진학프로석·박사 채용 접수 플랫폼

Université Gustave Eiffel is looking for a candidate to apply for a Postdoctoral Fellowship in the framework of the Marie-Sklodowska Curie Programme 2025.

The Candidate and Université Gustave Eiffel's supervisor will apply together to develop the following research project : AI Foundation Models for Multimodal Mobility Planning and Management

1. Context and Summary 

a- Context

Recent advancements in Artificial Intelligence (AI) have showcased the transformative potential of transformer-based architectures, driving breakthroughs in fields such as natural language processing (NLP) and computer vision. These architectures serve as the backbone of foundation models (FMs), which excel in solving fundamental and general tasks through pre-training on massive datasets and fine-tuning for specific applications. Renowned for their transferability, generalizability, and user-friendliness, foundation models enable unprecedented capabilities, from understanding complex relationships in multimodal data to adapting seamlessly across domains with minimal additional training. Their ability to unify diverse tasks and deliver interpretable, actionable solutions makes them a compelling choice for tackling challenges in domains like transport network planning and management, where the integration of diverse data sources and decision-making complexity have long been limiting factors.

However, despite their success, the application of FMs in domains requiring multimodal, spatio-temporal, and non-Euclidean data processing—such as Geographic Information Systems (GIS) and transport networks—remains underexplored. The challenges include computational complexity, the integration of heterogeneous data sources, and ethical considerations such as equity, data privacy, and bias. Despite these hurdles, the increasing availability of open-source pre-trained FMs offers promising avenues for researchers to leverage existing open-source models with feasible fine-tuning effort, potentially unlocking new possibilities in these complex domains (Balsebre et al., 2024; Li et al., 2024). Particularly the reasoning capability of LLMs combined with prediction and generative power of FMs on GIS and graph provide exciting possibilities (Liu et al., 2023; Zhang et al., 2024).

This research aims to address these challenges by developing and deploying AI foundation models tailored for multimodal mobility planning and management. Leveraging datasets from telecommunication, public transport, and GPS sources—enabled through partnerships with academic institutions and industrial stakeholders—the project seeks to revolutionize transport planning with actionable insights for sustainability, resilience, accessibility, and equity.

Objectives:
1. Data Fusion: Integrate multimodal mobility datasets to better understand travel behavior and traffic evolution.
2. AI-Powered Multi-modal Mobility Planning and Management: Develop an intelligent multimodal mobility planning and management system as a decision-making advisor with reasoning and interactive capabilities.
3. Scalability and Transferability: Assess the transferability of the AI models across different geographic and infrastructural contexts.

Potential Applications:
• Detecting bottlenecks and underutilized transport network capacity to enable targeted interventions for monitoring, management, and infrastructure planning.
• Providing data-driven solutions to optimize mobility infrastructure, including parking infrastructure, public transport routes and schedule optimization, traffic control, etc.
• Detecting and remedying accessibility issues.
• Classifying traffic patterns and multimodal transport network regimes, as well as detecting and anticipating anomalous situations.

Research Questions:
• How can multi-source, heterogeneous data be fused effectively for enhanced decision-making in mobility systems?
• What is the potential of FMs and LLMs for managing mobility under limited data conditions?
• What technical resources and methodologies are required to deploy these models in real-world scenarios?
• How can we address ethical risks, such as data privacy and bias, to ensure equitable AI solutions for transport planning?

Proposed Methodology:
1. Data Collection and Preprocessing: Identify and preprocess multimodal datasets (telecom, GPS, public transport data).
2. Model Selection and Adaptation: Identify suitable FMs (e.g., UrbanGPT, graph-based FMs) and fine-tune them for mobility-related tasks.
3. Model Evaluation: Iteratively refine the models for performance, accuracy, and generalizability, including zero-shot learning scenarios.
4. Case Study Application: Demonstrate the system’s capabilities in real-world scenarios, focusing on Lyon and US cities.
5. Industry Collaboration: Collaborate with a leading mobile phone operator, Orange, as well as metropolitan transport municipalities and operators to explore the transferability of research outcomes to industrial applications.

Expected Contributions:
• Advancement in spatio-temporal data modeling using AI.
• Novel tools and frameworks for sustainable urban mobility planning.
• Practical solutions for bottleneck detection, capacity optimization, anomaly detection and equitable, resilient and sustainable transport system design and management.

b- Summary

Position Overview:

The University Gustave Eiffel (France), in collaboration with the University of California, Berkeley (USA), and Orange (France), invites applications for a 24–36-month postdoctoral fellowship funded by the Marie Skłodowska-Curie Actions (MSCA) Global Postdoctoral Fellowship program. This unique opportunity allows the fellow to engage in cutting-edge research on the application of AI foundation models (FMs) for multimodal mobility planning and management, while gaining invaluable interdisciplinary, international, and inter-sectoral experience.

This fellowship brings together leading expertise from academia and industry. Candidates will benefit from a collaborative framework that combines cutting-edge research at University Gustave Eiffel and UC Berkeley with hands-on industrial experience at Orange. This unique synergy fosters interdisciplinary growth and real-world impact.

Research Context and Objectives:

Foundation models (FMs), including large language models (LLMs) and graph-based FMs, have revolutionized artificial intelligence. These models demonstrate remarkable adaptability across tasks by leveraging extensive pre-training and fine-tuning processes. However, their application to domains requiring the integration of multimodal, spatio-temporal, and graph-structured data—such as transport network modeling—remains underexplored.

The research aims to develop and deploy domain-specific FMs for multimodal mobility planning and management. By leveraging real and diverse datasets (telecommunication, GPS, and public transport), this project seeks to:
1. Develop Intelligent Management and Planning Tools: Create AI-powered systems to assist decision-makers in optimizing mobility infrastructure and operations.
2. Enhance Mobility Insights: Uncover novel patterns and anomalies in mobility behavior and transport system evolution.
3. Promote Sustainable and Resilient Mobility: Address urban challenges like congestion, underutilized capacity, response to critical and unforeseen events, and accessibility barriers with equity, resilience and sustainability in focus.

Technical Expectations:

The postdoctoral researcher will play a pivotal role in advancing the project through the following responsibilities:
• Data Integration and Preprocessing: Preprocess, integrate and analyze a variety of mobility-related datasets, ensuring harmonization across sources such as telecom data, public transport schedules, and GPS trajectories.
• Model Development and Adaptation: Identify, fine-tune, and validate foundation models (e.g., UrbanGPT, graph-based FMs) for mobility applications.
• Algorithmic Innovation: Design and evaluate methods for aligning model outputs with real-world needs, focusing on transport sustainability, resilience and equity.
• Case Studies and Applications: Test the models in real-world scenarios, including:
o Frency cities: Analysis of bottlenecks, multimodal accessibility, and traffic patterns.
o U.S. Cities: Refinement and testing of transferability and scalability of solutions.
• Ethical Considerations: Incorporate privacy safeguards and address biases to align the research with ethical AI principles.
• Industry Collaboration: Work closely with Orange during the final phase to validate and adapt research findings for industrial applications.

The candidate should demonstrate strong expertise in:
• Machine learning, particularly transformer-based models and deep learning frameworks (e.g., PyTorch, TensorFlow).
• Multimodal data processing and graph-based modeling.
• Programming languages such as Python.
• Interdisciplinary research, with a focus on transport systems or spatio-temporal data.

Collaboration Framework:

The fellowship is structured around collaboration with three key partners:
University Gustave Eiffel (France):
• Primary host institution led by Prof. Angelo Furno and Bahman Madadi, providing access to datasets and expertise in data-driven transport systems modeling and optimization.
• Supervision of foundational research and data preparation.
University of California, Berkeley (USA):
• Outgoing host led by Prof. Marta Gonzalez, a globally recognized expert in urban mobility and network science, and her team.
• Co-supervision during the outgoing phase, focusing on adapting and fine-tuning models to U.S.-specific urban contexts.

Orange (France):
• Industry partnership, led by Zbigniew Smoreda and Stefania Rubrichi, providing access to real-world telecom datasets and insights into practical applications.
• Collaboration during the final phase to explore the transferability of models from research to industry.

Timing Schedule and Milestones:

The fellowship will follow this detailed timeline:
Phase 1: Initial Research (3 months, University Gustave Eiffel):
• Establish research objectives, methodologies, and data-sharing agreements with partners.
• Identify, collect and analyse multi-source data, in collaboration with partners.
• Identify potential foundation models and prepare for the outgoing phase under Prof. Furno and Madadi’s supervision.

Phase 2: Advanced Research in the US (12 months, UC Berkeley):
• Refine and adapt models with U.S.-specific datasets under Prof. Gonzalez’s supervision.
• Test and validate the model architecture for scalability and generalization.
• Publish findings in high-impact journals and conferences.

Phase 3: Return Phase (12 months, University Gustave Eiffel):
• Focus on scalability and transferability to novel scenarios using Lyon as a case study.
• Conduct performance evaluations, including zero-shot learning scenarios.
• Address research questions related to sustainability, equity, and ethical AI.

Phase 4: Industrial Placement (6 months, Orange):
• Validate research findings at Orange’s premises.
• Collaborate with Orange’s R&D team and business division to adapt models for industrial use.
• Finalize deliverables, including tools and frameworks for transport system optimization.

Notes on the selection and funding process

This postdoctoral position is part of a proposal to be submitted for funding under the Marie Skłodowska-Curie Actions (MSCA) Global Postdoctoral Fellowship program. The fellowship is contingent upon the selection and approval of the proposal by the European Commission. The successful candidate will work closely with the supervisor to co-develop and submit the funding application, which will include a detailed research plan and a personalized Career Development Plan.

Interested candidates are encouraged to contact angelo.furno@univ-eiffel.fr and bahman.madadi@entpe.fr for an initial discussion about the proposal, funding scheme, and submission process. To facilitate this discussion, candidates may include a detailed CV with a list of publications to help assess their potential fit with the research topic.

2. Planned secondments

Outgoing phase at the University of California, Berkeley, under the supervision of Prof. Marta Gonzalez

Industrial Placement: 6 months at Orange facilities in Paris

3. Planned duration of the project

36 months