The NETT ITN addresses some key challenges in Neural Engineering through inter-related projects combining the skills of mathematicians, physicists, neuroscientists and bioengineers. It will generate transformative technologies for novel speech recognisers, neural-inspired laser networks for information processing, brain-computer interfaces (BCI), robots with cognitive skills and neural prosthetics for enhancing or repairing sensory-motor functions.
NETT’s primary objectives are to:
a) Bring together researchers from a variety of backgrounds and areas that feed into Neural Engineering in order to pool resources and tackle current challenges with an interdisciplinary team trained in core skills in complexity science.
b) Expose early career researchers to the different opportunities the field offers in academia and industry, training them in the methodology employed in both, with input from experienced visiting researchers and corporate research teams.
c) Progress the careers and employability of the fellows via internships with industry partners and a dedicated training programme in business, academic and social skills.
d) Demonstrate the interrelatedness of the various project areas throughout the consortium and to the wider scientific community, industry and the general public through regular study groups, workshops, public lectures and web presence.
The research highlights in NETT so far include:
• Characterisation of the microscopic and macroscopic dynamics of sparse pulse coupled neural networks (Papers under review in Physical Review E & Chaos, Solitons and Fractals)
• Development of SPIKY: Matlab tool for monitoring spike train synchrony (open source - http://wwwold.fi.isc.cnr.it/users/thomas.kreuz/Source-Code/SPIKY.html)
• Establishing the effect of transmission delays in the communication between brain areas (Paper in PLOS Computational Biology)
• Development of a novel algorithm for multi-photon scanning of neural tissue (presented at IEEE Chicago EMBS Meeting)
• Probed scale interaction in the brain through synchronisation (Paper accepted in Philosophical Transactions of the Royal Society B)
• Development of two-photon targeted patch clamping robot (presented at SFN 2013)
• Mathematical frameworks for oscillatory network dynamics in neuroscience (Paper in the Journal of Mathematical Neuroscience)
• Large scale model for brain activity and efficient numerical schemes for delayed neural field simulations (presented at ICMNS 2015)
• Dissemination of NETT activity on Artificial recognition of sounds in complex scenes from auditory neuronal activity to members of UK parliament (SET 2016)
• Development of a two-layer microfluidic device enabling the seeding, growth and delivery of agonists to a neuronal micro-culture at physiological time scales (link here)
• Development of a synchronous dual-arm movement planner for our humanoid robot ARoS.
• Implementation of a neural mass model with a source of temporally correlated noise, validation with experimental EEG recordings and subsequent predictions applicable to neurological disorders, in particular to epilepsy. Our preliminary conclusions were presented on a poster session on the annual meeting of Organization for Computational Neuroscience in Prague, 2015.
• Development of a Python toolkit (code here) to understand neural recordings from large populations of neurons.
• Development of free calibration schemes for rehabilitation therapies that use gait intention detectors. Results for stroke patients were published here.
The need for a European focus in Neural Engineering will be met by bringing together 7 main complementary research work plans (WP1-7). Each of these is driven by a set of problems (P1-18), whose resolution will rely on information and knowledge transfer between NETT members. The problems are networked so that theoreticians from one discipline will engage with engineering or experimental practitioners from another on projects driven by real-world problems.
WP1 Adaptive control methods
P1 aims to include insights from active learning, or experimental design, and stochastic optimal control theory together with state of the art techniques from computational neuroscience to develop a theoretical framework and practical algorithms for robotics or adaptive brain-computer interfaces (BCI). P2 will comprise the modeling of a new method of signal processing and the adaptation and validation of such a model for robotics or BCI. The objective is to integrate this model into commercial software.
Deliverable 1.1: Development of an adaptive control method.
Deliverable 1.2: Integration of stochastic optimal control principles in learning neural networks.
Deliverable 1.3: Demo of the previous principles in simulation.
Deliverable 2.1: Design of a new method of signal processing for robotics or BCI.
Deliverable 2:2. Software validation, integration into existing technology.
WP2 Synthetic Cognition
P3 will develop a neurocomputational model of how populations of spiking neurons in the audio-motor pathway select, prepare, encode and generate optimal trajectories for auditory-evoked eye-head gaze shifts in complex acoustic environments that contain many potential sound sources. This project links strongly with WP1 and WP3.
In P4, by integrating microfluidic modules with existing multi-electrode array (MEA) technology, we will control the network’s environment, deliver pharmacological agents, monitor firing activity and develop a theoretical framework. P5 has a direct objective of building an appropriate mathematical framework to understand the neurodynamics of spiking networks. P6 will develop a single word speech recogniser that receives as its input the output from a large population of real spiking neurons.
Deliverable 3.1: Demo of auditory model system with real time acoustic tracking.
Deliverable 3.2: Optimal control model of eye-head orienting.
Deliverable 3.3: Demo of audiometer model system.
Deliverable 4.1: Design and proof-of-principle integrated MEA microfluidics device.
Deliverable 4.2: Theoretical framework for understanding MEA microfluidic experiments.
Deliverable 5.1: Mathematical framework for spiking neural networks applied to cultures and robot control.
Deliverable 5.2: Stochastic neurodynamics beyond standard mean field analysis.
Deliverable 6.1: Neural recording of responses to the designed word set.
Deliverable 6.2: A single word speech recogniser working from neural responses.
Deliverable 6.3: Assessment of speech recognition using auditory cortical neurons.
WP3 Human-Robot Interaction (HRI)
P7 will analyse, numerically test and implement on a robotics system different neuro-plausible correlation-based learning rules and adaptation mechanisms. A direct objective of this work is to design autonomous robotic systems with human-like social and cognitive skills, with natural applications in health-care and real-world service tasks. P8 aims at endowing a high-degree of freedom robotics arm-hand system with movement capacities that reflect fundamental characteristics of human arm and hand trajectories. By providing physical assistance such robotic devices can be used for the rehabilitation of upper-limb motor functions after brain injury.
Deliverable 7.1: Theoretical study of learning and adaptation in the dynamic field architecture for HRI.
Deliverable 7.2: Demonstration of the robot’s learning and adaptation abilities.
Deliverable 8.1: Human-like movements of a high DOF robot arm.
Deliverable 8.2: Demo of the robot’s motor capacities in a a rehabilitation environment.
WP4 Neural Inspired Information Processing
P9 has two main goals. One is to develop semiconductor laser networks that mimic the information processing capabilities of small neuronal networks. The second goal of the project is to use knowledge generated from the study of behaviour of these laser networks to make predictions about the operation of neuronal networks. P10 addresses modelling at a mesoscopic level of neuronal activity to complement experimental data obtained using neural mass models; it will study the effects of noise, and relate the results obtained with potential coordination malfunctions leading to aberrant synchronized behaviour and subsequently to neurological disorders.
Deliverable 9.1: Relating structural coupling with dynamical correlation in laser networks.
Deliverable 9.2: Information processing capabilities of laser vs neuronal networks.
Deliverable 10.1: Sources of noise in neural mass models.
Deliverable 10.2: Ordering effects of random fluctuations at the mesoscopic level.
WP5 Neural Coding
P11 is aimed at developing information-theoretic algorithms for analysing data from large ensembles, such as can be recorded with new two-photon imaging techniques. Applications are to both BCI and basic scientific questions. P12 is aimed at developing two-photon targeted patch-clamping robotic technology; this is a specific instance of optogenetic BCI technology with significant prospects for high throughput in vivo drug characterisation as well as reverse engineering neural circuits. P13 concerns the implementation of a system for optimal sequential cellular recording in an inertial laser scanning microscope and the application of it to collect dataset with high cell count for characterisation of neural population code. P14 will aim to develop theoretical/computer vision approaches for the analysis of large scale optical neural recordings.
Deliverable 11.1: Neural decoding from 20-neuron patterns with development of a Matlab toolbox.
Deliverable 11.2: Neural decoding from 200-neuron patterns with development of a Matlab toolbox.
Deliverable 12.1: Development of a prototype LabView based system for automatic whole cell patch clamp recording
Deliverable 12.2: Demonstration of robotically automated whole cell patch clamp recording in vivo.
Deliverable 12.3: Demonstration of platform for combined two-photon targeted robotic patch clamping.
Deliverable 13.1: Implementation of novel scan algorithm in MATLAB and simulation of microscope.
Deliverable 13.2: LabView implementation of control algorithm on scope hardware.
Deliverable 13.3: Use of system to collect dataset with high cell count for characterization of neural population code.
Deliverable 14.1: A robust segmentation algorithm for two-photon calcium imaging.
Deliverable 14.2: Information theoretic analysis of signals from two-photon calcium imaging.
WP6 Emergent Neurodynamics
The principal aim of these projects is to study collective emergent dynamics in neural networks where neurons are represented by means of simple models. P15 plans to realise cultured neuronal networks with predefined topologies and to investigate, by employing multi electrode arrays technique, the possible emergence of coherent states in networks of cultured rat neurons. For P16, an ER will be also involved in the data analysis of experimentally measured spike train series and LFPs and particularly in Matlab software development. P16 will serve as a tool to be applied in all nodes of the ITN to teach innovative uni- and multi-variate data analysis and to help in developing nonlinear time series analysis specific to projects at the local laboratories.
Deliverable 15.1: Identification of Collective Solutions in scale free networks.
Deliverable 15.2: Robustness of collective solutions to the introduction of delay and inhomogeneities.
Deliverable 16.1: Analysis of LFP and single activity in cultured networks.
Deliverable 16.2: Relevance of hub neutrons for macroscopic oscillations in simulated and cultured neural networks.
Deliverable 16.3: Development of a Matlab toolbox for spike train time series analysis Ver1, Ver 2
WP7 Neural rehabilitation
P17 is related to robot-mediated motor rehabilitation therapies for patients after stroke, where one of the key points is to move on the rehabilitation task only when the patient is engaged and thus using their (damaged) motor cortex. In this context the objective is the development of signal processing and machine learning techniques able to detect in real-time neural processes preceding movement and motor imagery, i.e. intention of movement. P18 is related to any rehabilitation therapy such as stroke or neurofeedback systems for attention deficit disorders, where a key point of the therapy is to move on only when the patient is attending to the task. In this context the objective is the modelling and real-time detection of human cognitive information of the rehabilitation task related to the degree of attention.
Deliverable 17.1: EEG decoders to detect motor anticipatory potentials in motor therapies.
Deliverable 17.2: EEG decoders to detect motor anticipatory potentials in stroke patients without inter-session recalibration
Deliverable 18.1: EEG decoders of mind wandering.
Deliverable 18.2: EEG integrated platform for mind wandering detection in a neurorehabilitation application.