Overview
Developing efficient, adaptive, and trustworthy foundation model inference across the cloud-edge-IoT computing continuum.
CPI Lab Team
f-inference (Efficient Foundation Model Inference Across the Computing Continuum) is a Horizon Europe research project developing an open framework for efficient, adaptive, and trustworthy foundation model inference across cloud, edge, and Internet of Things (IoT) infrastructures.
The project addresses the limitations of cloud-centric AI deployments, including high computational costs, energy consumption, network latency, and concerns regarding data sovereignty.
By introducing a Resource-driven Computing Continuum (RCC), f-inference aims to distribute and orchestrate AI inference across heterogeneous computing resources, enabling intelligent applications to operate closer to their data sources while balancing performance, security, energy efficiency, and affordability.
Research Objectives
The project focuses on five research objectives:
Foundation Model Optimisation: Develop profiling, compression, early-exit strategies, and device-aware model adaptation techniques.
Efficient and Adaptive Inference: Enable distributed inference through dynamic routing, adaptive caching, and efficient communication across heterogeneous infrastructures.
Trustworthy Orchestration: Develop mechanisms for resource-aware deployment and orchestration while considering security, performance, economic constraints, and data sovereignty.
Industrial Validation: Evaluate the proposed technologies through representative use cases in digital industry, automotive systems, and healthcare.
Research and Societal Impact: Promote open-source tools, scientific publications, standardisation activities, and collaboration across European research and industrial communities.
Research and Innovation
f-inference combines model optimisation, distributed execution, and intelligent orchestration to support foundation model deployment across the computing continuum.
The research investigates how AI models can be adapted to available computing resources, how inference tasks can be distributed efficiently, and how deployment decisions can accommodate changing operational requirements.
The resulting framework aims to improve accessibility to advanced AI technologies while reducing infrastructure dependencies and supporting trustworthy data processing.
Use Cases
The project validates its research outcomes through three industrial pilot domains:
- Digital Industry: Foundation models for monitoring and managing computing and network resources in industrial environments.
- Automotive: Privacy-aware situational intelligence for cooperative, connected, and automated mobility.
- Healthcare: Predictive maintenance supporting medical-isotope reactor operations without interfering with safety-critical controls.
Project Information
| Details | |
|---|---|
| Programme | Horizon Europe |
| Grant Agreement | 101298393 |
| Duration | October 2026 – September 2029 |
| Coordinator | Institute of Communication and Computer Systems (ICCS), Greece |
| TU Delft Principal Investigator | Aaron Ding |
| Consortium | 19 organisations from 12 countries |
| EU Contribution | €7.98 million |
| Official Website | f-inference.eu |
