Data science research
Research activities: data analysis using artificial intelligence
The research activities of the Data Science department revolve around the following themes:
- User-centered AI, e.g. explainable AI, learning from few examples, information quality, data and algorithm security
- Multimodal and heterogeneous data processing and fusion, e.g., medical image analysis, text analysis, interaction analysis, remote sensing, time series, anomaly detection
- Decision support, e.g., combinatorial and multi-objective optimization, multi-criteria decision support, preference modeling.
The department comprises 7 teams from the 2 laboratories Lab-STICC (UMR CNRS 6285) and LaTIM (UMR 1101 INSERM):
- DECIDE (Lab-STICC), which provides decision-support solutions for decision-makers confronted with heterogeneous and complex data, through disciplines such as data and decision science or geographic information science.
- M3 (Lab-STICC), which develops technologies for the interpretation and fusion of heterogeneous marine data for accurate mapping of the marine environment.
- Matrix (Lab-STICC), which develops learning and modeling methods to extract information from complex temporal and spatial data.
- Cyber Health (LATIM), which protects and enhances the value of medical data throughout its life cycle, thanks to secure and innovative processing techniques.
- Vision (LATIM), which works on the development of optical devices and AI algorithms to improve human vision, both healthy and pathological.
- Action (LATIM), which optimizes multimodal imaging in oncology to improve diagnosis, treatment selection and radiotherapy.
- Imagine (LATIM), which develops innovative imaging and rehabilitation technologies to improve the lives of patients suffering from neuro-musculo-skeletal diseases.
In healthcare, cybersecurity is at the heart of the challenges of artificial intelligence (AI) empowered by massive multiscale data and new medical practices. It is also imposed by means of a complex broad set of strict ethic and legislative regulations. On one hand, the security of data should be ensured whatever their nature, transmission, processing and transformations they went through. On the other hand, methods that will be applied to exploit these data must themselves be safe. Healthcare AI systems are considered as of high risk by EU.
The healthcare sector (public and private) generates a vast amount of data from various sources, including electronic health records, advanced imaging techniques, high throughput sequencing, wearable devices, and population health data. The use of massive datasets, or “big data,” analyzed using sophisticated machine learning algorithms, has the potential to inform the development of more effective and personalized treatments, interventions, and policies, and to improve healthcare delivery and outcomes.However, the sensitive nature of personal health data, cybersecurity risks, biases in the data, and the lack of robustness of machine learning algorithms are all factors that currently limit the widespread use and exploitation of this data. These limitations thus hinder the potential benefits that can be obtained from massive health data analysis for the individuals and society.
The industry of the future invites itself into the metaverse
The metaverse is no longer limited to online gaming or social interaction: it is becoming a key technology for industry. Through the 5GMetaverse project, five Institut Mines-Télécom schools are seeking to adapt tomorrow's networks to the needs of augmented and virtual reality. In particular, the project aims to develop concrete solutions for industrial process optimization, remote assistance and man-machine collaboration.
This project aims to revolutionize data storage using DNA, offering long-term storage stability and ease of use. Faced with competition from North American consortia, the aim is to make French research an international leader in polymer-based storage. The project is spread over 7 years, with a budget of 20 million euros. It focuses on three major challenges: DNA synthesis, information compression and the exploration of new polymers. The 5-year objective is to achieve a DNA read/write cycle of 1 bit per second, enabling 10 GB of data to be stored in 24 hours. Demonstrators will be created for archiving, marking, computing and molecular engineering applications. The CNRS is coordinating this project involving over 20 interdisciplinary laboratories, with the aim of strengthening France's position in the national and European research communities.
This project aims to deepen our understanding of the role of the cerebellum in social cognition and social-emotional behavior, by studying patients with autosomal dominant spinocerebellar ataxias (SCA). The cerebellum, long associated with motor skills, is now recognized for its cognitive and emotional functions. Using advanced neuroimaging techniques and comprehensive neuropsychological assessment, we hope to characterize social-cognitive deficits in SCA and identify structural and functional abnormalities of the cerebellum. This multidisciplinary project could lead to non-pharmacological clinical interventions and offer psycho-educational support to patients and their families, filling an important gap in the management of ACS.
Intelligent orthopedic implants offer exciting prospects, particularly for improving post-surgical follow-up. Today, however, the technologies available for their power supply are not suited to powering the all-metal prostheses used in orthopedics. The Piezoknee project aims to exploit a wireless power transmission solution based on acoustic waves (APT: Acoustic Power Transmission) to power an intelligent knee implant (SOI: Smart Orthopedic Implant) capable of transmitting physiological information (temperature, pH, mechanical stress) to help prevent any complications.
While machine learning (ML) can lead to great advancements with respect to digital services and applications in the field of medicine, the training process based on real medical patient data is blocked by the fact that uncontrolled access to and exposure of such assets is not allowed by data protection legislation. The EU-funded PAROMA-MED project aims to develop novel technologies, tools, services and architectures for patients, health professionals, data scientists and health domain businesses so that they will be able to interact in the context of data and ML federations according to legal constraints and with complete respect to data owners rights from privacy protection to fine grained governance, without performance and functionality penalties of ML/AI workflows and applications.