Spiking representation learning for associative memories

Networks of interconnected neurons communicating through spiking signals offer the bedrock of neural computations. Our brains spiking neural networks have the computational capacity to achieve complex pattern recognition and cognitive functions effortlessly. However, solving real-world problems with artificial spiking neural networks (SNNs) has proved to be difficult for a variety of reasons. Crucially, scaling SNNs to large networks and processing large-scale real-world datasets have been challenging, especially when compared to their non-spiking deep learning counterparts. The critical operation that is needed of SNNs is the ability to learn distributed representations from data and use these representations for perceptual, cognitive and memory operations. In this work, we introduce a novel SNN that performs unsupervised representation learning and associative memory operations leveraging Hebbian synaptic and activity-dependent structural plasticity coupled with neuron-units modelled as Poisson spike generators with sparse firing (~1 Hz mean and ~100 Hz maximum firing rate). Crucially, the architecture of our model derives from the neocortical columnar organization and combines feedforward projections for learning hidden representations and recurrent projections for forming associative memories. We evaluated the model on properties relevant for attractor-based associative memories such as pattern completion, perceptual rivalry, distortion resistance, and prototype extraction.

Benchmarking Hebbian learning rules for associative memory

Associative memory or content addressable memory is an important component function in computer science and information processing and is a key concept in cognitive and computational brain science. Many different neural network architectures and learning rules have been proposed to model associative memory of the brain while investigating key functions like pattern completion and rivalry, noise reduction, and storage capacity. A less investigated but important function is prototype extraction where the training set comprises pattern instances generated by distorting prototype patterns and the task of the trained network is to recall the correct prototype pattern given a new instance. In this paper we characterize these different aspects of associative memory performance and benchmark six different learning rules on storage capacity and prototype extraction. We consider only models with Hebbian plasticity that operate on sparse distributed representations with unit activities in the interval [0,1]. We evaluate both non-modular and modular network architectures and compare performance when trained and tested on different kinds of sparse random binary pattern sets, including correlated ones. We show that covariance learning has a robust but low storage capacity under these conditions and that the Bayesian Confidence Propagation learning rule (BCPNN) is superior with a good margin in all cases except one, reaching a three times higher composite score than the second best learning rule tested.

Spiking neural networks with Hebbian plasticity for unsupervised representation learning

We introduce a novel spiking neural network model for learning distributed internal representations from data in an unsupervised procedure. We achieved this by transforming the non-spiking feedforward Bayesian Confidence Propagation Neural Network (BCPNN) model, employing an online correlation-based Hebbian-Bayesian learning and rewiring mechanism, shown previously to perform representation learning, into a spiking neural network with Poisson statistics and low firing rate comparable to in vivo cortical pyramidal neurons. We evaluated the representations learned by our spiking model using a linear classifier and show performance close to the non-spiking BCPNN, and competitive with other Hebbian-based spiking networks when trained on MNIST and F-MNIST machine learning benchmarks.

Hebbian fast plasticity and working memory

Theories and models of working memory (WM) were at least since the mid-1990s dominated by the persistent activity hypothesis. The past decade has seen rising concerns about the shortcomings of sustained activity as the mechanism for short-term maintenance of WM information in the light of accumulating experimental evidence for so-called activity-silent WM and the fundamental difficulty in explaining robust multi-item WM. In consequence, alternative theories are now explored mostly in the direction of fast synaptic plasticity as the underlying mechanism.The question of non-Hebbian vs Hebbian synaptic plasticity emerges naturally in this context. In this review we focus on fast Hebbian plasticity and trace the origins of WM theories and models building on this form of associative learning.

Brain-like combination of feedforward and recurrent network components achieves prototype extraction and robust pattern recognition

Associative memory has been a prominent candidate for the computation performed by the massively recurrent neocortical networks. Attractor networks implementing associative memory have offered mechanistic explanation for many cognitive phenomena. However, attractor memory models are typically trained using orthogonal or random patterns to avoid interference between memories, which makes them unfeasible for naturally occurring complex correlated stimuli like images. We approach this problem by combining a recurrent attractor network with a feedforward network that learns distributed representations using an unsupervised Hebbian-Bayesian learning rule. The resulting network model incorporates many known biological properties: unsupervised learning, Hebbian plasticity, sparse distributed activations, sparse connectivity, columnar and laminar cortical architecture, etc. We evaluate the synergistic effects of the feedforward and recurrent network components in complex pattern recognition tasks on the MNIST handwritten digits dataset. We demonstrate that the recurrent attractor component implements associative memory when trained on the feedforward-driven internal (hidden) representations. The associative memory is also shown to perform prototype extraction from the training data and make the representations robust to severely distorted input. We argue that several aspects of the proposed integration of feedforward and recurrent computations are particularly attractive from a machine learning perspective.

Semi-supervised learning with Bayesian Confidence Propagation Neural Network

Learning internal representations from data using no or few labels is useful for machine learning research, as it allows using massive amounts of unlabeled data. In this work, we use the Bayesian Confidence Propagation Neural Network (BCPNN) model developed as a biologically plausible model of the cortex. Recent work has demonstrated that these networks can learn useful internal representations from data using local Bayesian-Hebbian learning rules. In this work, we show how such representations can be leveraged in a semi-supervised setting by introducing and comparing different classifiers. We also evaluate and compare such networks with other popular semi-supervised classifiers.

StreamBrain: An HPC Framework for Brain-like Neural Networks on CPUs, GPUs and FPGAs

The modern deep learning method based on backpropagation has surged in popularity and has been used in multiple domains and application areas. At the same time, there are other -- less-known -- machine learning algorithms with a mature and solid theoretical foundation whose performance remains unexplored. One such example is the brain-like Bayesian Confidence Propagation Neural Network (BCPNN). In this paper, we introduce StreamBrain -- a framework that allows neural networks based on BCPNN to be practically deployed in High-Performance Computing systems. StreamBrain is a domain-specific language (DSL), similar in concept to existing machine learning (ML) frameworks, and supports backends for CPUs, GPUs, and even FPGAs. We empirically demonstrate that StreamBrain can train the well-known ML benchmark dataset MNIST within seconds, and we are the first to demonstrate BCPNN on STL-10 size networks. We also show how StreamBrain can be used to train with custom floating-point formats and illustrate the impact of using different bfloat variations on BCPNN using FPGAs.

Brain-like approaches to unsupervised learning of hidden representations -- a comparative study

Unsupervised learning of hidden representations has been one of the most vibrant research directions in machine learning in recent years. In this work we study the brain-like Bayesian Confidence Propagating Neural Network (BCPNN) model, recently extended to extract sparse distributed high-dimensional representations. The saliency and separability of the hidden representations when trained on MNIST dataset is studied using an external classifier, and compared with other unsupervised learning methods that include restricted Boltzmann machines and autoencoders.

Learning representations in Bayesian Confidence Propagation neural networks

Unsupervised learning of hierarchical representations has been one of the most vibrant research directions in deep learning during recent years. In this work we study biologically inspired unsupervised strategies in neural networks based on local Hebbian learning. We propose new mechanisms to extend the Bayesian Confidence Propagating Neural Network (BCPNN) architecture, and demonstrate their capability for unsupervised learning of salient hidden representations when tested on the MNIST dataset.

Characterization and Compensation of Network-Level Anomalies in Mixed-Signal Neuromorphic Modeling Platforms

Advancing the size and complexity of neural network models leads to an ever increasing demand for computational resources for their simulation. Neuromorphic devices offer a number of advantages over conventional computing architectures, such as high emulation speed or low power consumption, but this usually comes at the price of reduced configurability and precision. In this article, we investigate the consequences of several such factors that are common to neuromorphic devices, more specifically limited hardware resources, limited parameter configurability and parameter variations. Our final aim is to provide an array of methods for coping with such inevitable distortion mechanisms. As a platform for testing our proposed strategies, we use an executable system specification (ESS) of the BrainScaleS neuromorphic system, which has been designed as a universal emulation back-end for neuroscientific modeling. We address the most essential limitations of this device in detail and study their effects on three prototypical benchmark network models within a well-defined, systematic workflow. For each network model, we start by defining quantifiable functionality measures by which we then assess the effects of typical hardware-specific distortion mechanisms, both in idealized software simulations and on the ESS. For those effects that cause unacceptable deviations from the original network dynamics, we suggest generic compensation mechanisms and demonstrate their effectiveness. Both the suggested workflow and the investigated compensation mechanisms are largely back-end independent and do not require additional hardware configurability beyond the one required to emulate the benchmark networks in the first place. We hereby provide a generic methodological environment for configurable neuromorphic devices that are targeted at emulating large-scale, functional neural networks.