Spin Inspired Representations

点击翻译按钮获取中文摘要

Unconventional computing (UComp) devices are based on a diverse range of physical substrates, from atomic switch networks to carbon nanotubes, from liquid crystals to slime moulds. One goal of UComp is to exploit the properties of each substrates to compute somehow "better'' than its classical counterpart. This might be faster, or with less power, of in a device that can be more easily embedded/embodied in a larger system. To do so, it is important to have a computational model that fits "naturally'' on the material, rather than attempt to impose an inappropriate model that fights its implementation. Exploitation of these novel substrates desperately needs a principled general methodology for determining such natural models. Representation is central to software development: design the correct representation and operations on it (the algebra of the representation), and the task is half done. All classical computing representations, no matter how high level the language (strings, arrays, lists, graphs, structures, dictionaries, databases, ...), ultimately reduce to the basic unit of the bit, and bitwise operations. In other computational domains, this basic unit may be different: the qubit in quantum computing; the real number in continuous analogue computing; other low level representations in other UComp substrates. However, UComp fields currently have under-researched and therefore under-developed higher-order representations, leaving their programming closer to an "assembly language'' level, with all its problems of lack of abstraction and usability. This is holding back the application of novel computational materials.SpInspired will develop a methodology for discovering and exploiting good natural computational models of material computing, and for determining appropriate basic units that are both mathematically rich (so that they can be composed and combined into higher order representations) and efficiently physically realisable (so that they can be exploited as powerful combinational devices in unconventional domains and applications). We will do so by exploiting two diverse exemplar "sandpit" systems -- the well-characterised rich physical process of NMR spectroscopy, and ill-characterised carbon nanotube disordered material -- to guide the generic framework development. The successful project outcome will be a generic UComp development framework that will establish the gold standard for characterisation, analysis and comparison of different in materio computing systems. No such framework currently exists, which places a significant limitation of scope and general exploitation of current research undertaken in this field, as results are ad hoc and material-specific.