Generation Of Artificial Neural Networks Research Articles

Tuning Convolutional Spiking Neural Network With Biologically Plausible Reward Propagation.

Spiking neural networks (SNNs) contain more biologically realistic structures and biologically inspired learning principles than those in standard artificial neural networks (ANNs). SNNs are considered the third generation of ANNs, powerful on the robust computation with a low computational cost. The neurons in SNNs are nondifferential, containing decayed historical states and generating event-based spikes after their states reaching the firing threshold. These dynamic characteristics of SNNs make it difficult to be directly trained with the standard backpropagation (BP), which is also considered not biologically plausible. In this article, a biologically plausible reward propagation (BRP) algorithm is proposed and applied to the SNN architecture with both spiking-convolution (with both 1-D and 2-D convolutional kernels) and full-connection layers. Unlike the standard BP that propagates error signals from postsynaptic to presynaptic neurons layer by layer, the BRP propagates target labels instead of errors directly from the output layer to all prehidden layers. This effort is more consistent with the top-down reward-guiding learning in cortical columns of the neocortex. Synaptic modifications with only local gradient differences are induced with pseudo-BP that might also be replaced with the spike-timing-dependent plasticity (STDP). The performance of the proposed BRP-SNN is further verified on the spatial (including MNIST and Cifar-10) and temporal (including TIDigits and DvsGesture) tasks, where the SNN using BRP has reached a similar accuracy compared to other state-of-the-art (SOTA) BP-based SNNs and saved 50% more computational cost than ANNs. We think that the introduction of biologically plausible learning rules to the training procedure of biologically realistic SNNs will give us more hints and inspiration toward a better understanding of the biological system's intelligent nature.

Continuous learning of spiking networks trained with local rules

Artificial neural networks (ANNs) experience catastrophic forgetting (CF) during sequential learning. In contrast, the brain can learn continuously without any signs of catastrophic forgetting. Spiking neural networks (SNNs) are the next generation of ANNs with many features borrowed from biological neural networks. Thus, SNNs potentially promise better resilience to CF. In this paper, we study the susceptibility of SNNs to CF and test several biologically inspired methods for mitigating catastrophic forgetting. SNNs are trained with biologically plausible local training rules based on spike-timing-dependent plasticity (STDP). Local training prohibits the direct use of CF prevention methods based on gradients of a global loss function. We developed and tested the method to determine the importance of synapses (weights) based on stochastic Langevin dynamics without the need for the gradients. Several other methods of catastrophic forgetting prevention adapted from analog neural networks were tested as well. The experiments were performed on freely available datasets in the SpykeTorch environment.

A Survey of Supervised Learning Models for Spiking Neural Network

There has been a significant attempt to derive supervised learning models for training Spiking Neural Networks (SNN), which is the third and most recent generation of Artificial Neural Network (ANN). Supervised SNN learning models are considered more biologically plausible and thus exploits better the computational efficiency of biological neurons and also, are less computationally expensive than second generation ANN. SNN models have also produced competitive performance in most tasks when compared to second generation ANNs. These advantages, coupled with the difficulty in adopting the well established learning models for second generation networks to train SNN due to the difference in information coding led to the recent introduction of supervised learning models for training SNN.
 However, lack of comprehensive source of literature detailing strides made in this area, and the challenges and prospects of SNN serves as a hindrance to further exploration and application of SNN models. A comprehensive review of supervised learning methods in SNN is presented in this paper in which some widely used SNN neural models, learning models and their basic concepts, areas of applications, limitations, prospects and future research directions are discussed. The main contribution of this paper is that it presents and discusses trends in supervised learning in SNNwith the aim of providing a reference point for those desiring further knowledge and application of SNN methods.

Evolutionary Spiking Neural Networks for Solving Supervised Classification Problems.

This paper presents a grammatical evolution (GE)-based methodology to automatically design third generation artificial neural networks (ANNs), also known as spiking neural networks (SNNs), for solving supervised classification problems. The proposal performs the SNN design by exploring the search space of three-layered feedforward topologies with configured synaptic connections (weights and delays) so that no explicit training is carried out. Besides, the designed SNNs have partial connections between input and hidden layers which may contribute to avoid redundancies and reduce the dimensionality of input feature vectors. The proposal was tested on several well-known benchmark datasets from the UCI repository and statistically compared against a similar design methodology for second generation ANNs and an adapted version of that methodology for SNNs; also, the results of the two methodologies and the proposed one were improved by changing the fitness function in the design process. The proposed methodology shows competitive and consistent results, and the statistical tests support the conclusion that the designs produced by the proposal perform better than those produced by other methodologies.

On the estimation of stellar parameters with uncertainty prediction from Generative Artificial Neural Networks: application toGaiaRVS simulated spectra

Aims. We present an innovative artificial neural network (ANN) architecture, called Generative ANN (GANN), that computes the forward model, that is it learns the function that relates the unknown outputs (stellar atmospheric parameters, in this case) to the given inputs (spectra). Such a model can be integrated in a Bayesian framework to estimate the posterior distribution of the outputs. Methods. The architecture of the GANN follows the same scheme as a normal ANN, but with the inputs and outputs inverted. We train the network with the set of atmospheric parameters (Teff, logg, [Fe/H] and [alpha/Fe]), obtaining the stellar spectra for such inputs. The residuals between the spectra in the grid and the estimated spectra are minimized using a validation dataset to keep solutions as general as possible. Results. The performance of both conventional ANNs and GANNs to estimate the stellar parameters as a function of the star brightness is presented and compared for different Galactic populations. GANNs provide significantly improved parameterizations for early and intermediate spectral types with rich and intermediate metallicities. The behaviour of both algorithms is very similar for our sample of late-type stars, obtaining residuals in the derivation of [Fe/H] and [alpha/Fe] below 0.1dex for stars with Gaia magnitude Grvs<12, which accounts for a number in the order of four million stars to be observed by the Radial Velocity Spectrograph of the Gaia satellite. Conclusions. Uncertainty estimation of computed astrophysical parameters is crucial for the validation of the parameterization itself and for the subsequent exploitation by the astronomical community. GANNs produce not only the parameters for a given spectrum, but a goodness-of-fit between the observed spectrum and the predicted one for a given set of parameters. Moreover, they allow us to obtain the full posterior distribution...

Introducing Synaptic Delays in the NEAT Algorithm to Improve Modelling in Cognitive Robotics

This paper describes and tests an approach to improve the temporal processing capabilities of the neuroevolution of augmenting topologies (NEAT) algorithm. This algorithm is quite popular within the robotics community for the production of trained neural networks without having to determine a priori their size and topology. The main drawback of the traditional NEAT algorithm is that, even though it can implement recurrent synaptic connections, which allow it to perform some time related processing tasks, its capabilities are rather limited, especially when dealing with precise time dependent phenomena. NEAT's ability to capture the underlying dynamics that correspond to complex time series still has a lot of room for improvement. To address this issue, the paper describes a new implementation of the NEAT algorithm that is able to generate artificial neural networks (ANNs) with trainable time delayed synapses in addition to its previous capacities. We show that this approach, called $$\uptau $$?-NEAT improves the behavior of the neural networks obtained when dealing with complex time related processes. Several examples are presented, both dealing with the generation of ANNs that are able to produce complex theoretical signals such as chaotic signals or real data series, as in the case of the monthly number of international airline passengers or monthly $$\hbox {CO}_{2}$$CO2 concentrations. In these examples, $$\uptau $$?-NEAT clearly improves over the traditional NEAT algorithm in these tasks. A final example of the integration of this approach within a robot cognitive mechanism is also presented, showing the clear improvements it could provide in the modeling required for many cognitive processes.

Spiking Self-organizing Maps for Classification Problem

Abstract In Self-Organizing Maps (SOM) learning, preserving the map topology to simulate the real input features is one of the most important processes. This is done by training the weight values within the Best Matching Unit (BMU) neighborhood. Improper input feeding will cause failure in identifying the potential BMU which will lead to poor map topology. Many studies have been done to optimize the structure of SOM's topology using Artificial Neural Networks (ANN).Spiking Neural Network (SNN) is the third generation of ANN, where information are transferred from one neuron to other using spikes, and processed to trigger response as an output. Current researches have proven that SNN would be an alternative solution for enhancing ANN learning due to its superiority in capturing the internal relationship of neurons. This paper proposes embedded spiking neurons for Kohonen's Self-organizing Maps (SOM) learning to improve its learning process. The proposed Spiking SOM is divided into four main phases. Phase 1 involves the development of the training sample for SOM learning through neural coding schemes. In Phase 2, the spike values are fed into the training process and potential weights are generated. Phase 3 identifies and labels the outputs from the Spiking SOM classification based on the features and characteristics. Finally, in Phase 4, proposed Spiking SOM model is validated using classification accuracy, error quantization and statistical tests using Pearson correlation. Early experiment is conducted using the 1D coding schemes for transforming dataset into spike times with hexagonal lattice structure of SOM network. Result on cancer dataset shows that the tested model has produced feasible classification accuracy with low quantization error. It shows that the 1D coding is capable in preserving the features in the input neurons.

‘Query‐by Committee’— An Efficient Method to Select Information‐Rich Data for the Development of Peptide—HLA‐Binding Predictors

Rationale: We have previously demonstrated that bioinformatics tools such as artificial neural networks (ANNs) are capable of performing pathogen‐, genome‐ and HLA‐wide predictions of peptide–HLA interactions. These tools may therefore enable a fast and rational approach to epitope identification and thereby assist in the development of vaccines and immunotherapy. A crucial step in the generation of such bioinformatics tools is the selection of data representing the event in question (in casu peptide–HLA interaction). This is particularly important when it is difficult and expensive to obtain data. Herein, we demonstrate the importance in selecting information‐rich data and we develop a computational method, query‐by‐committee, which can perform a global identification of such information‐rich data in an unbiased and automated manner. Furthermore, we demonstrate how this method can be applied to an efficient iterative development strategy for these bioinformatics tools. Methods: A large panel of binding affinities of peptides binding to HLA A*0204 was measured by a radioimmunoassay (RIA). This data was used to develop multiple first generation ANNs, which formed a virtual committee. This committee was used to screen (or ‘queried’) for peptides, where the ANNs agreed (‘low‐QBC’), or disagreed (‘high‐QBC’), on their HLA‐binding potential. Seventeen low‐QBC peptides and 17 high‐QBC peptides were synthesized and tested. The high‐ or low‐QBC data were added to the original data, and new high‐ or low‐QBC second generation ANNs were developed, respectively. This procedure was repeated 40 times. Results: The high‐QBC‐enriched ANN performed significantly better than the low‐QBC‐enriched ANN in 37 of the 40 tests. Conclusion: These results demonstrate that high‐QBC‐enriched networks perform better than low‐QBC‐enriched networks in selecting informative data for developing peptide–MHC‐binding predictors. This improvement in selecting data is not due to differences in network training performance but due to the difference in information content in the high‐QBC experiment and in the low‐QBC experiment. Finally, it should be noted that this strategy could be used in many contexts where generation of data is difficult and costly.