Biography | Teaching | Research | Publications | Talks | Projects | Collaborations | Home
Paris AIoT
About
Along
with 18 other faculty members from 3 departments, we have launched the Paris
AIoT initiative, a collaboration between Telecom Paris and member industrial
companies to educate and train world-class researchers and PhDs to enable truly
ubiquitous sensing, computing and communication with fully distributed AI and
embedded intelligence. The Paris AIoT emphasizes application-oriented and system-oriented
research, its areas of interest include software and hardware at all levels of
the system stack offering innovative and pioneering solutions on the
convergence of AI and IoT. It represents a real strategic opportunity for
university-industry-government research partnerships in France and Europe in
the key areas of personalized health, autonomous driving, 6G, global mentoring
and personalized project-based education. It will forge close relationships
with industry leaders to facilitate rapid technology transfer and provide an
environment for research into
core issues of future generations of AIoT systems as well as a
platform to develop personalized and project-based training programs for MSc
and PhD students. The Paris AIoT consists of MSc and PhD students, post-doctoral
researchers and professors engaged in a pre-competitive research program that
is continuously evolving according to the interests of faculty members and our
industrial partners. Our program will build on the unique strengths of both
university and industry to enhance the productivity and competitiveness of
both. Research areas of interest include:
1.
AI for Training
Data Efficiency (Augmentation and Frugality) - Adequate
Development of Training Data;
2.
AI Defined
Network (Centralized, Hybrid and Distributed) and Distributed Intelligence;
3.
Energy and Memory
Efficient AI and Algorithms for IoT devices;
4.
Energy Efficient
and Hardware Accelerator, Smart Sensors and Low-Power Cognitive Connectivity;
5.
AI, Algorithms
and Encryption for Privacy and Security.
Scientific
Context
The vision of a massively digitized economy and society calls for
intelligent connectivity and service provision across a huge number of
heterogeneous domains, resources, and with an unlimited number of application
requirements. These covers the realization of a
unified and open communication and computing architecture beyond the current SoA. Such architecture will enable seamless operations and
service execution across a multiplicity of heterogeneous domains,
infrastructures, services, business, and application heterogeneous domains,
whilst providing secure and reliable scalability towards an unlimited number of
application requirements, hence paving the way towards massive digitization.
However, the massive adoption
of AI tools will exacerbate the problem of energy consumption of the ICT
infrastructure. Native
integration of AI/ML is in scope to implement adaptive decision making at
different time scales with expected impact on energy and performance efficiency
gains for such distributed multi-stakeholders’ systems. The adoption of these tools may trigger changes in the
existing architectures. Therefore, it will be crucial to devise energy
efficient architectures and computation algorithms to have energetically
sustainable communication and computing paradigms for future mobile
networks that adequately explore artificial intelligence
technologies. Research
needs to be done on: 1) distributed edge AI solutions, covering consensus
convergence, resource limitations, localized data management, transfer
learning; 2) adequate development of training data for telecommunications; 3)
AI security and comprehensibility of ML for the applications identified above;
4) strongly distributed AI/ML instrumentation integrated at every layer with an
end-to-end perspective.
The “softwarization” of network
components and functions is a steadily adopted strategy. Operators have started
to abandon the static, expensive proprietary hardware to adopt a more flexible
approach based on Virtual Network Functions (VNFs). This novel Network Function
Virtualization (NFV) paradigm advocates implementing network middleboxes (e.g.,
forwarding devices, firewalls, IDS, etc.) as pieces of software to be deployed
and executed on commercial off-the-shelf (COTS) hardware, usually on remote
cloud systems. A similar trend can be observed in 5G wireless networks, where
the traditional radio access is moving to cloud RAN or virtualized RAN running
on general-purpose computing and radio hardware.
Current IoT
architecture emphasizes centralized information processing, where the connected
objects mostly serve as data collection nodes. As the number of connected objects
increases exponentially, the centralized architecture of IoT networks suffers
from major bottlenecks, including connectivity
(interference), in latency (distant
servers), in energy consumption (cost
of distant communication), in centralized
data processing (data flooding) and in centralized
network management (number of objects). To address these bottlenecks, at
Paris AIoT Research Center, we propose a massive adoption of AI tools in the whole end to end system. Native integration of AI/ML is to implement adaptive decision
making at different time scales with expected impact on energy and performance
efficiency gains for such fully distributed systems. The adoption of these tools will trigger changes in the existing
architectures. Therefore, it will be crucial to devise energy efficient
architectures and computation algorithms to have energetically sustainable
communication and computing paradigms for future mobile networks that
adequately explore artificial intelligence technologies. In this direction, the network
management and computation will be fully decentralized, and the
connected objects (the nodes) will become active elements and perform
distributed tasks within the network including network sensing, connectivity
decision, security verification, data compression, data pruning and distributed
calculation. We will empower the intelligence of the individual nodes by
embedding them with Artificial Intelligence (AI) integrated with game-theoretic
decision analytics. AI embedded into IoT
nodes will improve security, privacy,
and help reduce latency and bandwidth.
Our proposed solution based on AIoT not only brings the compute closer to where
the data are generated, but adds the intelligence needed to improve robustness, reliability, efficiency,
availability and productivity.
Organization
Global
Mentoring and Talent Sourcing
In this model,
mentors develop parallel project-based learning programs abroad and help their
former PhD students or close collaborators (i.e., principal investigators) to
build a research center of excellence. The mentors (i.e., directors) and the
principal investigators will subsequently serve as remote mentors and internal
mentors, respectively, for young researchers and students. Additionally, the
mentors can provide direct mentorship to the scholars/students abroad by
hosting them in their own laboratories or by traveling periodically. In this
way, emerging scholars benefit not only from the mentorship provided by the
principal investigators on a daily basis, but also from the mentors over the
course of their training. The result is a sustainable model that ensures the
continuous transfer of knowledge and experience from one generation of
researchers to the next.
Outgoing Visiting
Researcher Fellowship
Incoming Visiting
Researcher Fellowship
Faculty Members
Presentations and
Documents : please email to Paris-AIoT@telecom-paris.fr
Télécom Paris is member of
