatic Languages, Databases and Artificial Intelligence) to transform data into information and then into knowledge.

DOING ( https://www.univ-orleans.fr/lifo/evenements/doing/ ) is based on an interdisciplinary collaboration to transform data in information and then in knowledge. This project is part of the DOING@MADICS action and the group DOING@DIAMS working group.

# SUBJECT The research work concerns task 1 of the DOING project, and is will focus mainly on the 2nd part concerning the extraction relationships.

(1) Detection and categorization of entities of interest (eg, pathology, treatment). Initial work has been done to develop a CRF-based system (Minard et al., 2020) in the as part of our participation in the DEFT 2020 evaluation campaign. One task will be to improve this system, possibly by using other corpora of clinical cases: corpora of DEFT 2019, 2020 annotated with clinical entities; current E3C corpus annotation (Magnini et al., 2020).

(2) Detection and categorization of relationships between entities. The extraction of relations makes it possible to bring out a meaning that will be represented in the data graph that we seek to construct. We will focus on two types of essential relationships: (2.1) Coreference relations: they exist between two linguistic units referring to the same entity speech (e.g. "Alice is feeling chills.

This symptom is related to his fever."). Current techniques are based on classic, generic neural models, trained from the corpora of the general domain, and suffer from a significant performance degradation when they are applied to another domain (Zhang et al., 2020). All in considering the specificity of co-reference in the language of medical specialty, we will seek to develop relatively generic models, more robust in terms of cross-domain performance and explainable.

To lift this last lock, we will use techniques interpretable statistical learning (trees of decision, random tree forests) proposed by our TAL group (Desoyer et al., 2014). In parallel, we we will be able to adapt the neural system proposed by the LATTICE laboratory whose designer is associated with LIFO (Grobol, 2020).

(2.2) Temporal relations: they make it possible to order the events concerning a patient (the appearance of symptoms, treatments followed, etc.).

We will work on the development of an extraction system temporal information and will test methods making it possible to compensate for the small quantity of data in the medical domain.

We will also continue the work started in the Temporal and ODIL projects by our group TAL (Lefeuvre-Halftermeyer et al., 2016), by evaluating the genericity of the proposed scheme and by studying the addition possible of the notion of containers (Pustejovsky and Stubbs, 2011).

(3) Interaction between information extraction, construction and use of graph databases. One method would aim to consider how relationship extraction can benefit from approaches used on graph databases (page rank, betweenness, etc.). The observations from T3 would come complement and integrate with T1 results for instantiation from the base.

CREW This postdoc is part of a collaborative work that involves the following researchers:

- LLL: Anne-Lyse Minard, Lotfi Abouda, Flora Badin, Emmanuel Schang - LIFO: Anaïs Lefeuvre-Haltermeyer - LIFAT: Jean-Yves Antoine - LISN: Agata Savary

REQUIRED PROFILE The candidate must:

- have a doctorate in computer science or linguistics with a specialization in NLP

- have prior knowledge of machine learning

- past experience on the following topics will be appreciated: resolution of coreferences, extraction of temporal relations, extraction of information in a specialized field

- have a level of French allowing to analyze the data and to discuss with the team

The research work is carried out at the Ligerian Laboratory of Linguistics (LLL) in Orléans. The recruited person must be present physically (it is not possible to work remotely).

TO APPLY You must send your application by email, no later than March 24 2023, to the following address: [email protected]

The application file must contain: - a detailed CV - a cover letter - the doctoral degree - pre-defence reports - two references

Gustave Eiffel University is recruiting an "IA" Research Officer, the competition is open, could you widely distribute the link below in your networks?

https://www.concours.developpement-durable.gouv.fr/IMG/pdf/crcn_20_uge_recrutement_2023.pdf

Deep Learning, Vision, OCR, Document Understanding, Automatic Language Processing

Contact person: Thierry Paquet, [email protected]

Pierrick Tranouez, [email protected]

Clement Chatelain, [email protected]

We are looking for a candidate currently registered in last year of a Master's degree or engineering school , with a background in machine learning, and significant experience in Deep Learning applied to vision or natural language processing.

Send CV, application letter

Thierry PAQUET __________________________________________ LITIS UR 4108 Laboratory www.litislab.eu

University of Rouen Normandy Tel:+33 2.32.95.50.13 GSM:+33 6.70.74.99.45 Skype: thierrypaquet

FR CNRS 3638 Normastic www.normastic.fr

Home: http://pagesperso.litislab.fr/tpaquet/