-
Computing with Clocks
Authors:
Jonathan Edwards,
Alex Yakovlev,
Simon O'Keefe
Abstract:
Clocks are a central part of many computing paradigms, and are mainly used to synchronise the delicate operation of switching, necessary to drive modern computational processes. Unfortunately, this synchronisation process is reaching a natural ``apocalypse''. No longer can clock scaling be used as a blunt tool to accelerate computation, we are up against the natural limits of switching and synchro…
▽ More
Clocks are a central part of many computing paradigms, and are mainly used to synchronise the delicate operation of switching, necessary to drive modern computational processes. Unfortunately, this synchronisation process is reaching a natural ``apocalypse''. No longer can clock scaling be used as a blunt tool to accelerate computation, we are up against the natural limits of switching and synchronisation across large processors. Therefore, we need to rethink how time is utilised in computation, using it more naturally in the role of representing data. This can be achieved by using a time interval delineated by discrete start and end events, and by re-casting computational operations into the time domain. With this, computer systems can be developed that are naturally scaleable in time and space, and can use ambient time references built to the best effort of the available technology.
Our ambition is to better manage the energy/computation time trade-off, and to explicitly embed the resolution of the data in the time domain. We aim to recast calculations into the ``for free'' format that time offers, and in addition, perform these calculations at the highest clock or oscillator resolution possible.
△ Less
Submitted 5 February, 2024;
originally announced February 2024.
-
A perspective on physical reservoir computing with nanomagnetic devices
Authors:
Dan A Allwood,
Matthew O A Ellis,
David Griffin,
Thomas J Hayward,
Luca Manneschi,
Mohammad F KH Musameh,
Simon O'Keefe,
Susan Stepney,
Charles Swindells,
Martin A Trefzer,
Eleni Vasilaki,
Guru Venkat,
Ian Vidamour,
Chester Wringe
Abstract:
Neural networks have revolutionized the area of artificial intelligence and introduced transformative applications to almost every scientific field and industry. However, this success comes at a great price; the energy requirements for training advanced models are unsustainable. One promising way to address this pressing issue is by develo** low-energy neuromorphic hardware that directly support…
▽ More
Neural networks have revolutionized the area of artificial intelligence and introduced transformative applications to almost every scientific field and industry. However, this success comes at a great price; the energy requirements for training advanced models are unsustainable. One promising way to address this pressing issue is by develo** low-energy neuromorphic hardware that directly supports the algorithm's requirements. The intrinsic non-volatility, non-linearity, and memory of spintronic devices make them appealing candidates for neuromorphic devices. Here we focus on the reservoir computing paradigm, a recurrent network with a simple training algorithm suitable for computation with spintronic devices since they can provide the properties of non-linearity and memory. We review technologies and methods for develo** neuromorphic spintronic devices and conclude with critical open issues to address before such devices become widely used.
△ Less
Submitted 9 December, 2022;
originally announced December 2022.
-
Reservoir Computing with Magnetic Thin Films
Authors:
Matthew Dale,
David Griffin,
Richard F. L. Evans,
Sarah Jenkins,
Simon O'Keefe,
Angelika Sebald,
Susan Stepney,
Fernando Torre,
Martin Trefzer
Abstract:
Advances in artificial intelligence are driven by technologies inspired by the brain, but these technologies are orders of magnitude less powerful and energy efficient than biological systems. Inspired by the nonlinear dynamics of neural networks, new unconventional computing hardware has emerged with the potential to exploit natural phenomena and gain efficiency, in a similar manner to biological…
▽ More
Advances in artificial intelligence are driven by technologies inspired by the brain, but these technologies are orders of magnitude less powerful and energy efficient than biological systems. Inspired by the nonlinear dynamics of neural networks, new unconventional computing hardware has emerged with the potential to exploit natural phenomena and gain efficiency, in a similar manner to biological systems. Physical reservoir computing demonstrates this with a variety of unconventional systems, from optical-based to memristive systems. Reservoir computers provide a nonlinear projection of the task input into a high-dimensional feature space by exploiting the system's internal dynamics. A trained readout layer then combines features to perform tasks, such as pattern recognition and time-series analysis. Despite progress, achieving state-of-the-art performance without external signal processing to the reservoir remains challenging. Here we perform an initial exploration of three magnetic materials in thin-film geometries via microscale simulation. Our results reveal that basic spin properties of magnetic films generate the required nonlinear dynamics and memory to solve machine learning tasks (although there would be practical challenges in exploiting these particular materials in physical implementations). The method of exploration can be applied to other materials, so this work opens up the possibility of testing different materials, from relatively simple (alloys) to significantly complex (antiferromagnetic reservoirs).
△ Less
Submitted 30 October, 2023; v1 submitted 29 January, 2021;
originally announced January 2021.
-
Deep Learning and Word Embeddings for Tweet Classification for Crisis Response
Authors:
Reem ALRashdi,
Simon O'Keefe
Abstract:
Tradition tweet classification models for crisis response focus on convolutional layers and domain-specific word embeddings. In this paper, we study the application of different neural networks with general-purpose and domain-specific word embeddings to investigate their ability to improve the performance of tweet classification models. We evaluate four tweet classification models on CrisisNLP dat…
▽ More
Tradition tweet classification models for crisis response focus on convolutional layers and domain-specific word embeddings. In this paper, we study the application of different neural networks with general-purpose and domain-specific word embeddings to investigate their ability to improve the performance of tweet classification models. We evaluate four tweet classification models on CrisisNLP dataset and obtain comparable results which indicates that general-purpose word embedding such as GloVe can be used instead of domain-specific word embedding especially with Bi-LSTM where results reported the highest performance of 62.04% F1 score.
△ Less
Submitted 26 March, 2019;
originally announced March 2019.
-
Cognition-inspired homeostasis can balance conflicting needs in robots
Authors:
James Stovold,
Simon O'Keefe,
Jon Timmis
Abstract:
Homeostasis keeps animals alive; it is a fundamental process that allows animals to adapt quickly to their environment. Artificial homeostasis can be used to help robots adapt to changing environments. Previous attempts at develo** artificial homeostasis for robots were driven by mimicry of the biochemical machinery that drives homeostasis in humans. By considering homeostasis from a cognitive p…
▽ More
Homeostasis keeps animals alive; it is a fundamental process that allows animals to adapt quickly to their environment. Artificial homeostasis can be used to help robots adapt to changing environments. Previous attempts at develo** artificial homeostasis for robots were driven by mimicry of the biochemical machinery that drives homeostasis in humans. By considering homeostasis from a cognitive perspective, we develop a comparatively simple robot controller named CogSis (COGnitive HomeostaSIS) and demonstrate that it can provide homeostasis to a robot, even when there are conflicting needs. We present experiments showing that a robot running CogSis is able to learn from previous experiences and use them to influence future behaviour; can maintain its charge level while attending to another task (warming itself in an area separate from the charging station); and is able to maintain its charge level while avoiding a conflicting need (kee** cool, when the charging station is placed in a hot region of the environment). Results are presented in simulation and from a real robot platform.
△ Less
Submitted 27 September, 2022; v1 submitted 25 November, 2018;
originally announced November 2018.
-
Discrete and Continuous Systems of Logic in Nuclear Magnetic Resonance
Authors:
Pedro M. Aguiar,
Robert Hornby,
Cameron McGarry,
Simon O'Keefe,
Angelika Sebald
Abstract:
We implement several non-binary logic systems using the spin dynamics of nuclear spins in nuclear magnetic resonance (NMR). The NMR system is a suitable test system because of its high degree of experimental control; findings from NMR implementations are relevant for other computational platforms exploiting particles with spin, such as electrons or photons. While we do not expect the NMR system to…
▽ More
We implement several non-binary logic systems using the spin dynamics of nuclear spins in nuclear magnetic resonance (NMR). The NMR system is a suitable test system because of its high degree of experimental control; findings from NMR implementations are relevant for other computational platforms exploiting particles with spin, such as electrons or photons. While we do not expect the NMR system to become a practical computational device, it is uniquely useful to explore strengths and weaknesses of unconventional computational approaches, such as non-binary logic.
△ Less
Submitted 1 December, 2015;
originally announced December 2015.
-
Weyl's law and quantum ergodicity for maps with divided phase space
Authors:
Jens Marklof,
Stephen O'Keefe,
Steve Zelditch
Abstract:
For a general class of unitary quantum maps, whose underlying classical phase space is divided into several invariant domains of positive measure, we establish analogues of Weyl's law for the distribution of eigenphases. If the map has one ergodic component, and is periodic on the remaining domains, we prove the Schnirelman-Zelditch-Colin de Verdiere Theorem on the equidistribution of eigenfunct…
▽ More
For a general class of unitary quantum maps, whose underlying classical phase space is divided into several invariant domains of positive measure, we establish analogues of Weyl's law for the distribution of eigenphases. If the map has one ergodic component, and is periodic on the remaining domains, we prove the Schnirelman-Zelditch-Colin de Verdiere Theorem on the equidistribution of eigenfunctions with respect to the ergodic component of the classical map (quantum ergodicity). We apply our main theorems to quantised linked twist maps on the torus. In the Appendix, S. Zelditch connects these studies to some earlier results on `pimpled spheres' in the setting of Riemannian manifolds. The common feature is a divided phase space with a periodic component.
△ Less
Submitted 23 September, 2004; v1 submitted 20 April, 2004;
originally announced April 2004.
-
Intermediate statistics in quantum maps
Authors:
Olivier Giraud,
Jens Marklof,
Stephen O'Keefe
Abstract:
We present a one-parameter family of quantum maps whose spectral statistics are of the same intermediate type as observed in polygonal quantum billiards. Our central result is the evaluation of the spectral two-point correlation form factor at small argument, which in turn yields the asymptotic level compressibility for macroscopic correlation lengths.
We present a one-parameter family of quantum maps whose spectral statistics are of the same intermediate type as observed in polygonal quantum billiards. Our central result is the evaluation of the spectral two-point correlation form factor at small argument, which in turn yields the asymptotic level compressibility for macroscopic correlation lengths.
△ Less
Submitted 17 March, 2004;
originally announced March 2004.
