Projects: Current


TimeStorm (EU project)

Time perception is a fundamental capacity of autonomous living biological and computational systems that plays a key role in the development of intelligence. In particular, time is important for encoding, revisiting and exploiting experiences (knowing), for making plans to accomplish timely goals at certain moments (doing), for maintaining the identity of self over time despite changing contexts (being). The TimeStorm research project studies the principles of time processing in the human brain and their replication in-silico, adopting a multidisciplinary research approach that involves developmental studies, brain imaging, computational modelling and embodied experiments. The research will contribute important scientific and technological advancements in the triad of knowing-doing-being by investigating and implementing for the first time artificial systems that consider the temporal aspects of cognition.

Website



Kinds of Intelligence (Leverhulme project)

Computers outperform humans on many tasks, but it is unclear whether and when the development of artificial general intelligence (AGI) will occur. One of the difficulties is that intelligence, both natural and artificial, is notoriously difficult to define. Much work has been done in AI both in defining general intelligence , and specifying different forms of intelligent behaviour and reasoning ability, drawing both on principles of mathematical logic and rationality, and on work in psychology and neurobiology. This project draws on leading work in these fields to further develop and critically assess notions of general intelligence used in AI.
Recent progress in neurobiology and psychology has illuminated aspects of general intelligence in biology, including its computational organization, evolutionary origins, relation to other capacities such as cultural intelligence, and association with traits such as brain size and processing speed. This has contributed to growing cross-fertilisation with machine learning; reflected in the research strategy of leading teams, including our industry partners Google DeepMind and Vicarious. This project will draw on the latest findings in neurobiology and psychology, combined with work in computer science and cognitive robotics, with the ultimate aim of mapping the space of possible intelligences – biological, artificial, and hybrid. This will enable more accurate predictions of AGI development and improved assessments of its benefits and risks.

Website



Brain Studio

Brain Studio has been developed by our lab at Imperial College, with an aim to automate and speed-up the process of designing and testing large-scale spiking neural networks (SNNs), through a suite of new simulation tools. In particular, this suite provides
(1) a framework for the compact representation of modular SNNs, suitable for large-scale brain circuits and cognitive systems,
(2) a generic and powerful simulator that allows the fast implementation, visualization and real-time adjustments of SNNs, and
(3) a fast optimization technique, which could be used in order to mimic the dynamical behaviour of specific neurons in the brain, using simple phenomenological models.
Finally, Brain Studio uses NeMo for CUDA hardware-accelerated simulation of phenomenological SNNs (such as Izhikevich neurons).

Website



NeMo Simulator

NeMo is a high-performance spiking neural network simulator which simulates networks of Izhikevich neurons on CUDA-enabled GPUs, aiming for real-time simulation of 100 000s of realistically connected spiking neurons.
NeMo is a C++ class library, and also has bindings for pure C, Matlab, and Python. NeMo is one of the supported backends for the PyNN common simulator interface (in PyNN's development branch). SpikeStream can be used as a graphical frontend for the simulator.

Website




NeuroBot

NeuroBot is a software system able to control avatars (bots) within the environment of the computer game Unreal Tournament 2004, in a human-like manner. To achieve this, its controller is governed by a computational model, based around a large-scale modular spiking neural network (~20.000 neurons) divided into specialized regions. The high level coordination of this network is carried out by an architecture based on global workspace theory.
The final version of NeuroBot achieved second place in the 2K BotPrize human-like bot competition at CIG 2011 in Seoul, South Korea.

Website