Synaptic Learning Rule Research Articles

Robust Sodium Carboxymethyl Cellulose-Based Neuromorphic Device with High Biocompatibility Engineered through Molecular Polarization for the Emulation of Learning Behaviors in the Human Brain.

Sodium carboxymethyl cellulose (CMC-Na), derived from natural cellulose and frequently employed as a biocompatible coating, thus renders it an ideal component for the construction of highly biocompatible neuromorphic devices aimed at biomachine interfaces. Here, an array of Mo/CMC-Na/ITO neuromorphic devices is fabricated, with CMC-Na serving as the functional layer. The devices exhibit capabilities to emulate various synaptic learning rules and demonstrate high endurance performance among biomaterial-based electronics, achieving stability over 2 × 104 pulses. Then, simulations of human brain-inspired learning and forgetting paradigms are conducted, highlighting the versatility of the device array in mimicking learning processes. Applications in pattern recognition leverage "learning-forgetting" paradigms, showcasing the potential of the device in cognitive tasks. Electrical measurements elucidate the mechanism of molecular polarization rotation, which offers insights into the modulation of synaptic weights within biocompatible biomaterial-based devices. Furthermore, the biocompatible properties of the devices are evaluated using human embryonic kidney 293 cells, confirming their excellent biocompatibility. The biodegradability of the devices is assessed by using physical transient tests to evaluate their sustainability in biomedical applications. Such advances represent pivotal improvements in implantable bioinspired electronics and show potential in biomachine interface and cognitive computing applications.

Synapses learn to utilize stochastic pre-synaptic release for the prediction of postsynaptic dynamics.

Synapses in the brain are highly noisy, which leads to a large trial-by-trial variability. Given how costly synapses are in terms of energy consumption these high levels of noise are surprising. Here we propose that synapses use noise to represent uncertainties about the somatic activity of the postsynaptic neuron. To show this, we developed a mathematical framework, in which the synapse as a whole interacts with the soma of the postsynaptic neuron in a similar way to an agent that is situated and behaves in an uncertain, dynamic environment. This framework suggests that synapses use an implicit internal model of the somatic membrane dynamics that is being updated by a synaptic learning rule, which resembles experimentally well-established LTP/LTD mechanisms. In addition, this approach entails that a synapse utilizes its inherently noisy synaptic release to also encode its uncertainty about the state of the somatic potential. Although each synapse strives for predicting the somatic dynamics of its postsynaptic neuron, we show that the emergent dynamics of many synapses in a neuronal network resolve different learning problems such as pattern classification or closed-loop control in a dynamic environment. Hereby, synapses coordinate themselves to represent and utilize uncertainties on the network level in behaviorally ambiguous situations.

Highly Promising 2D/1D BP-C/CNT Bionic Opto-Olfactory Co-Sensory Artificial Synapses for Multisensory Integration.

The development of high-performance artificial synaptic neuromorphic devices poses a significant challenge in the creation of biomimetic sensing neural systems that seamlessly integrate both sensory and computational functionalities. In pursuit of this objective, promising bionic opto-olfactory co-sensory artificial synapse devices are constructed utilizing the BP-C/CNT (2D/1D) hybrid filter membrane as the resistive layer. Experimental results demonstrated that the devices seamlessly integrated the light modulation, gas detection, and biological synaptic functions into a single device while addressing the challenge with separating artificial synaptic devices from sensors. These devices offered the following advantages: 1) Simulating visual synapses, they can effectively replicate fundamental synaptic functions under both electrical and optical stimulation. 2) By emulating olfactory synapse responses to specific gases, they can achieve ultra-low detection limits and rapid identification of ethanol and acetone gases. 3) They enable photo-olfactory co-sensing simulations that mimic synaptic function under light-modulated pulse conditions in distinct gas environments, facilitating the study of synaptic learning rules and Pavlovian responses. This work provides a pioneering approach for exploring highly stable 2D BP-based optoelectronics and advancing the development of biomimetic neural systems.

High-performance van der Waals antiferroelectric CuCrP2S6-based memristors

Layered thio- and seleno-phosphate ferroelectrics, such as CuInP2S6, are promising building blocks for next-generation nonvolatile memory devices. However, because of the low Curie point, the CuInP2S6-based memory devices suffer from poor thermal stability (<42 °C). Here, exploiting the electric field-driven phase transition in the rarely studied antiferroelectric CuCrP2S6 crystals, we develop a nonvolatile memristor showing a sizable resistive-switching ratio of ~ 1000, high switching endurance up to 20,000 cycles, low cycle-to-cycle variation, and robust thermal stability up to 120 °C. The resistive switching is attributed to the ferroelectric polarization-modulated thermal emission accompanied by the Fowler–Nordheim tunneling across the interfaces. First-principles calculations reveal that the good device performances are associated with the exceptionally strong ferroelectric polarization in CuCrP2S6 crystal. Furthermore, the typical biological synaptic learning rules, such as long-term potentiation/depression and spike amplitude/spike time-dependent plasticity, are also demonstrated. The results highlight the great application potential of van der Waals antiferroelectrics in high-performance synaptic devices for neuromorphic computing.

Biologically plausible local synaptic learning rules robustly implement deep supervised learning.

In deep neural networks, representational learning in the middle layer is essential for achieving efficient learning. However, the currently prevailing backpropagation learning rules (BP) are not necessarily biologically plausible and cannot be implemented in the brain in their current form. Therefore, to elucidate the learning rules used by the brain, it is critical to establish biologically plausible learning rules for practical memory tasks. For example, learning rules that result in a learning performance worse than that of animals observed in experimental studies may not be computations used in real brains and should be ruled out. Using numerical simulations, we developed biologically plausible learning rules to solve a task that replicates a laboratory experiment where mice learned to predict the correct reward amount. Although the extreme learning machine (ELM) and weight perturbation (WP) learning rules performed worse than the mice, the feedback alignment (FA) rule achieved a performance equal to that of BP. To obtain a more biologically plausible model, we developed a variant of FA, FA_Ex-100%, which implements direct dopamine inputs that provide error signals locally in the layer of focus, as found in the mouse entorhinal cortex. The performance of FA_Ex-100% was comparable to that of conventional BP. Finally, we tested whether FA_Ex-100% was robust against rule perturbations and biologically inevitable noise. FA_Ex-100% worked even when subjected to perturbations, presumably because it could calibrate the correct prediction error (e.g., dopaminergic signals) in the next step as a teaching signal if the perturbation created a deviation. These results suggest that simplified and biologically plausible learning rules, such as FA_Ex-100%, can robustly facilitate deep supervised learning when the error signal, possibly conveyed by dopaminergic neurons, is accurate.

Synchronization in STDP-driven memristive neural networks with time-varying topology

Synchronization is a widespread phenomenon in the brain. Despite numerous studies, the specific parameter configurations of the synaptic network structure and learning rules needed to achieve robust and enduring synchronization in neurons driven by spike-timing-dependent plasticity (STDP) and temporal networks subject to homeostatic structural plasticity (HSP) rules remain unclear. Here, we bridge this gap by determining the configurations required to achieve high and stable degrees of complete synchronization (CS) and phase synchronization (PS) in time-varying small-world and random neural networks driven by STDP and HSP. In particular, we found that decreasing P (which enhances the strengthening effect of STDP on the average synaptic weight) and increasing F (which speeds up the swapping rate of synapses between neurons) always lead to higher and more stable degrees of CS and PS in small-world and random networks, provided that the network parameters such as the synaptic time delay tau _c, the average degree langle k rangle, and the rewiring probability beta have some appropriate values. When tau _c, langle k rangle, and beta are not fixed at these appropriate values, the degree and stability of CS and PS may increase or decrease when F increases, depending on the network topology. It is also found that the time delay tau _c can induce intermittent CS and PS whose occurrence is independent F. Our results could have applications in designing neuromorphic circuits for optimal information processing and transmission via synchronization phenomena.

Learning the Vector Coding of Egocentric Boundary Cells from Visual Data.

The use of spatial maps to navigate through the world requires a complex ongoing transformation of egocentric views of the environment into position within the allocentric map. Recent research has discovered neurons in retrosplenial cortex and other structures that could mediate the transformation from egocentric views to allocentric views. These egocentric boundary cells respond to the egocentric direction and distance of barriers relative to an animal's point of view. This egocentric coding based on the visual features of barriers would seem to require complex dynamics of cortical interactions. However, computational models presented here show that egocentric boundary cells can be generated with a remarkably simple synaptic learning rule that forms a sparse representation of visual input as an animal explores the environment. Simulation of this simple sparse synaptic modification generates a population of egocentric boundary cells with distributions of direction and distance coding that strikingly resemble those observed within the retrosplenial cortex. Furthermore, some egocentric boundary cells learnt by the model can still function in new environments without retraining. This provides a framework for understanding the properties of neuronal populations in the retrosplenial cortex that may be essential for interfacing egocentric sensory information with allocentric spatial maps of the world formed by neurons in downstream areas, including the grid cells in entorhinal cortex and place cells in the hippocampus.SIGNIFICANCE STATEMENT The computational model presented here demonstrates that the recently discovered egocentric boundary cells in retrosplenial cortex can be generated with a remarkably simple synaptic learning rule that forms a sparse representation of visual input as an animal explores the environment. Additionally, our model generates a population of egocentric boundary cells with distributions of direction and distance coding that strikingly resemble those observed within the retrosplenial cortex. This transformation between sensory input and egocentric representation in the navigational system could have implications for the way in which egocentric and allocentric representations interface in other brain areas.

Spiking neural P systems with long-term potentiation and depression

Spiking neural P systems (SNP systems) are parallel, distributed, low energy consuming, non-deterministic, interpretable and simple structured computational models abstracted from biological neural systems. In this work, to make SNP systems have the ability to adjust synaptic weights according to changes in the environment and boost SNP systems' learning ability and intelligence, long-term potentiation/depression mechanisms in the signal transmission of two neurons are abstracted from biological neural systems and introduced into SNP systems. As result, a type of novel SNP system, called SNP systems with long-term potentiation and depression (LTPD-SNP systems), is developed. Based on novel synaptic weight learning rules in LTPD-SNP systems, the weights of synapses change dynamically depending on the intensity and timing of the spikes passing through them and their own properties. The universality of LTPD-SNP systems as number generation and acceptance devices is proved. A uniform solution to SAT problem is designed to study the computational efficiency of the LTPD-SNP systems. Simultaneously, an anomaly detection method for AC engines based on LTPD-SNP systems is constructed to show how the system can solve a learning task. This work improves the learning ability and intelligence of SNP systems, and offers some ideas for creating brain-like intelligent computational models.

α-Fe2O3-based artificial synaptic RRAM device for pattern recognition using artificial neural networks

We report on the α -Fe2O3-based artificial synaptic resistive random access memory device, which is a promising candidate for artificial neural networks (ANN) to recognize the images. The device consists of a structure Ag/α-Fe2O3/FTO and exhibits non-volatility with analog resistive switching characteristics. We successfully demonstrated synaptic learning rules such as long-term potentiation, long-term depression, and spike time-dependent plasticity. In addition, we also presented off-chip training to obtain good accuracy by backpropagation algorithm considering the synaptic weights obtained from α-Fe2O3 based artificial synaptic device. The proposed α-Fe2O3-based device was tested with the FMNIST and MNIST datasets and obtained a high pattern recognition accuracy of 88.06% and 97.6% test accuracy respectively. Such a high pattern recognition accuracy is attributed to the combination of the synaptic device performance as well as the novel weight mapping strategy used in the present work. Therefore, the ideal device characteristics and high ANN performance showed that the fabricated device can be useful for practical ANN implementation.

Inverse stochastic resonance in modular neural network with synaptic plasticity

This work explores the inverse stochastic resonance (ISR) induced by bounded noise and the multiple inverse stochastic resonance induced by time delay by constructing a modular neural network, where the modified Oja’s synaptic learning rule is employed to characterize synaptic plasticity in this network. Meanwhile, the effects of synaptic plasticity on the ISR dynamics are investigated. Through numerical simulations, it is found that the mean firing rate curve under the influence of bounded noise has an inverted bell-like shape, which implies the appearance of ISR. Moreover, synaptic plasticity with smaller learning rate strengthens this ISR phenomenon, while synaptic plasticity with larger learning rate weakens or even destroys it. On the other hand, the mean firing rate curve under the influence of time delay is found to exhibit a decaying oscillatory process, which represents the emergence of multiple ISR. However, the multiple ISR phenomenon gradually weakens until it disappears with increasing noise amplitude. On the same time, synaptic plasticity with smaller learning rate also weakens this multiple ISR phenomenon, while synaptic plasticity with larger learning rate strengthens it. Furthermore, we find that changes of synaptic learning rate can induce the emergence of ISR phenomenon. We hope these obtained results would provide new insights into the study of ISR in neuroscience.

H2Learn: High-Efficiency Learning Accelerator for High-Accuracy Spiking Neural Networks

Although spiking neural networks (SNNs) take benefits from the bioplausible neural modeling, the low accuracy under the common local synaptic plasticity learning rules limits their application in many practical tasks. Recently, an emerging SNN supervised learning algorithm inspired by backpropagation through time (BPTT) from the domain of artificial neural networks (ANNs) has successfully boosted the accuracy of SNNs, and helped improve the practicability of SNNs. However, current general-purpose processors suffer from low efficiency when performing BPTT for SNNs due to the ANN-tailored optimization. On the other hand, current neuromorphic chips cannot support BPTT because they mainly adopt local synaptic plasticity rules for simplified implementation. In this work, we propose H2Learn, a novel architecture that can achieve high efficiency for BPTT-based SNN learning, which ensures high accuracy of SNNs. At the beginning, we characterized the behaviors of BPTT-based SNN learning. Benefited from the binary spike-based computation in the forward pass and weight update, we first design look-up table (LUT)-based processing elements in the forward engine and weight update engine to make accumulations implicit and to fuse the computations of multiple input points. Second, benefited from the rich sparsity in the backward pass, we design a dual-sparsity-aware backward engine, which exploits both input and output sparsity. Finally, we apply a pipeline optimization between different engines to build an end-to-end solution for the BPTT-based SNN learning. Compared with the modern NVIDIA V100 GPU, H2Learn achieves <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$7.38\times $ </tex-math></inline-formula> area saving, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$5.74-10.20\times $ </tex-math></inline-formula> speedup, and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$5.25-7.12\times $ </tex-math></inline-formula> energy saving on several benchmark datasets.

Postsynaptic burst reactivation of hippocampal neurons enables associative plasticity of temporally discontiguous inputs.

A fundamental unresolved problem in neuroscience is how the brain associates in memory events that are separated in time. Here, we propose that reactivation-induced synaptic plasticity can solve this problem. Previously, we reported that the reinforcement signal dopamine converts hippocampal spike timing-dependent depression into potentiation during continued synaptic activity (Brzosko et al., 2015). Here, we report that postsynaptic bursts in the presence of dopamine produce input-specific LTP in mouse hippocampal synapses 10 min after they were primed with coincident pre- and post-synaptic activity (post-before-pre pairing; Δt = -20 ms). This priming activity induces synaptic depression and sets an NMDA receptor-dependent silent eligibility trace which, through the cAMP-PKA cascade, is rapidly converted into protein synthesis-dependent synaptic potentiation, mediated by a signaling pathway distinct from that of conventional LTP. This synaptic learning rule was incorporated into a computational model, and we found that it adds specificity to reinforcement learning by controlling memory allocation and enabling both 'instructive' and 'supervised' reinforcement learning. We predicted that this mechanism would make reactivated neurons activate more strongly and carry more spatial information than non-reactivated cells, which was confirmed in freely moving mice performing a reward-based navigation task.

Computational Modeling of Structural Synaptic Plasticity in Echo State Networks.

Most existing studies on computational modeling of neural plasticity have focused on synaptic plasticity. However, regulation of the internal weights in the reservoir based on synaptic plasticity often results in unstable learning dynamics. In this article, a structural synaptic plasticity learning rule is proposed to train the weights and add or remove neurons within the reservoir, which is shown to be able to alleviate the instability of the synaptic plasticity, and to contribute to increase the memory capacity of the network as well. Our experimental results also reveal that a few stronger connections may last for a longer period of time in a constantly changing network structure, and are relatively resistant to decay or disruptions in the learning process. These results are consistent with the evidence observed in biological systems. Finally, we show that an echo state network (ESN) using the proposed structural plasticity rule outperforms an ESN using synaptic plasticity and three state-of-the-art ESNs on four benchmark tasks.

Biologically plausible single-layer networks for nonnegative independent component analysis.

An important problem in neuroscience is to understand how brains extract relevant signals from mixtures of unknown sources, i.e., perform blind source separation. To model how the brain performs this task, we seek a biologically plausible single-layer neural network implementation of a blind source separation algorithm. For biological plausibility, we require the network to satisfy the following three basic properties of neuronal circuits: (i) the network operates in the online setting; (ii) synaptic learning rules are local; and (iii) neuronal outputs are nonnegative. Closest is the work by Pehlevan et al.(Neural Comput 29:2925-2954, 2017), which considers nonnegative independent component analysis (NICA), a special case of blind source separation that assumes the mixture is a linear combination of uncorrelated, nonnegative sources. They derive an algorithm with a biologically plausible 2-layer network implementation. In this work, we improve upon their result by deriving 2 algorithms for NICA, each with a biologically plausible single-layer network implementation. The first algorithm maps onto a network with indirect lateral connections mediated by interneurons. The second algorithm maps onto a network with direct lateral connections and multi-compartmental output neurons.

Top electrode modulated W/Ag/MgO/Au resistive random access memory for improved electronic synapse performance

Resistive random access memory (ReRAM) is touted to replace silicon-based flash memory due to its low operating voltage, fast access speeds, and the potential to scale down to nm range for ultra-high density storage. In addition, its ability to retain multi-level resistance states makes it suitable for neuromorphic computing application. Here, we develop a cationic ReRAM with a sputtered MgO as the insulating layer. The resistive switching properties of the Ag/MgO/Au ReRAM stack reveal a strong dependence on the sputtering conditions of MgO. Due to the highly stable sputtered MgO, repeatable resistive switching memory is achieved with a low ON voltage of ∼0.7 V and a memory window of ∼1 × 105. Limiting Ag diffusion through a modified top electrode in the W/Ag/MgO/Au stack significantly reduces the abruptness of resistive switching, thereby demonstrating analog switching capability. This phenomenon is evident in the improved linearity and symmetry of potentiation and depression weight modulation pulses, demonstrating ideal Hebbian synaptic learning rules.

Investigation of STDP mechanisms for memristor circuits

Memristive systems have become popular as a research field after the experimental realization of a memristor by the HP Labs research team. Researchers have demonstrated the synaptic behavior of the memristor in addition to its nonlinear and nonvolatile characteristics. Synapses play an important role in the learning process and spike-timing-dependent plasticity (STDP) is a fundamental synaptic learning rule. Because of problems with finding the memristor on the market as a discrete circuit element, researchers have designed many types of memristor circuits to emulate memristors. However, no information is available about the STDP behavior of these designed memristor emulators. This paper investigated the STDP property of some basic memristor circuits and the mathematical and simulation results demonstrated that none of them could support the STDP learning rule.

Natural-gradient learning for spiking neurons.

In many normative theories of synaptic plasticity, weight updates implicitly depend on the chosen parametrization of the weights. This problem relates, for example, to neuronal morphology: synapses which are functionally equivalent in terms of their impact on somatic firing can differ substantially in spine size due to their different positions along the dendritic tree. Classical theories based on Euclidean-gradient descent can easily lead to inconsistencies due to such parametrization dependence. The issues are solved in the framework of Riemannian geometry, in which we propose that plasticity instead follows natural-gradient descent. Under this hypothesis, we derive a synaptic learning rule for spiking neurons that couples functional efficiency with the explanation of several well-documented biological phenomena such as dendritic democracy, multiplicative scaling, and heterosynaptic plasticity. We therefore suggest that in its search for functional synaptic plasticity, evolution might have come up with its own version of natural-gradient descent.

A TSTDP memristive synapse based on a comprehensive mathematical model of memory-TFT threshold voltage shift

New emerging nano-scale technologies like hydrogenated noncrystalline-silicon thin-film transistors (TFTs) and memristors, fabricated at low temperatures and over large areas, permit low-cost processing and 3D integration with CMOS cores. Here, we aim to propose a mathematical model which explains the memory-TFT threshold voltage shift due to the gate bias instability. Then, based on this mathematical approach, we propose a novel learning synapse composed of a voltage/flux driven memristor in parallel with a common-source memory-TFT with a memristive load. The proposed device realizes the triplet-based spike-timing-dependent plasticity rule (TSTDP) as a more realistic form of learning than the purely pair-based STDP (PSTDP). PSTDP is a synaptic learning rule which utilizes a constant-frequency pairing protocol to induce synaptic weight change and cannot explain the modification due to the frequency changes of spike pairs, and also the outcomes of triplet and quadruplet experiments. However, TSTDP improves the learning capabilities of the conventional PSTDP and reproduces the results of more electrophysiological experiments. In this paper, we apply various spike patterns like different-frequency and different-timing spike pairs, spike triplets, and quadruplets to the proposed device. Our simulations confirm a close match with the experimental data sets of real biological synapses.

Analog Resistive Switching in Reduced Graphene Oxide and Chitosan‐Based Bio‐Resistive Random Access Memory Device for Neuromorphic Computing Applications

Bio‐resistive random access memory (Bio‐RRAM) devices are important because they are biocompatible and biodegradable. Herein, analog unipolar resistive switching and bipolar resistive switching in chitosan and reduced graphene oxide + chitosan (RGO+chitosan)‐based Bio‐RRAM devices are demonstrated, respectively. Endurance and retentivity of the Ag/RGO+chitosan/FTO‐based Bio‐RRAM device demonstrate good stability and nonvolatile in nature. Conductive atomic force microscope measurements infer the formation of the 1D conduction channels in the device, which further confirms the conduction channel mechanism in the RGO+chitosan‐based Bio‐RRAM device. Further, synaptic learning rules such as long‐term potentiation (LTP), long‐term depression, and spike time‐dependent plasticity on Ag/RGO+chitosan/FTO‐based device are also demonstrated. Estimated relaxation time infers more time for forgetting than learning. These results suggest that Ag/RGO+chitosan/FTO synaptic Bio‐RRAM device would indeed be a potential candidate for future neuromorphic computing applications.

Bidirectional synaptic plasticity rapidly modifies hippocampal representations.

Learning requires neural adaptations thought to be mediated by activity-dependent synaptic plasticity. A relatively non-standard form of synaptic plasticity driven by dendritic calcium spikes, or plateau potentials, has been reported to underlie place field formation in rodent hippocampal CA1 neurons. Here, we found that this behavioral timescale synaptic plasticity (BTSP) can also reshape existing place fields via bidirectional synaptic weight changes that depend on the temporal proximity of plateau potentials to pre-existing place fields. When evoked near an existing place field, plateau potentials induced less synaptic potentiation and more depression, suggesting BTSP might depend inversely on postsynaptic activation. However, manipulations of place cell membrane potential and computational modeling indicated that this anti-correlation actually results from a dependence on current synaptic weight such that weak inputs potentiate and strong inputs depress. A network model implementing this bidirectional synaptic learning rule suggested that BTSP enables population activity, rather than pairwise neuronal correlations, to drive neural adaptations to experience.