PhD Position F/M Enhancing AI-driven Predictive Analysis in Dynamic Contact

7 Feb 2024
Job Information

Organisation/Company

Inria

Research Field

Computer science

Researcher Profile

Recognised Researcher (R2)

Country

France

Application Deadline

7 Mar 2024 - 00:00 (UTC)

Type of Contract

To be defined

Job Status

Full-time

Hours Per Week

To be defined

Is the job funded through the EU Research Framework Programme?

Not funded by an EU programme

Is the Job related to staff position within a Research Infrastructure?

No

Offer Description

A propos du centre ou de la direction fonctionnelle

The Inria research centre in Lyon is the 9th Inria research centre, formally created in January 2022. It brings together approximately 300 people in 16 research teams and research support services.

Its staff are distributed at this stage on 2 campuses: in Villeurbanne La Doua (Centre / INSA Lyon / UCBL) on the one hand, and Lyon Gerland (ENS de Lyon) on the other.

The Lyon centre is active in the fields of software, distributed and high-performance computing, embedded systems, quantum computing and privacy in the digital world, but also in digital health and computational biology.

Contexte et atouts du poste

The doctoral position at Inria's AGORA research group, located at the La Doua Campus in Lyon, offers a unique opportunity to collaborate with esteemed experts such as Prof. Hervé Rivano, Prof. Razvan Stanica, and Dr. Juan Fraire. The appointee will use advanced software and simulators, enhancing their expertise in communication systems, wireless sensor networks, and urban network planning. This role is enriched by AGORA's strong international and academic-industrial collaborations. It offers the chance to delve into the smart city and satellite domains, exploring technologies pivotal to wireless sensor networks and massive machine-to-machine communications. This position is a gateway to cutting-edge research and professional growth in IoT and networking.

Mission confiée

Overview

This research delves into the dynamic contact networks within bike-to-infrastructure and Low-Earth Orbit (LEO) satellite systems.

In this context, a contact is a time-bound interaction allowing data exchange between two nodes, characterized by parameters like duration, nodes involved, signal attributes, and reliability metrics. A contact set encapsulates these interactions over time, forming a network topology that can be either a forward-looking contact plan or a historical contact trace . Central to this study is the hypothesis that AI can effectively translate historical contact traces into predictive contact plans . The research will leverage AI to analyze temporal patterns in contact data, aiming for accurate predictions of future network interactions.

This project will focus on two use cases: urban bike-to-infrastructure networks with complex, ever-changing data streams and LEO satellite networks characterized by their vast, evolving mega-constellations.

Use Cases

LEO Satellite Networks : Emerging Low Earth Orbit (LEO) Satellite Networks are set to revolutionize Earth Observation and broadband communication systems. Initially conceptualized in the early 1990s, these networks have gained renewed attention with the advent of Mega-Constellations. The constantly shifting satellite positions result in a continuously evolving network topology, traditionally represented as a sequence of 'snapshots' - stable network configurations over short periods, each comprising a set of contacts. Comprising numerous satellites interconnected through Inter-Satellite Links (ISLs), these networks facilitate low-latency, high-capacity communication vital for broadband services and EO missions. The dynamic motion of satellites in LEO presents unique challenges in maintaining effective communication links, as the ISLs represent temporal contacts with finite lifetimes due to continuous satellite movement. The challenge lies in efficiently predicting and managing these dynamic topologies to ensure uninterrupted and reliable data exchange, a vital component for the success of autonomous, heterogeneous satellite networks.

Bike-to-Infrastructure Connectivity: With the rapid urbanization since the mid-20th century, cities have witnessed a significant increase in challenges like traffic congestion, pollution, and unhealthy lifestyles. Biking has emerged as a sustainable alternative, leading to a worldwide trend in urban biking development. Bike-sharing systems, in particular, have grown tremendously, with thousands of bikes in over a thousand cities globally. This shift towards intelligent transport systems (ITS) has given rise to innovative approaches like the "Internet of Bikes" (IoB-DTN). IoB-DTN is a Delay Tolerant Network (DTN) protocol tailored for Internet of Things (IoT) applications in urban bike-sharing systems. It emphasizes data aggregation, leveraging the mobility of bikes to collect and relay data efficiently. The protocol explores spatial, temporal, and spatiotemporal data aggregation strategies, aiming to optimize network throughput and reduce energy consumption in data transmission. In this context, the project will focus on advanced contact modeling to understand and predict the interactions between bikes and urban infrastructure to enhance urban mobility experiences.

Objectives

General Objective : To enhance predictive analysis in dynamic contact networks through AI-driven methodologies, focusing on two distinct use cases: urban bike-to-infrastructure networks and LEO satellite systems. The aim is to develop and validate AI models that can effectively predict future network interactions by analyzing historical contact data and optimizing network efficiency and reliability .

Specific Objectives:

• Data Collection and Processing : Gather extensive contact trace data from urban bike test benches and satellite orbital datasets. Process and prepare the data for analysis, ensuring it is suitable for AI modeling.
• Development of Predictive Models : Utilize time series analysis and forecasting methods to predict future contacts in both use cases. Implement Graph Neural Networks (GNNs) to model the complex relationships between network nodes.
• Optimization and Routing : Apply Reinforcement Learning (RL) techniques to optimize network routing and decision-making processes. Develop clustering and community detection models to categorize bikes or satellites based on behavioral patterns.
• AI Algorithm Design and Implementation : Create and refine AI algorithms, focusing on predictive analytics and feature recognition specific to each use case. Integrate different ML techniques, including RNNs, LSTMs, GCNs, and RL methods, to address the unique challenges of each network.
• Model Validation and Testing : Test the developed AI models for bike-to-infrastructure and satellite networks within realistic scenarios. Evaluate the performance of the models against key metrics such as accuracy, efficiency, and reliability.
• Scholarly Dissemination : Publish research findings in premier academic conferences and journals. Share insights and methodologies developed during the project with the broader AI and network research communities.
• Application and Impact Assessment : Assess the practical applications of the research in enhancing urban mobility and satellite communication systems. Evaluate the potential societal and environmental impacts of improved network management and efficiency.

Methodology

The research plan considers a range of Machine Learning (ML) solutions to address the challenges of predictive analysis in dynamic contact networks, specifically for bike-to-infrastructure and LEO satellite networks. Below is A list of potential ML approaches and how they could be applied to the project.

• Time Series Analysis and Forecasting
• Models : Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs).
• Application : These models are well-suited for predicting future contacts based on temporal patterns in historical data. They can effectively handle sequential data, making them ideal for time-dependent contact trace analysis.
• Graph Neural Networks (GNNs):
• Models : Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs).
• Application : GNNs can model the complex relationships and interdependencies between different nodes in a network, which is crucial for understanding the dynamics of satellite networks and urban bike infrastructures.
• Reinforcement Learning (RL):
• Models : Deep Q-Networks (DQN), Policy Gradient methods, Actor-Critic models.
• Application : RL can optimize decision-making processes, such as routing in bike-to-infrastructure networks and satellite network operations, by learning the best actions to take in various states of the network.
• Clustering and Community Detection:
• Models : K-means, Hierarchical Clustering, DBSCAN, Louvain method for community detection.
• Application : These methods can segment bikes or satellites into groups based on similar characteristics or behaviors, aiding in the management and optimization of the networks.

References

Ruiz-De-Azua, Joan A., et al. "Assessment of satellite contacts using predictive algorithms for autonomous satellite networks." IEEE Access 8 (2020): 100732-100748. ( link )

Fontanesi, G., et al. "Artificial Intelligence for Satellite Communication and Non-Terrestrial Networks: A Survey." arXiv preprint arXiv:2304.13008 (2023). ( link )

Müller, Kevin. "Building Contact Graphs for Large-scale Constellations." Bachelor's thesis, Saarland University, 2023. ( link )

Magnana L. et al. "Implicit GPS-based bicycle route choice model using clustering methods and an LSTM network" PLoS ONE, 2022 ( link )

Delaine F. et al. "Rendez-vous Based Drift Diagnosis Algorithm For Sensor Networks Towards In Situ Calibration" IEEE Transactions on Automation Science and Engineering, 2022 ( link )

Zguira Y., et al. "Internet of Bikes: A DTN Protocol with Data Aggregation for Urban Data Collection" Sensors, 2018 ( link )

Principales activités

Data Collection : Acquiring extensive contact traces through connected bike test benches and satellite orbital data.

Predictive Modeling : Generating large-scale contact plans, utilizing AI to process and predict network dynamics.

Data Structuring : Creating an optimized data structure for contact sets that will feed into the AI models.

AI Development : Designing innovative AI algorithms for both domains, focusing on predictive analytics and feature recognition.

Model Validation : Testing the AI models within realistic network scenarios for bike-to-infrastructure and satellite communications.

Scholarly Contribution : Publishing findings in leading conferences and journals, contributing novel insights to AI in dynamic contact networks.

Compétences

We encourage applications from researchers with a Computer Science or Computer Engineering profile. Practical proficiency with programming languages (C/C++ and Python) is desirable. A solid understanding of mathematics, informatics, and mobile wireless networking is also preferred. Applicants must have fluency in English; proficiency in French is not a prerequisite but would be advantageous. We are seeking candidates who are empathetic, proactive, and self-motivated.

Avantages

• Subsidized meals
• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking (after 6 months of employment) and flexible organization of working hours
• Professional equipment available (videoconferencing, loan of computer equipment, etc.)
• Social, cultural and sports events and activities
• Access to vocational training
• Social security coverage

Rémunération

1st and 2nd year: 2100 euros gross salary /month

3rd year: 2190 euros gross salary / month

Requirements
Additional Information
Work Location(s)

Number of offers available

1

Company/Institute

Inria

Country

France

Geofield

Where to apply

Website

https://illbeback.ai/job/phd-position-f-m-enhancing-ai-driven-predictive-analys…

STATUS: EXPIRED