Reservoir Computing

We have implemented and demonstrated this work on a Pioneer mobile robot. We first detect people and objects in the scene using depth and vision sensors. We create a map of the environment and use off the shelf techniques for localization and path planning using SICK laser. We describe use of Dynamic Bayesian Network to maintain and update belief about each people and object location. We run separate particle filters for each person and object and for robots' own location. We update these particle filters with information from ...

Reservoir Computing is a recent pattern recognition tech- nique that combines temporal processing capabilities with fast learning rates and excellent convergence properties. The system consists of two parts: a recurrently connected network of simple nodes (e.g. neurons) called the reservoir, and a so-called readout function which can be any tra- ditional statistical technique and which computes the actual output. The choice of the node with which we build the reservoir is very broad, and in this case we use stochastich bitstream neurons. Classical sigmoidal neurons perform a weighted sum of their inputs, followed by a non-linearity. This uses a lot of additions and multiplications, which is not hardware efficient at all. Stochastic bitstream neurons circumvent this problem by communi- cating through stochastic bitstreams instead of analog values, which trans- forms additions and multiplications to simple bitwise operations, thus al- lowing an efficient hardware implementation. For t...

Abstract. This paper investigates the potential of recurrent spiking neurons for classification problems. It presents a hybrid approach based on the paradigm of Reservoir Computing. The practical applications based on recurrent spiking neurons are limited due to their ...

Common forecasting methods fail to accurately model the nonlinear and time-varying fluctuations of product demand. Reservoir computing (RC) utilizes a dynamical system to project time-series data to a higher-dimensional state representation extracting mathematical relations within complex demand functions. We demonstrate forecasting accuracy of RC on a multivariate product demand dataset.

Se propone un modelo de red discreta, capaz de ser implementada como una red booleana clásica, proclive a ser utilizada como medio excitable para un sistema de computación de reservorio.

La computación de reservorio es un marco de trabajo enfocado en el procesamiento de información con auxilio de un sistema dinámico, llamado reservorio o piscina, que gracias a su comportamiento es capaz de mapear las entradas que recibe a un espacio de estados que facilita su clasificación, haciendo las veces de una función kernel temporal. Se encuentra fuertemente relacionada con las redes neuronales recurrentes, pero no está limitada a las mismas.

Si bien se ha investigado de forma incipiente la viabilidad de Redes Booleanas Aleatorias como piscina, poseen profundas desventajas asociadas a las características propias de su espacio de estados y las transiciones entre los mismos, ya que favorecen un mapeado diferente para entradas idénticas, tan sólo por el momento en que son inyectadas al sistema.

Esta tesis presenta un formalismo para el concepto de «perturbación» aplicado a redes booleanas, y propone un modelo de red booleana con base en él, e inspirado en las Redes Neuronales de Picos (spiking neural networks), que favorece mapeados consistentes para entradas similares, dada su propensión a poseer una cuenca de atracción gigante con un atractor puntual.

Se exploran las capacidades computacionales del modelo propuesto y se comparan con las de Redes Booleanas Aleatorias típicas. Ambos modelos se someten a tareas de clasificación y cálculo, a fin de verificar si las predicciones teóricas sobre su desempeño son correctas.

Echo state networks (ESNs), belonging to the wider family of reservoir computing methods, are a powerful tool for the analysis of dynamic data. In an ESN, the input signal is fed to a fixed (possibly large) pool of interconnected neurons, whose state is then read by an adaptable layer to provide the output. This last layer is generally trained via a regularized linear least-squares procedure. In this paper, we consider the more complex problem of training an ESN for classification problems in a semi-supervised setting, wherein only a part of the input sequences are effectively labeled with the desired response. To solve the problem , we combine the standard ESN with a semi-supervised support vector machine (S 3 VM) for training its adaptable connections. Additionally, we propose a novel algorithm for solving the resulting non-convex optimization problem, hinging on a series of successive approximations of the original problem. The resulting procedure is highly cus-tomizable and also admits a principled way of parallelizing training over multiple processors/computers. An extensive set of experimental evaluations on audio classification tasks supports the presented semi-supervised ESN as a practical tool for dynamic problems requiring the analysis of partially labeled data.

The combination of machine learning and quantum computing has emerged as a promising approach for addressing previously untenable problems. Reservoir computing is an efficient learning paradigm that utilizes nonlinear dynamical systems for temporal information processing, i.e., processing of input sequences to produce output sequences. Here we propose quantum reservoir computing that harnesses complex dissipative quantum dynamics. Our class of quantum reservoirs is universal, in that any nonlinear fading memory map can be approximated arbitrarily closely and uniformly over all inputs by a quantum reservoir from this class. We describe a subclass of the universal class that is readily implementable using quantum gates native to current noisy gate-model quantum computers. Proof-of-principle experiments on remotely accessed cloud-based superconducting quantum computers demonstrate that small and noisy quantum reservoirs can tackle high-order nonlinear temporal tasks. Our theoretical and experimental results pave the path for attractive temporal processing applications of near-term gate-model quantum computers of increasing fidelity but without quantum error correction, signifying the potential of these devices for wider applications including neural modeling, speech recognition and natural language processing, going beyond static classification and regression tasks.

We present a novel gesture recognition system for the application of continuous gestures in mobile devices. We explain how meaningful gesture data can be extracted from the inertial measurement unit of a mobile phone and introduce a segmentation scheme to distinguish between different gesture classes. The continuous sequences are fed into an Echo State Network, which learns sequential data fast and with good performance. We evaluated our system on crucial network parameters and on our established metric to compute the number of successfully recognized gestures and the number of misclassifications. On a total of ten gesture classes, our framework achieved an average accuracy of 85%.

In this paper, we elaborate over the well-known interpretability issue in echo state networks. The idea is to investigate the dynamics of reservoir neurons with time-series analysis techniques taken from research on complex systems. Notably, we analyze time-series of neuron activations with Recurrence Plots (RPs) and Recurrence Quantification Analysis (RQA), which permit to visualize and characterize high-dimensional dynamical systems. We show that this approach is useful in a number of ways. First, the two-dimensional representation offered by RPs provides a way for visualizing the high-dimensional dynamics of a reservoir. Our results suggest that, if the network is stable, reservoir and input denote similar line patterns in the respective RPs. Conversely, the more unstable the ESN, the more the RP of the reservoir presents instability patterns. As a second result, we show that the Lmax measure is highly correlated with the well-established maximal local Lyapunov exponent. This suggests that complexity measures based on RP diagonal lines distribution provide a valuable tool to quantify the degree of network stability. Finally, our analysis shows that all RQA measures fluctuate on the proximity of the so-called edge of stability, where an ESN typically achieves maximum computational capability. We verify that the determination of the edge of stability provided by such RQA measures is more accurate than the one offered by the maximal local Lyapunov exponent. Therefore, we claim that RPs and RQA-based analyses can be used as valuable tools to design an effective network given a specific problem.

Ambient Assisted Living facilities provide assistance and care for the elderly, where it is useful to infer their daily activity for ensuring their safety and successful ageing. In this work, we present an Activity Recognition system that classifies a set of common daily activities exploiting both the data sampled by accelerometer sensors carried out by the user and the reciprocal Received Signal Strength (RSS) values coming from worn wireless sensor devices and from sensors deployed in the environment. To this end, we model the accelerometer and the RSS stream, obtained from a Wireless Sensor Network (WSN), using Recurrent Neural Networks implemented as efficient Echo State Networks (ESNs), within the Reservoir Computing paradigm. Our results show that, with an appropriate configuration of the ESN, the system reaches a good accuracy with a low deployment cost.

We used an Echo State Network (ESN) to investigate cognitive control tasks that involve temporal and policy abstraction, namely an n-back task and a Wisconsin Card Sorting task. In the task conditions where those abstractions were most relevant, separating the abstraction levels increased ESN performance, while the performance was reduced by such a separation in the conditions where the abstractions were less important. This result gives a possible explanation for the topological mapping of abstraction levels over the prefrontal cortex, because of wiring economy in the brain.

Several psycholinguistic models represent words as vectors in a high-dimensional state space, such that distances between vectors encode the strengths of paradigmatic relations between the represented words. This chapter argues that such an organization develops because it facilitates fast sentence processing. A model is presented in which sentences, in the form of word-vector sequences, serve as input to a recurrent neural network that provides random dynamics. The word vectors are adjusted by a process of self-organization, aimed at reducing fluctuations in the dynamics. As it turns out, the resulting word vectors are organized paradigmatically.

Keywords: Word representation; Sentence processing; Self-organization; Recurrent neural network; Reservoir computing.

A spiking neural network model is described for learning to discriminate among spatial patterns in an unsupervised manner. The network anatomy consists of source neurons that are activated by external inputs, a reservoir that resembles a generic cortical layer with an excitatory-inhibitory (EI) network and a sink layer of neurons for readout. Synaptic plasticity in the form of STDP is imposed on all the excitatory and inhibitory synapses at all times. While long-term excitatory STDP enables sparse and efficient learning of the salient features in inputs, inhibitory STDP enables this learning to be stable by establishing a balance between excitatory and inhibitory currents at each neuron in the network. The synaptic weights between source and reservoir neurons form a basis set for the input patterns. The neural trajectories generated in the reservoir due to input stimulation and lateral connections between reservoir neurons can be readout by the sink layer neurons. This activity is u...

Reservoir computing is a recently introduced, highly efficient bio-inspired approach for processing time dependent data. The basic scheme of reservoir computing consists of a non linear recurrent dynamical system coupled to a single input layer and a single output layer. Within these constraints many implementations are possible. Here we report an optoelectronic implementation of reservoir computing based on a recently proposed architecture consisting of a single non linear node and a delay line. Our implementation is sufficiently fast for real time information processing. We illustrate its performance on tasks of practical importance such as nonlinear channel equalization and speech recognition, and obtain results comparable to state of the art digital implementations.

Nous présentons ici l’implémentation optique d’un réseau de neurones artificiel de type reservoir computer. Il est réalisé sur une structure d’oscillateur optoélectronique utilisant une boucle à retard en fibre optique. Notre système est étudié en régime libre, puis soumis à des tests standardisés, y compris sa capacité de mémoire, pour mesurer ses performances en tant que reservoir computer.

History matching is a critical step within the reservoir management process to synchronize the simulation model with the production data. The history-matched model can be used for planning optimum field development and performing optimization and uncertainty quantifications. We present a novel history matching workflow based on a Bayesian framework that accommodates subsurface uncertainties. Our workflow involves three different model resolutions within the Bayesian framework: 1) a coarse low-fidelity model to update the prior range, 2) a fine low-fidelity model to represent the high-fidelity model, and 3) a high-fidelity model to re-construct the real response. The low-fidelity model is constructed by a multivariate polynomial function, while the high-fidelity model is based on the reservoir simulation model. We firstly develop a coarse low-fidelity model using a two-level Design of Experiment (DoE), which aims to provide a better prior. We secondly use Latin Hypercube Sampling (LH...

Advances in materials science have led to physical instantiations of self-assembled networks of memristive devices and demonstrations of their computational capability through reservoir computing. Reservoir computing is an approach that takes advantage of collective system dynamics for real-time computing. A dynamical system, called a reservoir, is excited with a time-varying signal and observations of its states are used to reconstruct a desired output signal. However, such a monolithic assembly limits the computational power due to signal interdependency and the resulting correlated readouts. Here, we introduce an approach that hierarchically composes a set of interconnected memristive networks into a larger reservoir. We use signal amplification and restoration to reduce reservoir state correlation, which improves the feature extraction from the input signals. Using the same number of output signals, such a hierarchical composition of heterogeneous small networks outperforms mono...

The mushroom body is the key network for the representation of learned olfactory stimuli in Drosophila and insects. The sparse activity of Kenyon cells, the principal neurons in the mushroom body, plays a key role in the learned classification of different odours. In the specific case of the fruit fly, the sparseness of the network is enforced by an inhibitory feedback neuron called APL, and by an intrinsic high firing threshold of the Kenyon cells. In this work we took inspiration from the fruit fly brain to formulate a novel machine learning algorithm that is able to optimize the sparsity level of a reservoir by changing the firing thresholds of the nodes. The sparsity is only applied on the readout layer so as not to change the timescales of the reservoir and to allow the derivation of a one-layer update rule for the firing thresholds. The proposed algorithm is a combination of learning a neuron-specific sparsity threshold via gradient descent and a global sparsity threshold via ...

It is a widely accepted fact that the computational
capability of recurrent neural networks is maximized on the so called “edge of criticality”. Once the network operates in this
configuration, it performs efficiently on a specific application
both in terms of (i) low prediction error and (ii) high short term memory capacity. Since the behavior of recurrent networks
is strongly influenced by the particular input signal driving
the dynamics, a universal, application-independent method for
determining the edge of criticality is still missing. In this paper,
we aim at addressing this issue by proposing a theoretically
motivated, unsupervised method based on Fisher information for
determining the edge of criticality in recurrent neural networks.
It is proven that Fisher information is maximized for (finite size) systems operating in such critical regions. However, Fisher
information is notoriously difficult to compute and either requires
the probability density function or the conditional dependence of
the system states with respect to the model parameters. The paper
takes advantage of a recently-developed non-parametric estimator of the Fisher information matrix and provides a method to
determine the critical region of echo state networks, a particular
class of recurrent networks. The considered control parameters,
which indirectly affect the echo state network performance, are
explored to identify those configurations lying on the edge of
criticality and, as such, maximizing Fisher information and computational performance. Experimental results on benchmarks
and real-world data demonstrate the effectiveness of the proposed
method.

Echo state networks are a relatively new type of recurrent neural networks that have shown great potentials for solving nonlinear, temporal problems. The basic idea is to transform the low dimensional temporal input into a higher dimensional state,
and then train the output connection weights to make the system output the target information. Because only the output weights are altered, training is typically quick and computationally efficient compared to training of other recurrent neural networks. This
paper investigates using an echo state network to learn the inverse kinematics model of a robot simulator with feedback-errorlearning. In this scheme teacher forcing is not perfect, and joint constraints on the simulator makes the feedback error inaccurate.
A novel trainingmethod which is less influenced by the noise in the training data is proposed and compared to the traditional ESN training method.

Reservoir Computing is a new paradigm for using Recurrent Neural Networks which shows promising results. However, as the recurrent part is created randomly, it typically needs to be large enough to be able to capture the dynamic features of the data considered. Moreover, this random creation is still lacking a strong methodology. We propose to study how pruning some connections from the reservoir to the readout can help on the one hand to increase the generalisation ability, in much the same way as regularisation techniques do, and on the other hand to improve the implementability of reservoirs in hardware. Furthermore we study the actual sub-reservoir which is kept after pruning which leads to important insights in what we have to expect from a good reservoir.

We implement a Liquid State Machine composed from a pool of chaotic spiking neurons. Furthermore, a synaptic plasticity mechanism operates on the connection weights between the neurons inside the pool. A special feature of the system’s classification capability is that it can learn the class of a set of time varying inputs when trained from positive examples only, thus, it is a one class classifier. To demonstrate the applicability of this novel neurocomputing architecture, we apply it for Online Signature Verification.

Convergent, contractive or incremental stability properties of nonlinear systems have attracted interest for control tasks such as observer design, output regulation and synchro- nization. The convergence property plays a central role in the neuromorphic (brain-inspired) paradigm of reservoir comput- ing, which seeks to harness the information processing capability of nonlinear systems. This paper presents a small-gain theorem for discrete-time output-feedback interconnected systems to be uniformly input-to-output convergent (UIOC) with outputs con- verging to a bounded reference output uniquely determined by the input. A small-gain theorem for interconnected time-varying discrete-time uniform input-to-output stable systems that could be of separate interest is also presented as an intermediate result. Applications of the UIOC small-gain theorem are illustrated in the design of observer-based controllers and interconnected nonlinear classical and quantum dynamical systems (as reservoir computers) for black-box system identification.