-
A Categorical Model for Retrosynthetic Reaction Analysis
Authors:
Ella Gale,
Leo Lobski,
Fabio Zanasi
Abstract:
We introduce a mathematical framework for retrosynthetic analysis, an important research method in synthetic chemistry. Our approach represents molecules and their interaction using string diagrams in layered props - a recently introduced categorical model for partial explanations in scientific reasoning. Such principled approach allows one to model features currently not available in automated re…
▽ More
We introduce a mathematical framework for retrosynthetic analysis, an important research method in synthetic chemistry. Our approach represents molecules and their interaction using string diagrams in layered props - a recently introduced categorical model for partial explanations in scientific reasoning. Such principled approach allows one to model features currently not available in automated retrosynthesis tools, such as chirality, reaction environment and protection-deprotection steps.
△ Less
Submitted 7 November, 2023;
originally announced November 2023.
-
Cluster Flow: how a hierarchical clustering layer make allows deep-NNs more resilient to hacking, more human-like and easily implements relational reasoning
Authors:
Ella Gale,
Oliver Matthews
Abstract:
Despite the huge recent breakthroughs in neural networks (NNs) for artificial intelligence (specifically deep convolutional networks) such NNs do not achieve human-level performance: they can be hacked by images that would fool no human and lack `common sense'. It has been argued that a basis of human-level intelligence is mankind's ability to perform relational reasoning: the comparison of differ…
▽ More
Despite the huge recent breakthroughs in neural networks (NNs) for artificial intelligence (specifically deep convolutional networks) such NNs do not achieve human-level performance: they can be hacked by images that would fool no human and lack `common sense'. It has been argued that a basis of human-level intelligence is mankind's ability to perform relational reasoning: the comparison of different objects, measuring similarity, gras** of relations between objects and the converse, figuring out the odd one out in a set of objects. Mankind can even do this with objects they have never seen before. Here we show how ClusterFlow, a semi-supervised hierarchical clustering framework can operate on trained NNs utilising the rich multi-dimensional class and feature data found at the pre-SoftMax layer to build a hyperspacial map of classes/features and this adds more human-like functionality to modern deep convolutional neural networks. We demonstrate this with 3 tasks. 1. the statistical learning based `mistakes' made by infants when attending to images of cats and dogs. 2. improving both the resilience to hacking images and the accurate measure of certainty in deep-NNs. 3. Relational reasoning over sets of images, including those not known to the NN nor seen before. We also demonstrate that ClusterFlow can work on non-NN data and deal with missing data by testing it on a Chemistry dataset. This work suggests that modern deep NNs can be made more human-like without re-training of the NNs. As it is known that some methods used in deep and convolutional NNs are not biologically plausible or perhaps even the best approach: the ClusterFlow framework can sit on top of any NN and will be a useful tool to add as NNs are improved in this regard.
△ Less
Submitted 27 April, 2023;
originally announced April 2023.
-
Icospherical Chemical Objects (ICOs) allow for chemical data augmentation and maintain rotational, translation and permutation invariance
Authors:
Ella Gale
Abstract:
Dataset augmentation is a common way to deal with small datasets; Chemistry datasets are often small. Spherical convolutional neural networks (SphNNs) and Icosahedral neural networks (IcoNNs) are a type of geometric machine learning algorithm that maintains rotational symmetry. Molecular structure has rotational invariance and is inherently 3-D, and thus we need 3-D encoding methods to input molec…
▽ More
Dataset augmentation is a common way to deal with small datasets; Chemistry datasets are often small. Spherical convolutional neural networks (SphNNs) and Icosahedral neural networks (IcoNNs) are a type of geometric machine learning algorithm that maintains rotational symmetry. Molecular structure has rotational invariance and is inherently 3-D, and thus we need 3-D encoding methods to input molecular structure into machine learning. In this paper I present Icospherical Chemical Objects (ICOs) that enable the encoding of 3-D data in a rotationally invariant way which works with spherical or icosahedral neural networks and allows for dataset augmentation. I demonstrate the ICO featurisation method on the following tasks: predicting general molecular properties, predicting solubility of drug like molecules and the protein binding problem and find that ICO and SphNNs perform well on all problems.
△ Less
Submitted 15 April, 2023;
originally announced April 2023.
-
Shape is (almost) all!: Persistent homology features (PHFs) are an information rich input for efficient molecular machine learning
Authors:
Ella Gale
Abstract:
3-D shape is important to chemistry, but how important? Machine learning works best when the inputs are simple and match the problem well. Chemistry datasets tend to be very small compared to those generally used in machine learning so we need to get the most from each datapoint. Persistent homology measures the topological shape properties of point clouds at different scales and is used in topolo…
▽ More
3-D shape is important to chemistry, but how important? Machine learning works best when the inputs are simple and match the problem well. Chemistry datasets tend to be very small compared to those generally used in machine learning so we need to get the most from each datapoint. Persistent homology measures the topological shape properties of point clouds at different scales and is used in topological data analysis. Here we investigate what persistent homology captures about molecular structure and create persistent homology features (PHFs) that encode a molecule's shape whilst losing most of the symbolic detail like atom labels, valence, charge, bonds etc. We demonstrate the usefulness of PHFs on a series of chemical datasets: QM7, lipophilicity, Delaney and Tox21. PHFs work as well as the best benchmarks. PHFs are very information dense and much smaller than other encoding methods yet found, meaning ML algorithms are much more energy efficient. PHFs success despite losing a large amount of chemical detail highlights how much of chemistry can be simplified to topological shape.
△ Less
Submitted 15 April, 2023;
originally announced April 2023.
-
ASL-Homework-RGBD Dataset: An annotated dataset of 45 fluent and non-fluent signers performing American Sign Language homeworks
Authors:
Saad Hassan,
Matthew Seita,
Larwan Berke,
Yingli Tian,
Elaine Gale,
Sooyeon Lee,
Matt Huenerfauth
Abstract:
We are releasing a dataset containing videos of both fluent and non-fluent signers using American Sign Language (ASL), which were collected using a Kinect v2 sensor. This dataset was collected as a part of a project to develop and evaluate computer vision algorithms to support new technologies for automatic detection of ASL fluency attributes. A total of 45 fluent and non-fluent participants were…
▽ More
We are releasing a dataset containing videos of both fluent and non-fluent signers using American Sign Language (ASL), which were collected using a Kinect v2 sensor. This dataset was collected as a part of a project to develop and evaluate computer vision algorithms to support new technologies for automatic detection of ASL fluency attributes. A total of 45 fluent and non-fluent participants were asked to perform signing homework assignments that are similar to the assignments used in introductory or intermediate level ASL courses. The data is annotated to identify several aspects of signing including grammatical features and non-manual markers. Sign language recognition is currently very data-driven and this dataset can support the design of recognition technologies, especially technologies that can benefit ASL learners. This dataset might also be interesting to ASL education researchers who want to contrast fluent and non-fluent signing.
△ Less
Submitted 8 July, 2022;
originally announced July 2022.
-
Are there any 'object detectors' in the hidden layers of CNNs trained to identify objects or scenes?
Authors:
Ella M. Gale,
Nicholas Martin,
Ryan Blything,
Anh Nguyen,
Jeffrey S. Bowers
Abstract:
Various methods of measuring unit selectivity have been developed with the aim of better understanding how neural networks work. But the different measures provide divergent estimates of selectivity, and this has led to different conclusions regarding the conditions in which selective object representations are learned and the functional relevance of these representations. In an attempt to better…
▽ More
Various methods of measuring unit selectivity have been developed with the aim of better understanding how neural networks work. But the different measures provide divergent estimates of selectivity, and this has led to different conclusions regarding the conditions in which selective object representations are learned and the functional relevance of these representations. In an attempt to better characterize object selectivity, we undertake a comparison of various selectivity measures on a large set of units in AlexNet, including localist selectivity, precision, class-conditional mean activity selectivity (CCMAS), network dissection,the human interpretation of activation maximization (AM) images, and standard signal-detection measures. We find that the different measures provide different estimates of object selectivity, with precision and CCMAS measures providing misleadingly high estimates. Indeed, the most selective units had a poor hit-rate or a high false-alarm rate (or both) in object classification, making them poor object detectors. We fail to find any units that are even remotely as selective as the 'grandmother cell' units reported in recurrent neural networks. In order to generalize these results, we compared selectivity measures on units in VGG-16 and GoogLeNet trained on the ImageNet or Places-365 datasets that have been described as 'object detectors'. Again, we find poor hit-rates and high false-alarm rates for object classification. We conclude that signal-detection measures provide a better assessment of single-unit selectivity compared to common alternative approaches, and that deep convolutional networks of image classification do not learn object detectors in their hidden layers.
△ Less
Submitted 2 July, 2020;
originally announced July 2020.
-
Subtle Sensing: Detecting Differences in the Flexibility of Virtually Simulated Molecular Objects
Authors:
Rhoslyn Roebuck Williams,
Xan Varcoe,
Becca R. Glowacki,
Ella M. Gale,
Alexander Jamieson-Binnie,
David R. Glowacki
Abstract:
During VR demos we have performed over last few years, many participants (in the absence of any haptic feedback) have commented on their perceived ability to 'feel' differences between simulated molecular objects. The mechanisms for such 'feeling' are not entirely clear: observing from outside VR, one can see that there is nothing physical for participants to 'feel'. Here we outline exploratory us…
▽ More
During VR demos we have performed over last few years, many participants (in the absence of any haptic feedback) have commented on their perceived ability to 'feel' differences between simulated molecular objects. The mechanisms for such 'feeling' are not entirely clear: observing from outside VR, one can see that there is nothing physical for participants to 'feel'. Here we outline exploratory user studies designed to evaluate the extent to which participants can distinguish quantitative differences in the flexibility of VR-simulated molecular objects. The results suggest that an individual's capacity to detect differences in molecular flexibility is enhanced when they can interact with and manipulate the molecules, as opposed to merely observing the same interaction. Building on these results, we intend to carry out further studies investigating humans' ability to sense quantitative properties of VR simulations without haptic technology.
△ Less
Submitted 7 May, 2020;
originally announced May 2020.
-
Isness: Using Multi-Person VR to Design Peak Mystical-Type Experiences Comparable to Psychedelics
Authors:
David R. Glowacki,
Mark D. Wonnacott,
Rachel Freire,
Becca R. Glowacki,
Ella M. Gale,
James E. Pike,
Tiu de Haan,
Mike Chatziapostolou,
Oussama Metatla
Abstract:
Studies combining psychotherapy with psychedelic drugs (PsiDs) have demonstrated positive outcomes that are often associated with PsiDs' ability to induce 'mystical-type' experiences (MTEs) - i.e., subjective experiences whose characteristics include a sense of connectedness, transcendence, and ineffability. We suggest that both PsiDs and virtual reality can be situated on a broader spectrum of ps…
▽ More
Studies combining psychotherapy with psychedelic drugs (PsiDs) have demonstrated positive outcomes that are often associated with PsiDs' ability to induce 'mystical-type' experiences (MTEs) - i.e., subjective experiences whose characteristics include a sense of connectedness, transcendence, and ineffability. We suggest that both PsiDs and virtual reality can be situated on a broader spectrum of psychedelic technologies. To test this hypothesis, we used concepts, methods, and analysis strategies from PsiD research to design and evaluate 'Isness', a multi-person VR journey where participants experience the collective emergence, fluctuation, and dissipation of their bodies as energetic essences. A study (N=57) analyzing participant responses to a commonly used PsiD experience questionnaire (MEQ30) indicates that Isness participants had MTEs comparable to those reported in double-blind clinical studies after high doses of psilocybin & LSD. Within a supportive setting and conceptual framework, VR phenomenology can create the conditions for MTEs from which participants derive insight and meaning.
△ Less
Submitted 5 May, 2020; v1 submitted 3 February, 2020;
originally announced February 2020.
-
Short and Wide Network Paths
Authors:
Lavanya Marla,
Lav R. Varshney,
Devavrat Shah,
Nirmal A. Prakash,
Michael E. Gale
Abstract:
Network flow is a powerful mathematical framework to systematically explore the relationship between structure and function in biological, social, and technological networks. We introduce a new pipelining model of flow through networks where commodities must be transported over single paths rather than split over several paths and recombined. We show this notion of pipelined network flow is optimi…
▽ More
Network flow is a powerful mathematical framework to systematically explore the relationship between structure and function in biological, social, and technological networks. We introduce a new pipelining model of flow through networks where commodities must be transported over single paths rather than split over several paths and recombined. We show this notion of pipelined network flow is optimized using network paths that are both short and wide, and develop efficient algorithms to compute such paths for given pairs of nodes and for all-pairs. Short and wide paths are characterized for many real-world networks. To further demonstrate the utility of this network characterization, we develop novel information-theoretic lower bounds on computation speed in nervous systems due to limitations from anatomical connectivity and physical noise. For the nematode Caenorhabditis elegans, we find these bounds are predictive of biological timescales of behavior. Further, we find the particular C. elegans connectome is globally less efficient for information flow than random networks, but the hub-and-spoke architecture of functional subcircuits is optimal under constraint on number of synapses. This suggests functional subcircuits are a primary organizational principle of this small invertebrate nervous system.
△ Less
Submitted 1 November, 2019;
originally announced November 2019.
-
My, and others', spiking memristors are true memristors: a response to R.S. Williams' question at the New Memory Paradigms: Memristive Phenomena and Neuromorphic Applications Faraday Discussion
Authors:
Ella M. Gale
Abstract:
At the Faraday Discussion, in the paper titled `Neuromorphic computation with spiking memristors: habituation, experimental instantiation of logic gates and a novel sequence-sensitive perceptron model' it was demonstrated that a large amount of computation could be done in a sequential way using memristor current spikes (d.c. response). As these spikes are found in many memristors (possibly all),…
▽ More
At the Faraday Discussion, in the paper titled `Neuromorphic computation with spiking memristors: habituation, experimental instantiation of logic gates and a novel sequence-sensitive perceptron model' it was demonstrated that a large amount of computation could be done in a sequential way using memristor current spikes (d.c. response). As these spikes are found in many memristors (possibly all), this novel approach could be highly useful for fast and reproducible memristor circuits. However, questions were raised as to whether these spikes were actually due to memristance or merely capacitance in the circuit. In this longer version of the Faraday Discussion response, as much information as is available from both published and unpublished data from my lab is marshalled together. We find that the devices are likely imperfect memristors with some capacitance, and that the spikes are related to the frequency effect seen in memristor hysteresis curves, thus are an integral part of memristance.
△ Less
Submitted 14 December, 2018;
originally announced December 2018.
-
When and where do feed-forward neural networks learn localist representations?
Authors:
Ella M. Gale,
Nicolas Martin,
Jeffrey S. Bowers
Abstract:
According to parallel distributed processing (PDP) theory in psychology, neural networks (NN) learn distributed rather than interpretable localist representations. This view has been held so strongly that few researchers have analysed single units to determine if this assumption is correct. However, recent results from psychology, neuroscience and computer science have shown the occasional existen…
▽ More
According to parallel distributed processing (PDP) theory in psychology, neural networks (NN) learn distributed rather than interpretable localist representations. This view has been held so strongly that few researchers have analysed single units to determine if this assumption is correct. However, recent results from psychology, neuroscience and computer science have shown the occasional existence of local codes emerging in artificial and biological neural networks. In this paper, we undertake the first systematic survey of when local codes emerge in a feed-forward neural network, using generated input and output data with known qualities. We find that the number of local codes that emerge from a NN follows a well-defined distribution across the number of hidden layer neurons, with a peak determined by the size of input data, number of examples presented and the sparsity of input data. Using a 1-hot output code drastically decreases the number of local codes on the hidden layer. The number of emergent local codes increases with the percentage of dropout applied to the hidden layer, suggesting that the localist encoding may offer a resilience to noisy networks. This data suggests that localist coding can emerge from feed-forward PDP networks and suggests some of the conditions that may lead to interpretable localist representations in the cortex. The findings highlight how local codes should not be dismissed out of hand.
△ Less
Submitted 11 June, 2018;
originally announced June 2018.
-
Spiking memristor logic gates are a type of time-variant perceptron
Authors:
Ella M. Gale
Abstract:
Memristors are low-power memory-holding resistors thought to be useful for neuromophic computing, which can compute via spike-interactions mediated through the device's short-term memory. Using interacting spikes, it is possible to build an AND gate that computes OR at the same time, similarly a full adder can be built that computes the arithmetical sum of its inputs. Here we show how these gates…
▽ More
Memristors are low-power memory-holding resistors thought to be useful for neuromophic computing, which can compute via spike-interactions mediated through the device's short-term memory. Using interacting spikes, it is possible to build an AND gate that computes OR at the same time, similarly a full adder can be built that computes the arithmetical sum of its inputs. Here we show how these gates can be understood by modelling the memristors as a novel type of perceptron: one which is sensitive to input order. The memristor's memory can change the input weights for later inputs, and thus the memristor gates cannot be accurately described by a single perceptron, requiring either a network of time-invarient perceptrons or a complex time-varying self-reprogrammable perceptron. This work demonstrates the high functionality of memristor logic gates, and also that the addition of theasholding could enable the creation of a standard perceptron in hardware, which may have use in building neural net chips.
△ Less
Submitted 8 January, 2018;
originally announced January 2018.
-
Non-ideal memristors for a non-ideal world
Authors:
Ella Gale
Abstract:
Memristors have pinched hysteresis loops in the $V-I$ plane. Ideal memristors are everywhere non-linear, cross at zero and are rotationally symmetric. In this paper we extend memristor theory to produce different types of non-ideality and find that: including a background current (such as an ionic current) moves the crossing point away from zero; including a degradation resistance (that increases…
▽ More
Memristors have pinched hysteresis loops in the $V-I$ plane. Ideal memristors are everywhere non-linear, cross at zero and are rotationally symmetric. In this paper we extend memristor theory to produce different types of non-ideality and find that: including a background current (such as an ionic current) moves the crossing point away from zero; including a degradation resistance (that increases with experimental time) leads to an asymmetry; modelling a low resistance filament in parallel describes triangular $V-I$ curves with a straight-line low resistance state. A novel measurement of hysteresis asymmetry was introduced based on hysteresis and it was found that which lobe was bigger depended on the size of the breaking current relative to the memristance. The hysteresis varied differently with each type of non-ideality, suggesting that measurements of several device I-V curves and calculation of these parameters could give an indication of the underlying mechanism.
△ Less
Submitted 9 November, 2016;
originally announced November 2016.
-
TiO$_2$-based Memristors and ReRAM: Materials, Mechanisms and Models (a Review)
Authors:
Ella Gale
Abstract:
The memristor is the fundamental non-linear circuit element, with uses in computing and computer memory. ReRAM (Resistive Random Access Memory) is a resistive switching memory proposed as a non-volatile memory. In this review we shall summarise the state of the art for these closely-related fields, concentrating on titanium dioxide, the well-utilised and archetypal material for both. We shall cove…
▽ More
The memristor is the fundamental non-linear circuit element, with uses in computing and computer memory. ReRAM (Resistive Random Access Memory) is a resistive switching memory proposed as a non-volatile memory. In this review we shall summarise the state of the art for these closely-related fields, concentrating on titanium dioxide, the well-utilised and archetypal material for both. We shall cover material properties, switching mechanisms and models to demonstrate what ReRAM and memristor scientists can learn from each other and examine the outlook for these technologies.
△ Less
Submitted 9 November, 2016;
originally announced November 2016.
-
Single Memristor Logic Gates: From NOT to a Full Adder
Authors:
Ella Gale
Abstract:
Memristors have been suggested as a novel route to neuromorphic computing based on the similarity between them and neurons (specifically synapses and ion pumps). The d.c. action of the memristor is a current spike which imparts a short-term memory to the device. Here it is demonstrated that this short-term memory works exactly like habituation (e.g. in \emph{Aplysia}). We elucidate the physical ru…
▽ More
Memristors have been suggested as a novel route to neuromorphic computing based on the similarity between them and neurons (specifically synapses and ion pumps). The d.c. action of the memristor is a current spike which imparts a short-term memory to the device. Here it is demonstrated that this short-term memory works exactly like habituation (e.g. in \emph{Aplysia}). We elucidate the physical rules, based on energy conservation, governing the interaction of these current spikes: summation, `bounce-back', directionality and `diminishing returns'. Using these rules, we introduce 4 different logical systems to implement sequential logic in the memristor and demonstrate how sequential logic works by instantiating a NOT gate, an AND gate, an XOR gate and a Full Adder with a single memristor. The Full Adder makes use of the memristor's short-term memory to add together three binary values and outputs the sum, the carry digit and even the order they were input in. A memristor full adder also outputs the arithmetical sum of bits, allowing for a logically (but not physically) reversible system. Essentially, we can replace an input/output port with an extra time-step, allowing a single memristor to do a hither-to unexpectedly large amount of computation. This makes up for the memristor's slow operation speed and may relate to how neurons do a similarly-large computation with such slow operations speeds. We propose that using spiking logic, either in gates or as neuron-analogues, with plastic rewritable connections between them, would allow the building of a neuromorphic computer.
△ Less
Submitted 19 October, 2015;
originally announced October 2015.
-
Evolving Spiking Networks with Variable Resistive Memories
Authors:
Gerard David Howard,
Larry Bull,
Ben de Lacy Costello,
Andrew Adamatzky,
Ella Gale
Abstract:
Neuromorphic computing is a brainlike information processing paradigm that requires adaptive learning mechanisms. A spiking neuro-evolutionary system is used for this purpose; plastic resistive memories are implemented as synapses in spiking neural networks. The evolutionary design process exploits parameter self-adaptation and allows the topology and synaptic weights to be evolved for each networ…
▽ More
Neuromorphic computing is a brainlike information processing paradigm that requires adaptive learning mechanisms. A spiking neuro-evolutionary system is used for this purpose; plastic resistive memories are implemented as synapses in spiking neural networks. The evolutionary design process exploits parameter self-adaptation and allows the topology and synaptic weights to be evolved for each network in an autonomous manner. Variable resistive memories are the focus of this research; each synapse has its own conductance profile which modifies the plastic behaviour of the device and may be altered during evolution. These variable resistive networks are evaluated on a noisy robotic dynamic-reward scenario against two static resistive memories and a system containing standard connections only. Results indicate that the extra behavioural degrees of freedom available to the networks incorporating variable resistive memories enable them to outperform the comparative synapse types.
△ Less
Submitted 17 May, 2015;
originally announced May 2015.
-
Experimental evidence of chaos from memristors
Authors:
L. V. Gambuzza,
L. Fortuna,
M. Frasca,
E. Gale
Abstract:
Until now, most memristor-based chaotic circuits proposed in the literature are based on mathematical models which assume ideal characteristics such as piece-wise linear or cubic non-linearities. The idea, illustrated here and originating from the experimental approach for device characterization, is to realize a chaotic system exploiting the non-linearity of only one memristor with a very simple…
▽ More
Until now, most memristor-based chaotic circuits proposed in the literature are based on mathematical models which assume ideal characteristics such as piece-wise linear or cubic non-linearities. The idea, illustrated here and originating from the experimental approach for device characterization, is to realize a chaotic system exploiting the non-linearity of only one memristor with a very simple experimental set-up using feedback. In this way a simple circuit is obtained and chaos is experimentally observed and is confirmed by the calculation of the largest Lyapunov exponent. Numerical results using the Strukov model support the existence of robust chaos in our circuit. This is the first experimental demonstration of chaos in a real memristor circuit and suggests that memristors are well placed for hardware encryption.
△ Less
Submitted 24 April, 2015;
originally announced April 2015.
-
Uniform and Piece-wise Uniform Fields in Memristor Models
Authors:
Ella Gale
Abstract:
The Strukov model was the phenomenological model that accompanied the announcement of the first recognised physical instantiation of the memristor and, as such, it has been widely used. This model described the motion of a boundary, $w$, between two types of inter-converting material, $R_{\mathrm{off}}$ and $R_{\mathrm{on}}$, seemingly under a uniform field across the entire device. In fact, what…
▽ More
The Strukov model was the phenomenological model that accompanied the announcement of the first recognised physical instantiation of the memristor and, as such, it has been widely used. This model described the motion of a boundary, $w$, between two types of inter-converting material, $R_{\mathrm{off}}$ and $R_{\mathrm{on}}$, seemingly under a uniform field across the entire device. In fact, what was intended was a field with a discontinuity at $w$, that was uniform between $0<x<w$. In this paper we show that the discontinuity is required for the Strukov model derivation to be completed, and thus the derivation as given does not describe a situation with a uniform field across the entire device. The discontinuity can be described as a Heaviside function, $H$, located on $w$, for which there are three common single-valued approximations for $H(w)$. The Strukov model as intended includes an approximation for the Heaviside function (the field is taken to be the same as that across the $R_{\mathrm{on}}$ part of the device). We compare approximations and give solutions. We then extend the description of the field to a more-realistic continuously varying sigmoidal transition between two uniform fields and demonstrate that the centro-symmetric approximation model (taking the field as being the average of the fields across $R_{\mathrm{on}}$ and $R_{\mathrm{off}}$) is a better single-point model of that situation: the other two approximations over or underestimate the field.
△ Less
Submitted 23 April, 2014; v1 submitted 22 April, 2014;
originally announced April 2014.
-
Boolean Logic Gates From A Single Memristor Via Low-Level Sequential Logic
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
By using the memristor's memory to both store a bit and perform an operation with a second input bit, simple Boolean logic gates have been built with a single memristor. The operation makes use of the interaction of current spikes (occasionally called current transients) found in both memristors and other devices. The sequential time-based logic methodology allows two logical input bits to be used…
▽ More
By using the memristor's memory to both store a bit and perform an operation with a second input bit, simple Boolean logic gates have been built with a single memristor. The operation makes use of the interaction of current spikes (occasionally called current transients) found in both memristors and other devices. The sequential time-based logic methodology allows two logical input bits to be used on a one-port by sending the bits separated in time. The resulting logic gate is faster than one relying on memristor's state switching, low power and requires only one memristor. We experimentally demonstrate working OR and XOR gates made with a single flexible Titanium dioxide sol-gel memristor.
△ Less
Submitted 17 February, 2014;
originally announced February 2014.
-
Is Spiking Logic the Route to Memristor-Based Computers?
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
Memristors have been suggested as a novel route to neuromorphic computing based on the similarity between neurons (synapses and ion pumps) and memristors. The D.C. action of the memristor is a current spike, which we think will be fruitful for building memristor computers. In this paper, we introduce 4 different logical assignations to implement sequential logic in the memristor and introduce the…
▽ More
Memristors have been suggested as a novel route to neuromorphic computing based on the similarity between neurons (synapses and ion pumps) and memristors. The D.C. action of the memristor is a current spike, which we think will be fruitful for building memristor computers. In this paper, we introduce 4 different logical assignations to implement sequential logic in the memristor and introduce the physical rules, summation, `bounce-back', directionality and `diminishing returns', elucidated from our investigations. We then demonstrate how memristor sequential logic works by instantiating a NOT gate, an AND gate and a Full Adder with a single memristor. The Full Adder makes use of the memristor's memory to add three binary values together and outputs the value, the carry digit and even the order they were input in.
△ Less
Submitted 17 February, 2014;
originally announced February 2014.
-
Connecting Spiking Neurons to a Spiking Memristor Network Changes the Memristor Dynamics
Authors:
Deborah Gater,
Attya Iqbal,
Jeffrey Davey,
Ella Gale
Abstract:
Memristors have been suggested as neuromorphic computing elements. Spike-time dependent plasticity and the Hodgkin-Huxley model of the neuron have both been modelled effectively by memristor theory. The d.c. response of the memristor is a current spike. Based on these three facts we suggest that memristors are well-placed to interface directly with neurons. In this paper we show that connecting a…
▽ More
Memristors have been suggested as neuromorphic computing elements. Spike-time dependent plasticity and the Hodgkin-Huxley model of the neuron have both been modelled effectively by memristor theory. The d.c. response of the memristor is a current spike. Based on these three facts we suggest that memristors are well-placed to interface directly with neurons. In this paper we show that connecting a spiking memristor network to spiking neuronal cells causes a change in the memristor network dynamics by: removing the memristor spikes, which we show is due to the effects of connection to aqueous medium; causing a change in current decay rate consistent with a change in memristor state; presenting more-linear $I-t$ dynamics; and increasing the memristor spiking rate, as a consequence of interaction with the spiking neurons. This demonstrates that neurons are capable of communicating directly with memristors, without the need for computer translation.
△ Less
Submitted 17 February, 2014;
originally announced February 2014.
-
The Short-term Memory (D.C. Response) of the Memristor Demonstrates the Causes of the Memristor Frequency Effect
Authors:
Ella Gale,
Ben de Lacy Costello,
Victor Erokhin,
Andrew Adamatzky
Abstract:
A memristor is often identified by showing its distinctive pinched hysteresis curve and testing for the effect of frequency. The hysteresis size should relate to frequency and shrink to zero as the frequency approaches infinity. Although mathematically understood, the material causes for this are not well known. The d.c. response of the memristor is a decaying curve with its own timescale. We show…
▽ More
A memristor is often identified by showing its distinctive pinched hysteresis curve and testing for the effect of frequency. The hysteresis size should relate to frequency and shrink to zero as the frequency approaches infinity. Although mathematically understood, the material causes for this are not well known. The d.c. response of the memristor is a decaying curve with its own timescale. We show via mathematical reasoning that this decaying curve when transformed to a.c. leads to the frequency effect by considering a descretized curve. We then demonstrate the validity of this approach with experimental data from two different types of memristors.
△ Less
Submitted 17 February, 2014;
originally announced February 2014.
-
Does the D.C. Response of Memristors Allow Robotic Short-Term Memory and a Possible Route to Artificial Time Perception?
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
Time perception is essential for task switching, and in the mammalian brain appears alongside other processes. Memristors are electronic components used as synapses and as models for neurons. The d.c. response of memristors can be considered as a type of short-term memory. Interactions of the memristor d.c. response within networks of memristors leads to the emergence of oscillatory dynamics and i…
▽ More
Time perception is essential for task switching, and in the mammalian brain appears alongside other processes. Memristors are electronic components used as synapses and as models for neurons. The d.c. response of memristors can be considered as a type of short-term memory. Interactions of the memristor d.c. response within networks of memristors leads to the emergence of oscillatory dynamics and intermittent spike trains, which are similar to neural dynamics. Based on this data, the structure of a memristor network control for a robot as it undergoes task switching is discussed and it is suggested that these emergent network dynamics could improve the performance of role switching and learning in an artificial intelligence and perhaps create artificial time perception.
△ Less
Submitted 17 February, 2014;
originally announced February 2014.
-
Design of a Hybrid Robot Control System using Memristor-Model and Ant-Inspired Based Information Transfer Protocols
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
It is not always possible for a robot to process all the information from its sensors in a timely manner and thus quick and yet valid approximations of the robot's situation are needed. Here we design hybrid control for a robot within this limit using algorithms inspired by ant worker placement behaviour and based on memristor-based non-linearity.
It is not always possible for a robot to process all the information from its sensors in a timely manner and thus quick and yet valid approximations of the robot's situation are needed. Here we design hybrid control for a robot within this limit using algorithms inspired by ant worker placement behaviour and based on memristor-based non-linearity.
△ Less
Submitted 17 February, 2014;
originally announced February 2014.
-
Which Memristor Theory is Best for Relating Devices Properties to Memristive Function?
Authors:
Ella M. Gale,
Benjamin de Lacy Costello,
Andrew Adamatzky
Abstract:
There are three theoretical models which purport to relate experimentally-measurable or fabrication-controllable device properties to the memristor's operation: 1. Strukov et al's phenomenological model; 2. Georgiou et al's Bernoulli rewrite of that phenomenological model; 3. Gale's memory-conservation model. They differ in their prediction of the effect on memristance of changing the electrode si…
▽ More
There are three theoretical models which purport to relate experimentally-measurable or fabrication-controllable device properties to the memristor's operation: 1. Strukov et al's phenomenological model; 2. Georgiou et al's Bernoulli rewrite of that phenomenological model; 3. Gale's memory-conservation model. They differ in their prediction of the effect on memristance of changing the electrode size and factors that affect the hysteresis. Using a batch of TiO$_2$ sol-gel memristors fabricated with different top electrode widths we test and compare these three theories. It was found that, contrary to model 2's prediction, the `dimensionless lumped parameter', $β$, did not correlate to any measure of the hysteresis. Contrary to model 1, memristance was found to be dependent on the three spatial dimensions of the TiO$_2$ layer, as was predicted by model 3. Model 3 was found to fit the change in resistance value with electrode size. Simulations using model 3 and experimentally derived values for contact resistance gave hysteresis values that were linearly related to (and only one order of magnitude out) from the experimentally-measured values. Memristor hysteresis was found to be related to the ON state resistance and thus the electrode size (as those two are related). These results offer a verification of the memory-conservation theory of memristance and its association of the vacancy magnetic flux with the missing magnetic flux in memristor theory. This is the first paper to experimentally test various theories pertaining to the operation of memristor devices.
△ Less
Submitted 16 December, 2013;
originally announced December 2013.
-
Slime Mould Memristors
Authors:
Ella Gale,
Andrew Adamatzky,
Ben de Lacy Costello
Abstract:
In laboratory experiments we demonstrate that protoplasmic tubes of acellular slime mould \emph{Physarum polycephalum} show current versus voltage profiles consistent with memristive systems and that the effect is due to the living protoplasm of the mould. This complements previous findings on memristive properties of other living systems (human skin and blood) and contributes to development of se…
▽ More
In laboratory experiments we demonstrate that protoplasmic tubes of acellular slime mould \emph{Physarum polycephalum} show current versus voltage profiles consistent with memristive systems and that the effect is due to the living protoplasm of the mould. This complements previous findings on memristive properties of other living systems (human skin and blood) and contributes to development of self-growing bio-electronic circuits. Distinctive asymmetric $V$-$I$ curves which were occasionally observed when the internal current is on the same order as the driven current, are well-modelled by the concept of active memristors.
△ Less
Submitted 2 April, 2014; v1 submitted 14 June, 2013;
originally announced June 2013.
-
Comparison of Ant-Inspired Gatherer Allocation Approaches using Memristor-Based Environmental Models
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
Memristors are used to compare three gathering techniques in an already-mapped environment where resource locations are known. The All Site model, which apportions gatherers based on the modeled memristance of that path, proves to be good at increasing overall efficiency and decreasing time to fully deplete an environment, however it only works well when the resources are of similar quality. The L…
▽ More
Memristors are used to compare three gathering techniques in an already-mapped environment where resource locations are known. The All Site model, which apportions gatherers based on the modeled memristance of that path, proves to be good at increasing overall efficiency and decreasing time to fully deplete an environment, however it only works well when the resources are of similar quality. The Leaf Cutter method, based on Leaf Cutter Ant behaviour, assigns all gatherers first to the best resource, and once depleted, uses the All Site model to spread them out amongst the rest. The Leaf Cutter model is better at increasing resource influx in the short-term and vastly out-performs the All Site model in a more varied environments. It is demonstrated that memristor based abstractions of gatherer models provide potential methods for both the comparison and implementation of agent controls.
△ Less
Submitted 4 February, 2013;
originally announced February 2013.
-
Beyond Markov Chains, Towards Adaptive Memristor Network-based Music Generation
Authors:
Ella Gale,
Oliver Matthews,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
We undertook a study of the use of a memristor network for music generation, making use of the memristor's memory to go beyond the Markov hypothesis. Seed transition matrices are created and populated using memristor equations, and which are shown to generate musical melodies and change in style over time as a result of feedback into the transition matrix. The spiking properties of simple memristo…
▽ More
We undertook a study of the use of a memristor network for music generation, making use of the memristor's memory to go beyond the Markov hypothesis. Seed transition matrices are created and populated using memristor equations, and which are shown to generate musical melodies and change in style over time as a result of feedback into the transition matrix. The spiking properties of simple memristor networks are demonstrated and discussed with reference to applications of music making. The limitations of simulating composing memristor networks in von Neumann hardware is discussed and a hardware solution based on physical memristor properties is presented.
△ Less
Submitted 4 February, 2013;
originally announced February 2013.
-
Evolution of Plastic Learning in Spiking Networks via Memristive Connections
Authors:
Gerard Howard,
Ella Gale,
Larry Bull,
Ben de Lacy Costello,
Andy Adamatzky
Abstract:
This article presents a spiking neuroevolutionary system which implements memristors as plastic connections, i.e. whose weights can vary during a trial. The evolutionary design process exploits parameter self-adaptation and variable topologies, allowing the number of neurons, connection weights, and inter-neural connectivity pattern to emerge. By comparing two phenomenological real-world memristor…
▽ More
This article presents a spiking neuroevolutionary system which implements memristors as plastic connections, i.e. whose weights can vary during a trial. The evolutionary design process exploits parameter self-adaptation and variable topologies, allowing the number of neurons, connection weights, and inter-neural connectivity pattern to emerge. By comparing two phenomenological real-world memristor implementations with networks comprised of (i) linear resistors (ii) constant-valued connections, we demonstrate that this approach allows the evolution of networks of appropriate complexity to emerge whilst exploiting the memristive properties of the connections to reduce learning time. We extend this approach to allow for heterogeneous mixtures of memristors within the networks; our approach provides an in-depth analysis of network structure. Our networks are evaluated on simulated robotic navigation tasks; results demonstrate that memristive plasticity enables higher performance than constant-weighted connections in both static and dynamic reward scenarios, and that mixtures of memristive elements provide performance advantages when compared to homogeneous memristive networks.
△ Less
Submitted 14 December, 2012;
originally announced December 2012.
-
Emergent Spiking in Non-Ideal Memristor Networks
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
Memristors have uses as artificial synapses and perform well in this role in simulations with artificial spiking neurons. Our experiments show that memristor networks natively spike and can exhibit emergent oscillations and bursting spikes. Networks of near-ideal memristors exhibit behaviour similar to a single memristor and combine in circuits like resistors do. Spiking is more likely when filame…
▽ More
Memristors have uses as artificial synapses and perform well in this role in simulations with artificial spiking neurons. Our experiments show that memristor networks natively spike and can exhibit emergent oscillations and bursting spikes. Networks of near-ideal memristors exhibit behaviour similar to a single memristor and combine in circuits like resistors do. Spiking is more likely when filamentary memristors are used or the circuits have a higher degree of compositional complexity (i.e. a larger number of anti-series or anti-parallel interactions). 3-memristor circuits with the same memristor polarity (low compositional complexity) are stabilised and do not show spiking behaviour. 3-memristor circuits with anti-series and/or anti-parallel compositions show richer and more complex dynamics than 2-memristor spiking circuits. We show that the complexity of these dynamics can be quantified by calculating (using partial auto-correlation functions) the minimum order auto-regression function that could fit it. We propose that these oscillations and spikes may be similar phenomena to brainwaves and neural spike trains and suggest that these behaviours can be used to perform neuromorphic computation.
△ Less
Submitted 10 March, 2014; v1 submitted 30 October, 2012;
originally announced October 2012.
-
The Effect of Electrode Size on Memristor Properties: An Experimental and Theoretical Study
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
The width of the electrodes is not included in the current phenomenological models of memristance, but is included in the memory-conservation (mem-con) theory of memristance. An experimental study of the effect of changing the top electrode width was performed on titanium dioxide sol-gel memristors. It was demonstrated that both the on resistance, Ron, and the off resistance, Roff, decreased with…
▽ More
The width of the electrodes is not included in the current phenomenological models of memristance, but is included in the memory-conservation (mem-con) theory of memristance. An experimental study of the effect of changing the top electrode width was performed on titanium dioxide sol-gel memristors. It was demonstrated that both the on resistance, Ron, and the off resistance, Roff, decreased with increasing electrode size. The memory function part of the mem-con model could fit the relationship between Ron and electrode size. Similarly, the conservation function fits the change in Roff. The experimentally measured hysteresis did not fit the phenomenological model's predictions. Instead the size of the hysteresis increased with increasing electrode size, and correlated well to decreasing Ron.
△ Less
Submitted 30 July, 2012;
originally announced July 2012.
-
Filamentary Extension of the Mem-Con theory of Memristance and its Application to Titanium Dioxide Sol-Gel Memristors
Authors:
Ella Gale,
Ben de Lacy Costello,
Andrew Adamatzky
Abstract:
Titanium dioxide sol-gel memristors have two different modes of operation, believed to be dependent on whether there is bulk memristance, i.e. memristance throughout the whole volume or filamentary memristance, i.e. memristance caused by the connection of conducting filaments. The mem-con theory of memristance is based on the drift of oxygen vacancies rather than that of conducting electrons and h…
▽ More
Titanium dioxide sol-gel memristors have two different modes of operation, believed to be dependent on whether there is bulk memristance, i.e. memristance throughout the whole volume or filamentary memristance, i.e. memristance caused by the connection of conducting filaments. The mem-con theory of memristance is based on the drift of oxygen vacancies rather than that of conducting electrons and has been previously used to describe bulk memristance in several devices. Here, the mem-con theory is extended to model memristance caused by small filaments of low resistance titanium dioxide and it compares favorably to experimental results for filamentary memristance in sol-gel devices.
△ Less
Submitted 30 July, 2012;
originally announced July 2012.
