Resistive memory crossbars can dramatically reduce the energy required to perform computations in neural algorithms by three orders of magnitude when compared to an optimized digital ASIC [1]. For data intensive applications, the computational energy is dominated by moving data between the processor, SRAM, and DRAM. Analog crossbars overcome this by allowing data to be processed directly at each memory element. Analog crossbars accelerate three key operations that are the bulk of the computation in a neural network as illustrated in Fig 1: vector matrix multiplies (VMM), matrix vector multiplies (MVM), and outer product rank 1 updates (OPU)[2]. For an NxN crossbar the energy for each operation scales as the number of memory elements O(N2) [2]. This is because the crossbar performs its entire computation in one step, charging all the capacitances only once. Thus the CV2 energy of the array scales as array size. This fundamentally better than trying to read or write a digital memory. Each row of any NxN digital memory must be accessed one at a time, resulting in N columns of length O(N) being charged N times, requiring O(N3) energy to read a digital memory. Thus an analog crossbar has a fundamental O(N) energy scaling advantage over a digital system. Furthermore, if the read operation is done at low voltage and is therefore noise limited, the read energy can even be independent of the crossbar size, O(1) [2].
The dentate gyrus forms a critical link between the entorhinal cortex and CA3 by providing a sparse version of the signal. Concurrent with this increase in sparsity, a widely accepted theory suggests the dentate gyrus performs pattern separation-similar inputs yield decorrelated outputs. Although an active region of study and theory, few logically rigorous arguments detail the dentate gyrus's (DG) coding.We suggest a theoretically tractable, combinatorial model for this action. The model provides formal methods for a highly redundant, arbitrarily sparse, and decorrelated output signal. To explore the value of this model framework, we assess how suitable it is for two notable aspects of DG coding: how it can handle the highly structured grid cell representation in the input entorhinal cortex region and the presence of adult neurogenesis, which has been proposed to produce a heterogeneous code in the DG.We find tailoring themodel to grid cell input yields expansion parameters consistent with the literature. In addition, the heterogeneous coding reflects activity gradation observed experimentally. Finally,we connect this approach with more conventional binary threshold neural circuit models via a formal embedding.
Biological neural networks continue to inspire new developments in algorithms and microelectronic hardware to solve challenging data processing and classification problems. Here, we survey the history of neural-inspired and neuromorphic computing in order to examine the complex and intertwined trajectories of the mathematical theory and hardware developed in this field. Early research focused on adapting existing hardware to emulate the pattern recognition capabilities of living organisms. Contributions from psychologists, mathematicians, engineers, neuroscientists, and other professions were crucial to maturing the field from narrowly-tailored demonstrations to more generalizable systems capable of addressing difficult problem classes such as object detection and speech recognition. Algorithms that leverage fundamental principles found in neuroscience such as hierarchical structure, temporal integration, and robustness to error have been developed, and some of these approaches are achieving world-leading performance on particular data classification tasks. In addition, novel microelectronic hardware is being developed to perform logic and to serve as memory in neuromorphic computing systems with optimized system integration and improved energy efficiency. Key to such advancements was the incorporation of new discoveries in neuroscience research, the transition away from strict structural replication and towards the functional replication of neural systems, and the use of mathematical theory frameworks to guide algorithm and hardware developments.
For decades, neural networks have shown promise for next-generation computing, and recent breakthroughs in machine learning techniques, such as deep neural networks, have provided state-of-the-art solutions for inference problems. However, these networks require thousands of training processes and are poorly suited for the precise computations required in scientific or similar arenas. The emergence of dedicated spiking neuromorphic hardware creates a powerful computational paradigm which can be leveraged towards these exact scientific or otherwise objective computing tasks. We forego any learning process and instead construct the network graph by hand. In turn, the networks produce guaranteed success often with easily computable complexity. We demonstrate a number of algorithms exemplifying concepts central to spiking networks including spike timing and synaptic delay. We also discuss the application of cross-correlation particle image velocimetry and provide two spiking algorithms; one uses time-division multiplexing, and the other runs in constant time.
The effort to develop larger-scale computing systems introduces a set of related challenges: Large machines are more difficult to synchronize. The sheer quantity of hardware introduces more opportunities for errors. New approaches to hardware, such as low-energy or neuromorphic devices are not directly programmable by traditional methods. These three challenges may be addressed, at least for a subset of interesting problems, by a dynamical systems approach. The initial state of system represents the problem, and the final state of the system represents the solution. By carefully controlling the attractive basin of the system, we can move it between these two points while tolerating errors, which appear as perturbations. Here we describe both conventional and neural computers as dynamical systems, and show how to construct algorithms with resilience to noise, using traditional numerical problems as a special case. This suggests a reduction from numerical problems to spiking neural hardware such as IBM's TrueNorth.
Through various means of structural and synaptic plasticity enabling online learning, neural networks are constantly reconfiguring their computational functionality. Neural information content is embodied within the configurations, representations, and computations of neural networks. To explore neural information content, we have developed metrics and computational paradigms to quantify neural information content. We have observed that conventional compression methods may help overcome some of the limiting factors of standard information theoretic techniques employed in neuroscience, and allows us to approximate information in neural data. To do so we have used compressibility as a measure of complexity in order to estimate entropy to quantitatively assess information content of neural ensembles. Using Lempel-Ziv compression we are able to assess the rate of generation of new patterns across a neural ensemble's firing activity over time to approximate the information content encoded by a neural circuit. As a specific case study, we have been investigating the effect of neural mixed coding schemes due to hippocampal adult neurogenesis.
Resistive memories enable dramatic energy reductions for neural algorithms. We propose a general purpose neural architecture that can accelerate many different algorithms and determine the device properties that will be needed to run backpropagation on the neural architecture. To maintain high accuracy, the read noise standard deviation should be less than 5% of the weight range. The write noise standard deviation should be less than 0.4% of the weight range and up to 300% of a characteristic update (for the datasets tested). Asymmetric nonlinearities in the change in conductance vs pulse cause weight decay and significantly reduce the accuracy, while moderate symmetric nonlinearities do not have an effect. In order to allow for parallel reads and writes the write current should be less than 100 nA as well.