Work packages

Work package 1: Materials Processing and Evaluation

Work package 1 will supply nanomaterial networks to members of the consortium in order to provide thin film systems that will be used as the basis for evolutionary computation – top-down approach. However, a key objective of this work package is to achieve a physical understanding of the evolutionary/training processes. The emerging behaviour of a simple materials network will be investigated and modelled using a bottom-up approach.

We shall endeavour to provide answers to the following (essential to develop a platform for more complex evolution). Can we alter the electrical behaviour of a simple material network using applied electrical stimuli? These stimuli will be dc voltages, pulses and ac signals. Which of these will be the most appropriate? What device/material configurations should we use? What changes can we achieve and are these reversible? Is reversibility essential for the sort of devices we hope to evolve? Can we fix or freeze-in the final configuration, say by distributing the electro-active elements in a monomer and then cross-linking with UV light? The work package will comprise a series of tasks.

Work package 2: Interface (hardware)

The interface work package must provide a basic platform able to interface to different materials with different properties, i.e. timing, signal levels, etc., without major revisions. That is, within the two main objectives, the work package must provide an interface that can be adapted to the range of current potential substrates, but should also be able to interface to materials that most likely will emerge during the NASCENCE project.

The explorative engine in the NASCENCE project is based on Evolution, e.g. Evolutionary Algorithms (EAs) including Cartesian Genetic Programming (CGP), Genetic Algorithms (GAs), Genetic Programming (GP) and Evolutionary Strategies (ES), all operating in the digital discrete world of computers. Target substrates for computing are not constraint by the strict timing and levels of the digital domain; the rich dynamics of matter operates in continuous time expressing non-discrete analogue values. Interfacing the digital domain to the analogue operation of computational matter requires an interface that can use the power of evolution to explore the rich dynamics and possible computational properties of the matter.

A key factor in this interface is to avoid the digital domain, running the EA, to constrain exploration of possible computational power in the matter. As such, the interface must include a mapping from the digital genetic representation to the analogue domain for configuration, and incorporate a corresponding flexible feedback interface from the physical parameters of materials to the digital domain of evolutionary processes. In order to keep the explorative approach unconstrained, the actual input parameter exploited by the matter should not be defined in a way that restricts what signal property that can be exploited; matter can exploit whatever parameter the input data provide, e.g. variation in level, timing, power spectrum, etc. Likewise the interface must be able to detect similar response from the matter.

Work package 3: Interface (software)

Work package 3 develop a software interface to allow search algorithms to interface with the hardware and develop evolutionary algorithms. Since the mapping from applied electrical signals to device configurations is not known, there is no way to directly program the desired device configuration. This leads to the challenge of how to get the materials to perform a particular computation on a set of input data. 

In previous work with liquid crystals and FPGAs, this problem was addressed by using artificial evolution. Using this approach, the job of programming becomes that of describing the problem in the form of a fitness function, which measures how close the computation performed is to the desired computation. By selecting the most “fit” electrical stimuli for reproduction, from a population of candidates, the evolutionary search is guided toward a successful configuration.

Work package 4: Computational tasks (experimental)

Work package 4 is to demonstrate that one can perform computation using a variety of nanoscale materials. Investigations will be undertaken that will identify suitable computational tasks that can be successfully achieved. Some of the tasks will be standard computational tasks using in machine learning and artificial neural networks. This will allow comparisons between the evolved physical computational systems developed in the project and conventional approaches.

However, other tasks may have to be devised that better utilize the type of computation that is possible in material systems. A complementary strand of research undertaken in this work package will be to attempt to build a physical analogue of a biological neuron. In this work package, the hardware platform developed in WP2 (NTNU) that uses materials developed in WP1 (UDUR) is used together with the software developed in WP3 (SUPSI-IDSIA). The aim is to identify suitable computational tasks that are feasible for the evolvable device. These tasks will be devised in association with WP5 (UT) as we want to compare advantages and disadvantages and use the simulations of WP5 (UT) to investigate whether it is likely that these computational tasks can be performed by the device.

After these tasks have been identified and successfully simulated, experiments will be conducted to investigate how well the evolvable material platform can solve these tasks, how stable the computation is, and how environmentally sensitive the computation is. Data produced in this work package will be analysed in WP5 (UT) and WP6 (SUPSI-IDSIA). An important aspect of the research outcome in this work package is reproducibility, consequently once working systems solving computational tasks have been obtained, the experimental details will be communicated to NTNU, so that experiments can be undertaken that show that such behaviour and performance can be replicated. This is one of the reasons why the project requires the construction of a number of (almost) identical evolution-in-materio platforms.

Work package 5: Computational tasks (simulation)

Work package 5 is intimately related to WP7 and is aimed at developing the mathematical and computational foundations for describing the functionality and complexity, and for predicting the behavior of the evolvable systems developed in WP1, WP2 and WP3, in order to support the task of identifying and choosing suitable computational tasks in WP that are likely to be feasible for the evolvable device.

Work package 6: Evaluation and data mining

Work package 6 use machine learning to automatically explore the computational properties of the system and data mining to find interesting observations about the system.

Open ended learning systems will be used to automatically learn what interesting functionality can be extracted from our devices. Examples of open ended learning systems include artificial curiosity and scouting. In traditional machine learning, a problem would be posed by a humanand then the machine learning algorithm applied to solve it. This is acceptable when there are clear ideasabout what you are trying to learn about and there is a clearly defined goal. However, when there are lots ofunknowns and a new domain to explore, this approach could stifle our deeper understanding of what thesystem can do and how it can do it. With the open ended approaches, the machine learning continuouslyinvents new experiments to try and sees if it is possible to solve them.

The data mining approaches of  this work package will use data from other experiments to discover behaviours that were not initially apparent. For example, were there environmental changes that affected thebehaviour of the device and that should be accounted for in future work? The outcomes of this will then feedback into other work packages to inform the design of experiments and increase the likelihood of success

Work package 7: Mathematical foundations

The aim of work package7  is to develop a new theoretical foundation for the evolvable systems that will be developed in WP1, WP2 and WP3, and to apply it in the simulations of WP5. First of all we shall compare existing models for particle interactions that are based on continuous mathematical concepts and discrete mathematical concepts. Based on the simulations in WP5 we shall see which of them are best suited in terms of quality of solutions and computational efforts and resources. We shall also explore the use of graph theoretical concepts in order to find simpler models for particle interactions and flows in network structures of nanoparticles, and for future usage in large scale simulations.

Work package 8: Dissemination and exploitation

Dissemination of the project results is considered as an important activity by the consortium members, since it will improve the scientific exploitation of the project. A distinction is made between public and scientific dissemination. The former gives visibility to the project and consortium and makes the project broadly based and accepted, whereas the latter disseminates the innovations of the project to a focused scientific and research community. The dissemination and exploitation of the results of the project includes the creation of a web site that will enable the visibility of the project results. Information about the main topics of the project and a state-of the-art review and technology watch for those topics will be made available.

The results of the project will be disseminated via:

  • the project web site
  • publications in conference proceedings, journals, magazines, books, etc. 
  • presentations at conferences, seminars, workshops, etc.
  • demonstrations at conference exhibitions, seminars, workshops, etc.
  • press releases and marketing materials.

There are no requirements related to confidentiality of research results. If this is feasible, we shall develop an evolvable platform that can be evolved remotely via the web site.

Work package 9: Project management

The project is a multidisciplinary venture that deals with research and engineering topics. All required expertises are present within the project consortium. The project has a well-structured project management that will exploit the potential of the consortium. The financial administration of the project is placed in a separate task for the Twente group. Their neutral financial department will enforce strict control and supervision of the project budget. The Twente group will perform the overall coordination of activities within the work plan and monitor overall performance, time and costs of the project. As such the main objectives can be summarized as:

  • Support the technical work packages on administrative, financial, political and ethical issues
  • Stimulate and maintain communication, coordination and cooperation between the work packages
  • Supervise that current project specifications contribute to the defined objectives of the task and overall project

Reports on the deliverables of the NASCENCE project can be found under Approved Deliverables.