Funding

We welcome philanthropists, Maecenas, and good patrons and patronesses.

Free Energy Principle for adaptive cognitive architectures

PI/CIs: Michaël Garcia-Ortiz / Esther Mondragón and Eduardo Alonso
Funder/scheme: DSTL, UK-France Joint Research PhD Programme
Duration: 3 years (2020-23)
Amount: £98,000

Reinforcement Learning (RL) is widely used to teach robots how to make decisions and act. Tremendous advances were recently made; however, agents still suffer from a lack of generalization, adaptation and transfer of their knowledge to novel environments. The Free Energy Principle (FEP) takes inspiration from Information Theory, Neuroscience and Cognitive Sciences. It formulates perception, reasoning and action in a probabilistic framework, and elegantly accounts for the emergence of high-level cognitive functions. It is motivated by the need for life to adapt to its environment, and therefore complements RL approaches. We hypothesize that FEP can improve RL for Robotics by offering more efficient and flexible architectures. We propose first to enhance current RL architectures with neuro-inspired algorithms that were identified as key components of intelligence in the FEP theory, then to implement optimization mechanisms inspired from the FEP to provide adaptability to agents.

Two full-time PhD studentships available to pursue cutting-edge research in Artificial Intelligence. For any further enquiries contact Michael Garcia-Ortiz.
Apply Here.

Learning, Approximating and Minimising Streaming Automata for Large-scale Optimisation

PI: Laure Daviaud
Funder/scheme: EPSRC / New Investigator Award
Duration: 3 years (2020-23)
Amount: £324,660.48

This project links formal methods used in verification and artificial intelligence, proposing to apply learning techniques to improve the efficiency of some algorithms which certify computer systems and to compute fast accurate models for real-life systems. Automata are one of the mathematical tools used in verification to model computer or real-life systems. Giving certifications on these systems often boils down to running some algorithms on the corresponding automata. The efficiency of such algorithms usually depends on the size of the considered automaton. Minimising automata is thus a paramount problem in verification, as a way to verify large computer or real-life systems faster.
This proposal aims at studying the minimisation of some streaming models of quantitative automata using machine learning techniques. The kind of automata we are going to focus on, are streaming models, in the sense that the input is not stored but received as a stream of data and dealt with on the fly, thus being particularly suitable for the treatment of big data. They are also suited to deal with optimisation problems such as minimising the resource consumption of a system or computing the worst-case running time of a program. Minimising these kind of automata is highly challenging and linked with the long-standing open problem of the determinisation of max-plus automata.


Learning to Represent Scenes for Artificial General Intelligence and Robotics

PI: Michaël Garcia-Ortiz
Funder/scheme: City, University of London / City Pump Prime Funding
Duration: 1 year (2020-21)
Amount: £19,468.69

Today, robots and agents are limited in their learning capabilities, and can’t adapt to novel environments, which is a strong issue that lead to new developments in transfer learning, abstraction, and generalisation. To adapt to the unknown, a robot must build representations of the world, how to act in it, its physics, and develop a common-sense knowledge akin to the one present in animals and humans. This project is at the intersection of multiple scientific fields that would allow to learn such representations: continual learning and unsupervised learning of scene decomposition.


Building a partnership with Sony Computer Science Laboratories (Sony CSL) and Japanese Universities

PI/CIs: Michaël Garcia-Ortiz / Eduardo Alonso and Nabil Aouf
Funder/scheme: City, University of London / Global Partnership Seed Fund
Duration: 1 year (2020-21)
Amount: £4,985

This award will finance visits to potential partners in Japan, in order to build long-lasting relations with SMCSE on the topics of Artificial Intelligence and Robotics. In particular, Sony CSL and CitAI share a common interest in the domains of Artificial General Intelligence and Developmental Robotics.


Algorithms for predictive maintenance of vehicles in a connected environment

PI/CI: Eduardo Alonso / Michaël Garcia-Ortiz
Funder/scheme: EIT-Digital and Bosch AA-AS / Industrial AI Doctoral Training Program
Duration: 3 years (2020-23)
Amount: £154,000

Novel algorithms will be developed for predicting and specifying repair and maintenance requirements of vehicles in connected environments, where the vehicle and the repair workshop are in two-way data communication with cloud-based services. These may include algorithms for identifying what data to collect, identifying and predicting faults using collected data, advising vehicle operators of repair and maintenance requirements and optimising the repair and maintenance tasks to be performed.


FourCmodelling: Conflict, Competition, Cooperation and Complexity: Using Evolutionary Game Theory to model realistic populations

PI/CIs: Mark Broom / Eduardo Alonso, Andrea Baronchelli and Anne Kandler
Partners: Biology Center, Ceske Budejovice, Eotvos Lorand University, Maastricht University, University of Szeged, University of Torino, Arizona State University, Moffitt Cancer Center, University of Illinois at Chicago, University of North Carolina at Greensboro and Wilfird Laurier University
Funder/scheme: EU / H2020-MSCA-RISE-2015
Duration: 3 years (2016-19)
Amount: €243,000

Real animals and human populations are complex, involving structural relationships depending upon space and time and varied interactions between potentially many individuals. Human societies feature family units, communities, companies and nations. Some animal also have complex societies, such as primate groups and social insect colonies. Single organisms themselves can be thought of as complex ecosystems, host to many interacting life forms. Models of populations are necessarily idealised, and most involve either simple pairwise interactions or "well-mixed" structureless populations, or both. In this project we are developing game-theoretical models, both general and focused on specific real population scenarios, which incorporate population structure and within population interactions which are both complex in character. We focus on the themes of Conflict, Competition, Cooperation and Complexity inherent in the majority of real populations. There are four complementary sub-projects within the overall project. The first focuses on developing a general theory of modelling multiplayer evolutionary games in structured populations, and feeds into each of the other three sub-projects. The second considers complex foraging games, in particular games under time constraints and involving sequential decisions relating to patch choice. The third involves modelling human social behaviours, a particular example being epidemic cascades on social networks. The final sub-project models cancer as a complex adaptive system, where a population of tumour, normal and immune cells evolve within a human ecosystem. The four sub-projects have been developed in parallel fostered by frequent research visits and interactions, each involving a team comprising of EU and North American researchers, and feed into each other through regular interactions and meetings. The aim is to develop a rich, varied but consistent theory with wide applicability.


Innovative Technology for District Heating and Cooling (InDeal)

PI/CIs: Nicos Karcanias/ Eduardo Alonso and George Halikias
Partners: NAITEC, The Centre for Research and Technology-Hellas (CERTH), The Institute for Research and Technology Thessaly (IRETETH), The French Alternative Energies and Atomic Energy Commission (CEA Tech), IZNAB, The Center for Technology Research and Innovation (CETRI), Net Technologies Finland (NET), PROMAR Ltd, Energetika Projekt, Syndicat national du chauffage urbain et de la climatisation urbaine (SNCU), FEDENE | Fédération des Services Energie Environnement, and SERM (Société d'Équipement de la Region Montpelliéraine).
Funder/scheme: EU / H2020-EE-2015-2-RIA
Duration: 3 years (2016-19)
Amount: €1,992,726.25

Challenged by climate change, and coupled with the need to secure sustainable economic growth and social cohesion, Europe must achieve a genuine energy revolution to reverse present-day unsustainable trends and live up to the ambitious policy expectations. A rational, consistent and far-sighted approach to heating and cooling is key for ensuring such transformation. Toward this direction, district heating and cooling systems need to be more efficient, intelligent and cheaper. InDeal project will offer an innovative platform that will impose a fairly distribution of heating and cooling among the network’s buildings by: (i) real – time energy consumption data gathering via artificial intelligent meters, (ii) identifying and evaluating the network’s buildings’ need and demand for heating and cooling depending to their energy efficiency, energy consumption and type of building (EDP tool), (iii) predicting the short-term and long-term weather conditions and forthcoming need for heating and cooling (EDP tool), (iv) monitoring and control the level of energy stored in network’s storage stations and substations (SMT), (v) 24/7 monitoring of the DHC system by a central control platform and (vi) minimizing heat losses via novel pipe design solutions and innovative insulation materials. The target of InDeal is to turn the current DHCS into a new next-level automated DHCS that will guarantee the increase of the overall energy efficiency of the system accomplishing a fairly distribution of heating and cooling energy demands. In light of this, InDeal will make a significant step forward contributing to wider use of intelligent district heating and cooling systems and integration of renewables, waste and storage.


Learning in Autism: A systematic computational approach

PI/CI: Sebastian Gaigg / Eduardo Alonso
Funder/scheme: Baily Thomas Charitable Fund
Duration: 3 years (2016-19)
Amount: £70,000

Recent theories suggest that reward learning abnormalities may lie at the root of Autism Spectrum Disorders (ASD). However, relevant empirical evidence remains scarce, especially concerning those individuals who have additional intellectual impairments. In a series of pilot studies we develop the experimental methods necessary to systematically examine reward learning processes across the entire autism spectrum. The work builds foundations for a fuller understanding of the aetiology of ASD and directly inform interventions that promote learning through rewards.


Deep Learning for Route Planning

PI: Eduardo Alonso
Funder/scheme: MBDA UK Limited
Duration: 3 years (2016-19)
Amount: £9,000

The aim is to use an evaluate deep learning and reinforcement learning techniques to extract features and contextual information from terrain data, and to find a best route through the previously generated cost spaces.


Projects

  • FEP for Adaptive Cognition
  • Automata
  • Learning Scenes Representations
  • Partnership award
  • AI for vehicles
  • FourCmodelling
  • InDeal
  • Learning in Autism
  • DL for route planning