Institute for Intelligent Networked Systems

FreeML: Engineering Networked Machine Learning via Meta-Free Energy Minimisation
PI: Simeone
Sponsor: Engineering and Physical Sciences Research Council

Inspired by neuroscience, informed by information-theoretic principles, and motivated by modern wireless systems architectures integrating artificial intelligence (AI) and communications, this Fellowship sets out to develop a paradigm-shifting framework for networked machine learning (ML) that is centred on free energy minimisation, networked meta-learning, and native integration of wireless communication and learning.

NeuroComm: Brain-Inspired Wireless Communications – From Theoretical Foundations to Implementation for 6G and Beyond
PIs: Simeone and Rajendran
Sponsor: Engineering and Physical Sciences Research Council & US National Science Foundation

This project explores integrated neuromorphic sensing and computing which is emerging as alternative, brain-inspired, paradigms for efficient data collection and semantic signal processing that build on event-driven measurements, in-memory computing, spike-based information processing, reduced precision and increased stochasticity, and adaptability via learning in hardware.

CONTRACT: Conformal Calibration for Reliable AI-Based Wireless Communications
PI: Simeone
Sponsor: European Research Council

CONTRACT sets out to investigate a novel, theoretically principled, framework for the reliable
deployment of artificial intelligence (AI)-based black-box models in wireless systems.

Given the highly non-deterministic patterns of traffic and connectivity conditions, AI is currently viewed as an essential technology to ensure the support of communication services with widely heterogeneous performance requirements, ranging from cloud gaming to Industry 4.0.

However, AI “apps” are, by and large, black boxes, whose introduction within the larger wireless network raises critical concerns about reliability.

The main goal of CONTRACT is to endow for the first time AI-based wireless deployments with formal reliability guarantees.

To this end, CONTRACT will introduce novel methodologies for: (i) the offline calibration of pre-trained AI apps via hyperparameter optimization prior to deployment; and (ii) the online run-time monitoring of key performance indicators and conflicts among simultaneously deployed apps. In CONTRACT, hyperparameter optimization and online monitoring will leverage contextual information describing traffic and connectivity conditions, as well as synthetic data from a digital twin, i.e., a simulator, of the wireless network.

Verifiably Robust Conformal Probes
PIs: Nicola Paoletti (King’s College London), Simeone

This project develops ways of detecting misaligned LLM behaviours with a high degree of confidence, attaching a quantitative value to that level of trust. The goal is to enable safer and more socially beneficial models and empower policy makers to make better decisions about any necessary guardrails on the technology. 

SGAI: Smart and Green AI
PI: Rajendran
Sponsor: Engineering and Physical Sciences Research Council

This EPSRC fellowship will lay the foundations for a new AI paradigm featuring algorithms based on the free energy principle (FEP) and hardware platforms leveraging the stochasticity of novel nanoscale devices based on 2-dimensional materials, enabling embedded systems with unprecedented efficiency.

NeuroSoC: A multiprocessor system on chip with in-memory neural processing unit
PI: Rajendran
Sponsor: Horizon Europe & InnovateUK

The overarching objective of NeuroSoC is to develop a flexible computing system where an analogue in memory computing (IMC)-based neural processing unit is integrated into a multi-processor functional safe and secure system-on-chip to tackle the requirements of a wide set of edge-AI applications.

NeuMatmul: Neuromorphic Accelerators for Learning and Inference with LLMs
PI: Rajendran
Sponsor: ARIA

The project will develop brain-inspired algorithms and hardware accelerators for fine-tuning large language models, avoiding the energy-intensive backpropagation methods used in current AI training.

By combining these biological learning principles with in-memory computing hardware, the team aims to significantly reduce the energy and time required to train and adapt AI models.