Scientific direction Development of key enabling technologies
Transfer of knowledge to industry

PhD : selection by topics

Technological challenges >> Artificial intelligence & Data intelligence
21 proposition(s).

See all positions [+]

AI processing of time series for smart sensors

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire Infrastructure et Ateliers Logiciels pour Puces

01-10-2020

SL-DRT-20-0261

marielle.malfante@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Today, sensors are used to acquire data of a given modality (acoustics, pressure, image, etc.). Usually, such data are stored before being analysed, for instance with machine learning methods. The relevant information is thereby extracted. A large variety of sensors and use cases can be considered: - Microphones for the automatic classification of acoustic landscapes, - Pressure sensors to study deformation and monitor various architectures (brides, dams, wind turbines, etc.), - Seismometers to detect signals warning of a seisms or of a volcanic eruption, - Smart watches or bracelets to detect stress phases, Etc. The issue of smart sensors consists in creating and designing sensors whose output is the relevant information, straightaway (see Fig. 1). Most of the time the raw signal no longer has to be transferred and stored. Smart sensors are a challenge in numerous fields, typically when sensors have to run autonomously in remote environments, or with limited power and storage access. For instance when studying acoustic landscapes pour environmental monitoring (forest, underwater areas, etc). IoT and wearable sensors are also targeted. Turning a sensor into a smart sensor presents a challenge at many levels. For instance, efficient AI methods to process the data need to be designed, with constraints in term of computation power and energy. Another challenge consists in building those analysis tools from small datasets, or from weakly supervised datasets. CEA is already conducting researches on those issues, and AI based methods are particularly relevant. This PhD subjects focuses on sensors recording time series: IMU, microphones, connected bracelets, etc. The core of the issue is to work on AI methods for time series, in one or several applicative fields. The PhD registers in a larger subject, namely AI reliability (anomaly detection, detection of events of class unseen during the training stage, etc.), but also the development of AI methods under labelling constraint. The topic is ambitious and several approaches are considered.

Download the offer (.zip)

Embedded AI for the semantic interpretation of a probabilistic environment model

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire Infrastructure et Ateliers Logiciels pour Puces

01-10-2020

SL-DRT-20-0291

tiana.rakotovao@cea.fr

Artificial intelligence & Data intelligence (.pdf)

The perception and modelling of an environment is a major issue when developing autonomous vehicles. How to model the surroundings of a vehicle? How to detect and identify the various obstacles? What about free spaces, and areas safe to drive on? Which sensor combination is the most appropriate to reach an exhaustive description and modelling of the environment? Those questions all have beginnings of answers, but still remain open and not yet solved. There also is a strong constraint regarding the need for embedding systems, which is one of the CEA focuses. Which processing and analysis can be considered while targeting embedded systems? Occupancy Grid is a model used to represent the surroundings of a vehicle and present various advantages. Several sensors of various modalities are used to compute the grid: each modality brings a specific information. For instance, infra-red is efficient by night, LIDAR offers a 360° field of view but is not robust to bad weather conditions, in which case a radar would be preferable. Ultrasound sensors on the contrary are used to analyse very short distances. CEA has developed approaches based on Bayesian fusion to produce SigmaFusion library. SigmaFusion is a tool to fuse the information of different sensors to produce an occupancy grid, which evolves with time. A strong point of SigmaFusion is the computing optimization: the technology is particularly efficient and competitive under strong embedded constraints (low cost integration with low energy consumption on micro-controller certified for critical task for the automotive market). An issue currently addressed is the use of EdgeAI methods to gain a semantic interpretation of an occupancy grid. A typical question is the level of knowledge and interpretation that can be reached while respecting the embedding constraint. Is it possible to detect the object evolving in a grid automatically, in real time and at low energetic cost (pedestrians, cyclists, cars, etc.)?

Download the offer (.zip)

Lensless imaging and artificial intelligence for rapid diagnosis of infections

Département Microtechnologies pour la Biologie et la Santé (LETI)

Laboratoire Systèmes d'Imagerie pour le Vivant

01-10-2020

SL-DRT-20-0518

caroline.paulus@cea.fr

Artificial intelligence & Data intelligence (.pdf)

The objective of the thesis is to develop a portable technology for pathogen identification. Indeed, in a context of spread of medical deserts and resurgence of antibiotic-resistant infections, it is urgent to develop innovative techniques for rapid diagnosis of infections in isolated regions. Among optical techniques for pathogen identification, lens free imaging methods draws attention because they are the only ones currently able to offer simultaneous characterization of a large number of colonies, all with low-cost, portable and energy-efficient technology. The objective of the thesis is to explore the potential of lensless imaging combined with artificial intelligence algorithms to identify bacterial colonies present in a biological fluid. The thesis will aim to optimize the sizing of the imaging system (sources, sensors) and to study image processing and machine learning algorithms necessary for colony identification. Two cases of clinical applications will be studied.

Download the offer (.zip)

Applied formal semantics in hardware compiler frameworks

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire composants logiciels pour la Sûreté et la Sécurité des Systèmes

01-10-2020

SL-DRT-20-0540

Mihail.Asavoae@cea.fr

Artificial intelligence & Data intelligence (.pdf)

The development of RISC-V instruction set architecture (ISA) is supported by new methodologies and tools which are dedicated to increase the productivity of hardware designs (i.e., high-level design languages and specialized compilation chains). At the language level, Chisel and FIRRTL Hardware Description Languages (HDLs) aim to raise the level of abstraction of hardware design. It thus becomes appealing to formally reason on functional and temporal properties of these high-level designs and rely on appropriate compilation extensions to transfer these high-level properties down to the level of generated Verilog, for example. In this PhD proposal, we target a formal verification framework for computer architectures to support the specification and verification of timing-related safety and security properties. The following two contributions are expected of this PhD thesis: 1) the design and implementation of a verification infrastructure based on formal executable semantics of Chisel and FIRRTL HDLs and 2) the design and implementation of an assertion language to express timing safety and security properties, which are to be verified with the aforementioned formal infrastructure. The scientific contributions of this thesis are expected to evaluated on a selection of the rich-set of architecture designs provided by the RISC-V ecosystem.

Download the offer (.zip)

Federated learning: collaboration vs. personalization

Département Métrologie Instrumentation et Information (LIST)

Laboratoire Science des Données et de la Décision

01-09-2020

SL-DRT-20-0661

aurelien.mayoue@cea.fr

Artificial intelligence & Data intelligence (.pdf)

In 2016, Google introduced the founding principles of federated learning [1] which opened up a brand new computing paradigm for AI. Until now, most of deep machine learning approaches adopt a centralized way to train a model. It requires the data to be stored in a datacenter. This is practically what giant AI companies have been doing over the years. However, this centralized approach is privacy-intrusive as users of the service have to send their data to the service provider which manages the datacenter. Federated learning is a collaborative process which leaves the training data distributed on the client devices and learns a shared model by aggregating locally-computed updates. As the data remains in its original location, the privacy is improved and the cost communication also decreases. The independent and identically distributed (IID) sampling of the training data is a key point to train a machine learning model. It ensures that the stochastic gradient is an unbiased estimate of the full gradient. But, in a decentralized learning process, it is unrealistic to assume that the local data on each edge device is always IID. ?Non-iidness? can be the cause of a significant decrease of the model accuracy. To improve the performance of federated learning whatever the local distribution of data, we investigate a way of personalizing the models at edge which permits each node to fine-tune its model locally while continuing to train the shared model. [1] Google AI blog: https://ai.googleblog.com/2017/04/federated-learning-collaborative.html

Download the offer (.zip)

Robust and distributed multi-agent-system-based data-stream learning in a collaborative environment

Département Métrologie Instrumentation et Information (LIST)

Laboratoire Intelligence Artificielle et Apprentissage Automatique

01-09-2019

SL-DRT-20-0665

sandra.garciarodriguez@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Nowadays, data streams are present in more and more applications and domains where dynamism and speed truly matter. Research challenges open lines about the generation and processing of these streams, especially in distributed, heterogeneous and collaborative environments. Existing ones lack, in general, the means of collaborating, negotiating, sharing, or validating data streams on such kind of heterogeneous environments. Multi-agent systems principles enable some of these features. However there is still work to do in order to comply them with the characteristics of data streams. The main line of this proposal is using distributed agents in data streams to deal with different challenges as managing non-synchronised streams from different sources, increasing the robustness of online models that deal with such streams (to make them reliable through unexpected environment changes) and generating new metrics to evaluate the needs mentioned before. This work would rely in the "Streamer" framework already existing in the laboratory.

Download the offer (.zip)

Resistance level modulation in PCM memory for neuromorphic applications

Département Composants Silicium (LETI)

Laboratoire de Caractérisation et Test Electrique

01-03-2020

SL-DRT-20-0740

carlo.Cagli@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Since the last 50 years, the processors are based on the von Neumann architecture and the progress in the integration on a very large scale made it possible to realize this computational architecture on a suitable technological substrate. However, today the miniaturization of electronic components is no longer sufficient to increase performance and reduce the power consumption of conventional architectures. In addition new applications, first of all the artificial intelligence, requires completely new paradighms and calculation approaches. New computational architectures inspired by biology have recently been proposed to overcome these difficulties. The main difference between a neuromorphic circuit and a classical architecture is the memory organization: networks of biological neurons are characterized by co-localization of memory (synapses) and computing centers (neurons). PCM memories are see as strong candidates for emulation of synaptic behavior but the ability to modulate their resistance level is still a challenge. This is a key point to enable PCM cell as synaptic component. This work proposal will start with a characterization phase where the ability of PCM to be modulated in different resistance levels will be scrutinized. The collected data will feed a PCM multilevel model, which is essencial to enable new circuit architecture. In a final stage, innovative circuit can be proposed, based on PCM technology, as proof of concept for neurorphic applications.

Download the offer (.zip)

Formal Specification for Machine Learning Algorithms

Département Ingénierie Logiciels et Systèmes (LIST)

Laboratoire pour la Sûreté du Logiciel

SL-DRT-20-0764

zakaria.chihani@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Machine Learning techniques, Neural Networks in particular, are going through an impressive expansion, permeating various domains from autonomous vehicles to judicial and medical assistance. This effervescence, however, may hold more than benefits, as it slowly but surely reaches critical systems such as autonomous transportation, robotics, or banking. Indeed, the remarkable efficiency of neural nets comes at a price: weakness to adversarial perturbations, lack of formal specification of desired NN properties, behavioral unpredictability... Recent efforts aim at adapting the plethora of methods that help validate "traditional" software to Machine Learning. This thesis follows this line of research by focusing on an essential component of the V&V discipline: formal specifications. In other words, under what form the properties of an AI system can be expressed so as to be readily "understood" by a computer. This thesis is complementary to another thesis that already started in our lab, that focuses on verification technologies.

Download the offer (.zip)

Framework development for qualification/certification of Artifical Intelligence based systems

Département Ingénierie Logiciels et Systèmes (LIST)

Labo.conception des systèmes embarqués et autonomes

01-04-2020

SL-DRT-20-0790

morayo.adedjouma@cea.fr

Artificial intelligence & Data intelligence (.pdf)

The open PhD position is related to qualification/ certification of autonomous systems that used Artifical intelligent (AI) components at operation time.Currently, for non-autonomous system, the safety is assessed prior to the system deployment and the safety assessment results into a safety case remains valid through life. For AI-based system, such regulatory regimes are not adequate as the system can exhibit new behavior in front of unknown situations during operation. The thesis objective is to develop incremental (dynamic) assurance case strategies and continuous certification approach with the aim to demonstrate that an sytsem using AI components may still be safe when they adapt their behavior in operation. The framework will be evaluated in a number of industry-relevant AI-based applications and scenarios.

Download the offer (.zip)

Transfer Learning and Optimal Transport applied to additive manufacturing process driving

Département Métrologie Instrumentation et Information (LIST)

Laboratoire Science des Données et de la Décision

01-10-2020

SL-DRT-20-0792

fred-maurice.ngole-mboula@cea.fr

Artificial intelligence & Data intelligence (.pdf)

This PhD thesis aims at exploring possible contributions of optimal transportation field to transfer learning through the following directions: - building a knowledge transferability criterion between a source and a target task based on the regularity of the transportation plan between the source and the target data distributions; - integrating priors on the tasks similarity through the transportation ground metric; - applying Wasserstein barycenter to multi-task learning problems. These works might find multiple practical use-cases of interest in the lab, including additive manufacturing. A more detailed presentation of this PhD thesis subject can be found via the following link: https://drive.google.com/file/d/1TmoIYeK9RKRWV7aHFGeqY48tpobuBJeB/view?usp=sharing

Download the offer (.zip)

Few-shot Learning and Domain Adaptation for Information Extraction

Département Intelligence Ambiante et Systèmes Interactifs (LIST)

Laboratoire Analyse Sémantique Textes et Images

01-10-2020

SL-DRT-20-0862

romaric.besancon@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Automated Information Extraction from texts is necessary for all applications that require the processing of large amounts of textual data, but the development of such systems, adapted to a specific domain, is still very costly because it requires to specify very precisely the information to extract or to produce lots of annotated data in order to train machine learning models. The goal of this thesis is to work on new ways to reduce the cost of domain adapation for information extraction systems. The proposed approach is based on two axes: (1) the development of a generic information extraction model, based on a model of semantic frames representing general events that can be made more generic by the use of word embeddings, such as BERT or ELMO and (2) the study of methods to automatically adapt this generic model to a new domain, exploiting only few annotated examples: this adaptation will rely on approaches such as distant supervision, active learning or transfer learning.

Download the offer (.zip)

Optimized coding technics for the design of deep neural network hardware accelerators

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire Adéquation Algorithmes Architecture

01-09-2020

SL-DRT-20-0869

johannes.thiele@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Artificial neural network based approaches have significantly improved performance in many areas such as classification, segmentation, and so on. The effectiveness of this approach is well established and the number of future applications increases. However, due to their computational complexity and their memory need, these networks are difficult to embed on low power platforms. When porting these networks on embedded platforms, a large variety of hardware constraints have to be taken into account. To overcome these difficulties several research works have produced different techniques that allow to reduce memory and computation footprint of artificial neural networks: reduction of the number of parameters, low precision quantization, etc. This thesis aims to go further into the optimization by working on the data coding. On this thesis, we proposed to explore a new method by working directly on the information coding of the neural network. This coding method would aim to unify two existing coding models: the vector model and the spike model, while keeping in perspective the hardware implementation.

Download the offer (.zip)

Transformation to simplify classification functions

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire Adéquation Algorithmes Architecture

SL-DRT-20-0871

marc.duranton@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Many of the problems addressed by Artificial Intelligence are problems of classifying complex input data into different classes. The functions transforming the complex input space into a simpler, linearly separable space are done either by learning (deep convolutional networks) or by projecting input data into a high-dimensional space in order to obtain a "rich" non-linear representation of the inputs, then having a linear matching between the high-dimensional space and the output units (the "reservoir computing" approach). These concepts are also linked to the Support Vector Machines (work of Vapnik 1966-1995). The objective of the thesis is to study this type of transformations that can be applicable for real applications, and to define an optimized, generic architecture for a given application domain, allowing data to be pre-processed in order to prepare them for a classification requiring a minimum of operations and which can, for example, be done on the fly (continuous learning). The targeted research results are multiple: - From a theoretical point of view, an approach unifying the transformations done by deep learning networks, "reservoir computing" and approaches that transform a complex input space into an essentially linearly separable space. - Define which transformations should be done in practice for a given class of problems (e. g. object recognition) by simplifying them as much as possible (depending on the error rate, false positives, etc.). - Propose optimized architectures, making the best use of advanced technologies (semiconductor, 3D stacking, photonics, etc.). The final result would be the proposal of an optimized module, which could be used as preprocessing unit, to help efficiently perform transfer learning, one shot learning and continuous learning functions for example.

Download the offer (.zip)

Distributed Memory-centric Computing architecture for AI applications, using advanced 3D and NVM Technologies

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire Intégration Silicium des Architectures Numériques

01-09-2020

SL-DRT-20-0913

ivan.miro-panades@cea.fr

Artificial intelligence & Data intelligence (.pdf)

With the revolution of AI applications, AI algorithms are getting more and more demanding in terms of computing and memory requirements, while it is envisioned that new devices implementing these AI features should be available at the ?edge?, meaning close to the final user (portable devices, automotive, IoT, etc) and not anymore only in the cloud. This implies very strong requirement in terms of memory capacity to enable learning at the edge with compliant energy efficiency. The PhD consists in proposing and exploring new Distributed Memory-centric Computing architecture for AI applications, using advanced technologies to overcome the current issues (memory capacity, memory bandwidth, energy per inference, and learning capability). Recent Non Volatile Memory (NVM) technologies and 3D integration technologies offer dense memory integration while bringing the memory closer to computing cores. The architecture challenge consists in defining the adequate system partitioning, the distributed communication mechanisms, the memory/computing ratios requirements, to in-fine obtain the targeted distributed memory centric computing architecture. The work will consist in architecture and application system exploration through system modeling, and may lead to a testchip to validate the proposed concept. The PhD will take place in an active collaboration between University of Stanford (CA, USA) and CEA (Grenoble, France).

Download the offer (.zip)

Digital twin in structural health monitoring for aerospace using machine learning

Département Imagerie Simulation pour le Contrôle (LIST)

Laboratoire Méthodes CND

01-06-2020

SL-DRT-20-0938

olivier.mesnil@cea.fr

Artificial intelligence & Data intelligence (.pdf)

This PhD is funded by the Marie Curie program of European Union through the innovative training network (ITN) GW4SHM on Guided Waves for Structural Health Monitoring. Structural Health Monitoring (SHM) relies on the permanent integration of sensors to create smart structures by enabling the monitoring of structural health over time. In the aerospace industry, an SHM system installed on an aircraft or a reusable launcher would provide a prognostic of the residual life of the structure, leading to safer and cheaper structures through a customized maintenance plan. In SHM, the use of Guided elastic Wave (GW) is particularly promising in the aerospace industry due to the limited added mass of the sensors and high defect sensitivity. Typically in GW-SHM, a sensor network is instrumented on the structure in order to generate and measure propagated elastic waves, carrying information on invisible but potentially harmful defect(s). Advanced post-processing techniques can then be employed to reliably detect, locate and quantify the defect(s). Despites its promises and relative maturity, the field of GW-SHM does not currently have any large scale application in the aerospace industry. The underlying reason is the very high sensitivity of the GW to a large set of parameters, including of course the presence of the defect, but also the local micro structure of the propagating medium, the environmental and operating conditions (such as temperature or loads) or any geometric singularity. The interpretation of GW signals outside a laboratory environment on a known sample is therefore a challenge. Moreover, in order to demonstrate the performance of SHM towards the certification of such systems, extensive simulation campaigns must be conducted to study the results in a large set of conditions and avoid prohibitively expensive experimental trials, which requires a digital twin (i.e. a faithful digital replica of a structure instrumented by an SHM system). However, the digital twin concept only holds in SHM if all the influencing parameters are fully known, which is barely possible for the simplest configurations in well-controlled environments. The goal of this thesis is to expand the concept of digital twin for GW-SHM in order to enable the use of simulation for the reliable interpretation of GW-SHM data in uncontrolled experimental conditions for the detection and characterization of flaws. To achieve this goal, simulation tools developed at CEA-List and integrated in the CIVA software will be extensively used to generate a set of simulations describing a not-fully known experiment. Machine learning tools will then trained to exploit this a priori knowledge in order to extract the defect(s) signature(s) from the signals, enabling the application of diagnostic processes such as guided wave imaging. Reliability of the developed methodology will be of interest to demonstrate its efficiency in representative conditions. The process will then be tested and validated for the SHM of advanced aeronautics composites structures in realistic environments. Working Context The PhD thesis will be hosted by CEA-List at Saclay, France (20km from Paris) with a secondment at Safran Tech at Magny-Les-Hameaux (7km from Saclay) to study the reliability of the developed approach on realistic use-cases. At CEA-List, the department of imaging and simulation for nondestructive testing (DISC) is a department of about 120 people dedicated to the R&D related to simulation, instrumentation and methods for nondestructive testing and SHM. In SHM, activities revolve around novel instrumentations, integrated SHM systems, passive monitoring techniques, simulation, performance demonstration and artificial intelligence, with applications in multiple industries including aerospace, nuclear, petrochemical or land transportation. Safran inaugurated on January 2015, Safran's new Research & Technology Center, Safran Tech, at France's leading science and technology cluster, in Saclay, near Paris. The Safran Tech center houses a research workforce of around 500 scientists and engineers organized in six research units and four platforms. The research units are covering energy & propulsion, materials & processes, sensors, electronics, information & communication technologies, and digital simulation for engineering. Eligibility criteria Marie Sklodowska-Curie Actions Eligibility rules apply. Particularly: Mobility Rule: The researcher (any nationality) must not have resided or carried out his/her main activity (work, studies, etc.) in France for more than 12 months in the 3 years immediately prior to his/her recruitment (excluding vacations and short stays). Early Stage Researchers (ESR): ESR shall, at the date of recruitment, be within the first four years of full-time equivalent research experience(*) and have not been awarded a doctoral degree. The 4-years period is measured from the date when the researcher obtained the degree which formally entitle him/her to embark on a doctorate either in the country in which the degree was obtained or the country in which the researcher is recruited. (*) Research and development activities (R&D) in any institution (companies, research organizations, government agencies, universities, etc.) are considered as research experience. Teaching in universities is also considered as research experience since it's part of the research & academic career. Mandatory skills: - Scientific curiosity and motivation for challenging tasks - M2 or equivalent in one of the following area (or related fields) o Mechanical engineering ? elastic or acoustic waves o Machine learning ? data analysis ? signal processing Optional skills: - Hands-on experience in machine learning and data analysis is a plus - Experience in a laboratory and data acquisition

Download the offer (.zip)

Compact Neural Network topology learning for in-sensor inference

Département d'Optronique (LETI)

Laboratoire Architecture Systèmes Photoniques

01-10-2020

SL-DRT-20-0951

william.guicquero@cea.fr

Artificial intelligence & Data intelligence (.pdf)

This thesis will be as a ?CIFRE? thesis as a collaboration between ST Microelectronics and CEA-Leti, in the context of the Grenoble Multidisciplinary Institute in Artificial Intelligence (MIAI). This is part of a collaborative project between Gipsa-Lab, Tima-Lab, ST and CEA-Leti, called ?Near-Sensor Neural Computing?. This specific project envisions three different theses (i.e., 2 other thesis subjects). The goal of this thesis is to provide methods and tools for producing a network topology given as priors the hardware resources available as well as the task (detection/classification) to perform and the image database used for training and test. The topology optimization will benefit from recent advances in hardware-friendly neural network design namely dimensionality reduction via compressive sensing, low quantization (e.g., binary or ternary) of the weights and activations as well as the dynamic selective execution of network parts. The technical inputs such as use cases and hardware-related limitations will be provided by ST; carefully taking into account its product and market knowledge. On the other hand, the CEA supervision will provide a solid background on machine learning for hardware-oriented design. Among the expected outcomes of this work is a demonstrator of such network topology generator for a given task with the tests demonstrating the validity of the approach.

Download the offer (.zip)

Online characterization and classification of radiological signals using embedded machine learning

Département Métrologie Instrumentation et Information (LIST)

Laboratoire Capteurs et Architectures Electroniques

01-10-2019

SL-DRT-20-0956

gwenole.corre@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Machine learning classification methods have become ubiquitous in the fields of signal and image processing. Nevertheless, these classification methods remain very little used today in the fields of embedded applications. In fact, several studies have shown that machine learning classification methods provide satisfactory performance in classifying signals received from radiological sensors. However, most of these studies and solutions have been developed and experimented offline, and there are few or no real-time neural network learning based solutions. The aim of the thesis is to propose learning methods that can be embedded in real-time portable measurement devices. The proposed methods have to meet the radiological signal classification constraints and to enhance the performance of these measurement devices.

Download the offer (.zip)

Crowdsensing for identifying sources of air pollution using a network of low-cost multimodal gas sensors

Département Systèmes (LETI)

Laboratoire Signaux et Systèmes de Capteurs

SL-DRT-20-0966

sylvain.leirens@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Crowdsensing for the identification of air pollution sources relies on the use of a large number of low-cost multimodal gas sensors. The subject of the thesis covers the implementation of this network of sensors for the identification of high emitters in urban road traffic. It is now recognized that a small number of highly polluting vehicles is responsible for a large fraction of the pollution generated by the road traffic, while urban air quality is becoming a growing concern in Europe. The thesis work will consist in formulating and solving a problem of separation and localization of mobile air pollution sources in urban environment with a network of mobile sensors. Measurement campaigns will be carried out to validate experimentally the approach developed in the thesis.

Download the offer (.zip)

Machine learning methods using uncertain labels, human stress estimation application

Département Systèmes (LETI)

Laboratoire Signaux et Systèmes de Capteurs

01-10-2019

SL-DRT-20-0981

christelle.godin@cea.fr

Artificial intelligence & Data intelligence (.pdf)

With wearable sensors development, it is now possible to monitor physiological parameters and activity. Many studies show that stress level assessment can be done by using those measurements. Supervised machine learning methods used for it are flourishing. They suppose that for each measurement, a ?ground truth? stress level is available. However, while doing experiments for database construction it is not possible to attribute an exact stress label to each event but subjective values are easily available. The goal of the PhD is to take into account data with uncertain, fuzzy, redundant, contradictory or missing values in order to obtain a better stress estimator. This kind of approach should be useful for a lot of applications including other mental state estimation like drowsiness detection, mental disease diagnosis, emotion estimation.

Download the offer (.zip)

Detection and classification of emotional state change using machine learning.

Département Systèmes (LETI)

Laboratoire Signaux et Systèmes de Capteurs

01-10-2020

SL-DRT-20-1028

vincent.heiries@cea.fr

Artificial intelligence & Data intelligence (.pdf)

This PhD work focuses on the analysis of human stress by sensor data fusion and statistical methods. There are several types of stress affecting humans, including chronic stress and acute stress. In this study, we will focus on acute stress or peak stress (panic attack) which is problematic because it can be traumatic, and especially disabling / inhibiting during a crucial task to be performed. Acute stress occurs when the stressor appears suddenly and provokes a rapid adaptation response. The issue raised by this subject is therefore the detection or even anticipation of the occurrence of a stress peak which can be considered as an unusual, relatively rare event of variable amplitude. In terms of signal processing methods, this work may use specific event-driven processing methods. Moreover, the detection and classification of these events leading to a stress peak will be obtained through a fusion of multi-sensor multi-physical data. These sensors may be signal sensors worn by the person (ECG, PPG, EDA, breathing sensor, accelerometer, etc.) or sensors monitoring the individual remotely (video facial analysis, semantic field of speech, prosody). Algorithms allowing the classification of peak stress events from the signals obtained by this set of sensors will be developed from the family of machine learning methods (for example, non-limiting: CNN, RNN, or Deep Q-learning type learning algorithms). The applications of this stress detection and classification system are broad, ranging for example from the surgeon to the industrial operator subjected to a high mental load, from the air traffic controller to the firefighter, or in general to any person subjected to a crucial task to be performed and for which any loss of capacity could be critical. This system will pave the way for the implementation of appropriate stress management adapted to each individual.

Download the offer (.zip)

Human-in-the-Loop Learning and Adaptation under Uncertain and Unpredictable Situations in AI-based Autonomous Systems

Département Ingénierie Logiciels et Systèmes (LIST)

Labo.conception des systèmes embarqués et autonomes

01-10-2020

SL-DRT-20-1108

huascar.espinoza@cea.fr

Artificial intelligence & Data intelligence (.pdf)

Autonomous systems are evolving towards self-adaptive systems, being boosted by Artificial Intelligence (AI) techniques such as Machine/Deep Learning (M/DL). The emergence of autonomy means that software has to operate in an open and highly dynamic world, being capable of adapting themselves autonomously at run time to new environment conditions or unpredictable situations. In particular, this thesis aims to explore the combination of the capabilities of humans and algorithms to detect uncertainty regions and avoid dangerous situations in the real world and transfer control between a machine and a human (or to the safest agent). Learning-enabled systems based on deep learning are first trained in simulation environments before deploying them in the real world. While simulators are providing increasingly realistic training environments, there is always a gap between simulation and training, because training data does not capture some feature spaces and the model does not learn about them due to the incompleteness of the simulator to reflect the complexity of the real world. Our goal is to find methods for detecting unknown unknowns by combining simulation training with human input from demonstration data.

Download the offer (.zip)

See all positions