Hebbian Research Articles

Continual learning with hebbian plasticity in sparse and predictive coding networks: a survey and perspective

Abstract Recently, the use of bio-inspired learning techniques such as Hebbian learning and its closely-related Spike-Timing-Dependent Plasticity (STDP) variant have drawn significant attention for the design of compute-efficient AI systems that can continuously learn on-line at the edge. A key differentiating factor regarding this emerging class of neuromorphic continual learning system lies in the fact that learning must be carried using a data stream received in its natural order, as opposed to conventional gradient-based offline training, where a static training dataset is assumed available a priori and randomly shuffled to make the training set independent and identically distributed (i.i.d). In contrast, the emerging class of neuromorphic continual learning systems covered in this survey must learn to integrate new information on the fly in a non-i.i.d manner, which makes these systems subject to catastrophic forgetting. In order to build the next generation of neuromorphic AI systems that can continuously learn at the edge, a growing number of research groups are studying the use of Sparse and Predictive Coding-based Hebbian neural network architectures and the related Spiking Neural Networks (SNNs) equipped with STDP learning. However, since this research field is still emerging, there is a need for providing a holistic view of the different approaches proposed in the literature so far. To this end, this survey covers a number of recent works in the field of neuromorphic continual learning based on state-of-the-art Sparse and Predictive Coding technology; provides background theory to help interested researchers quickly learn the key concepts; and discusses important future research questions in light of the different works covered in this paper. It is hoped that this survey will contribute towards future research in the field of neuromorphic continual learning.

Applying Numerical Simulation to Model Varying Process and Cell Parameters during the Electrolyte Filling Process of Lithium‐Ion Batteries

The electrolyte filling process is considered one of the bottlenecks of lithium‐ion battery production due mainly to the long electrolyte wetting times. Additionally, the required experimental process design is time and material‐intensive, increasing the development costs of new materials or cell designs. A model of the filling process would allow for more efficient cell production, but until now, the published models have mainly been focused on individual components on a pore scale. Within the scope of this work, the model setup for a holistic examination of the electrolyte filling process is shown, allowing the study of the electrolyte wetting on a cell scale. The characteristic values of a porous medium, such as the permeability, are calculated with a microsimulation of an anode and a cathode pore structure. These values are then transferred to the ANSYS porous media model, and cell scale simulations are performed. Two cell formats and variations in the evacuation pressure and electrolyte temperature are simulated and compared to experimental wetting data. The results show that the simulation successfully models the wetting behavior for the investigated cell formats and cell assembly types, validating the model with experimental data both qualitatively and quantitatively.

Quantifying Heterogeneous Degradation Pathways and Deformation Fields in Solid‐State Batteries

AbstractSolid‐state batteries are compelling candidates for next‐generation energy storage devices, promising both high energy density and improved safety, by utilizing metallic Li as the negative electrode. However, they suffer from poor cyclability and rate capability, which limits their wide application. Degradation in these devices occurs through complex mechanical, chemical and electrochemical pathways, all of which produce heterogeneous deformation fields. Therefore, isolating solid‐state degradation mechanisms, and explicitly linking them to the associated deformation fields requires a multimodal characterization strategy. Here, a novel 3‐D, in situ methodology for linking degradation to deformation in solid‐state cells is presented. X‐ray imaging is used to measure the morphological degradation, and combined with X‐ray diffraction to quantify (electro)chemical aspects. Finally, the heterogeneous stress fields from these various pathways are mapped in situ. This heterogeneity is shown globally, from the interface to the bulk electrolyte, as well as locally, around features such as cracks and voids. Through these analyses, it is possible to delineate the effects of solid electrolyte processing, cell assembly, and cycling on the end‐of‐life state of the cell. Moreover, the importance of stress mitigation in these cells is highlighted, with mean stresses around the interface and some cracks comfortably exceeding the elastic limit of Li.

Learning in Associative Networks Through Pavlovian Dynamics

Abstract Hebbian learning theory is rooted in Pavlov’s classical conditioning While mathematical models of the former have been proposed and studied in the past decades, especially in spin glass theory, only recently has it been numerically shown that it is possible to write neural and synaptic dynamics that mirror Pavlov conditioning mechanisms and also give rise to synaptic weights that correspond to the Hebbian learning rule. In this letter we show that the same dynamics can be derived with equilibrium statistical mechanics tools and basic and motivated modeling assumptions. Then we show how to study the resulting system of coupled stochastic differential equations assuming the reasonable separation of neural and synaptic timescale. In particular, we analytically demonstrate that this synaptic evolution converges to the Hebbian learning rule in various settings and compute the variance of the stochastic process. Finally, drawing from evidence on pure memory reinforcement during sleep stages, we show how the proposed model can simulate neural networks that undergo sleep-associated memory consolidation processes, thereby proving the compatibility of Pavlovian learning with dreaming mechanisms.

Enhanced capacitance of nickel ferrite decorated laser-induced graphene nanocomposite for symmetric supercapacitor device

The demand for high-performance energy storage devices, such as supercapacitors, has driven the exploration of hybrid electrode materials with enhanced charge storage capabilities. This study investigates the development of a novel NiFe2O4 (NFO) decorated LIG nanocomposite as an advanced electrode material for supercapacitor applications. Firstly, the LIG was synthesized by direct laser-conversion of polyimide to graphene followed by in-situ decoration of NFO nanoparticles by the drop-casting method. This facile strategy resulted in NFO/LIG nanocomposite with well-dispersed NFO nanoparticles, as verified by Raman spectroscopy. Furthermore, the morphological and structural analysis of the nanocomposite was carried out by using FESEM, EDX, and HRTEM. Additionally, XPS analysis revealed the existence of Ni2+ and Fe3+ ions which create redox active sites within NFO/LIG and permit the diffusion of electrolyte ions to form redox species. The unique physicochemical properties of graphene, combined with the pseudocapacitive characteristics of NFO, are leveraged to enhance specific capacitance, energy density, and overall electrochemical performance. Electrochemical results showed a remarkable increase in the specific capacitance of NFO/LIG nanocomposite (198 mF/cm2 at 1.5 mA/cm2), as compared to pure LIG (65 mF/cm2 1.5 mA/cm2). When utilized as a symmetric supercapacitor, the device offers areal specific capacitance of 44 mF/cm2 at 1.5 mA/cm2. In addition to this, a pouch cell assembly was designed on a flexible substrate using PVA/KOH gel electrolyte demonstrating 18.7 mF/cm2 at 5 mV/s, highlighting the potential use of NFO/LIG electrodes in energy storage applications.

Study on Mitigating Membrane Degradation in Degraded Membrane Electrode Assembly of Polymer Electrolyte Membrane Fuel Cell through Temperature and Humidity Control

Study on Mitigating Membrane Degradation in Degraded Membrane Electrode Assembly of Polymer Electrolyte Membrane Fuel Cell through Temperature and Humidity Control

Endometrial extracellular vesicles regulate processes related to embryo development and implantation in human blastocysts.

What is the transcriptomic response of human blastocysts following internalization of extracellular vesicles (EVs) secreted by the human endometrium? EVs secreted by the maternal endometrium induce a transcriptomic response in human embryos that modulates molecular mechanisms related to embryo development and implantation. EVs mediate intercellular communication by transporting various molecules, and endometrial EVs have been postulated to be involved in the molecular regulation of embryo implantation. Our previous studies showed that endometrial EVs carry miRNAs and proteins associated with implantation events that can be taken up by human blastocysts; however, no studies have yet investigated the transcriptomic response of human embryos to this EV uptake, which is crucial to demonstrate the functional significance of this communication system. A prospective descriptive study was performed. Primary human endometrial epithelial cells (pHEECs), derived from endometrial biopsies collected from fertile oocyte donors (n = 20), were cultured in vitro to isolate secreted EVs. Following EV characterization, Day 5 human blastocysts (n = 24) were cultured in the presence or absence of the EVs for 24 h and evaluated by RNA-sequencing. EVs were isolated from the conditioned culture media using ultracentrifugation, and characterization was performed using western blot, nanoparticle tracking analysis, and transmission electron microscopy. Human blastocysts were devitrified, divided into two groups (n = 12/group), and cultured in vitro for 24 h with or without previously isolated EVs. RNA-sequencing analysis was performed, and DESeq2 was used to identify differentially expressed genes (DEGs) (FDR < 0.05). QIAGEN Ingenuity Pathway Analysis was used to perform the functional enrichment analysis and integration with our recently published data from the pHEECs' EV-miRNA cargo. Characterization confirmed the isolation of EVs from pHEECs' conditioned culture media. Among the DEGs in blastocysts co-cultured with EVs, we found 519 were significantly upregulated and 395 were significantly downregulated. These DEGs were significantly enriched in upregulated functions related to embryonic development, cellular invasion and migration, cell cycle, cellular organization and assembly, gene expression, and cell viability; and downregulated functions related to cell death and DNA fragmentation. Further, the intracellular signaling pathways regulated by the internalization of endometrial EVs were previously related to early embryo development and implantation potential, for their role in pluripotency, cellular homeostasis, early embryogenesis, and implantation-related processes. Finally, integrating data from miRNA cargo of EVs, we found that the miRNAs carried by endometrial EVs targeted nearly 80% of the DEGs in human blastocysts. This is an in vitro study in which conditions of endometrial cell culture could not mimic the intrauterine environment. This study provides novel insights into the functional relevance of EVs secreted by the human endometrium, and particularly the role of EV-miRNA regulation on global transcriptome behavior of human blastocysts during early embryogenesis and embryo implantation. It provides potential biomarkers that could become useful diagnostic targets for predicting implantation success. This study was supported by the Spanish Ministry of Education through FPU awarded to M.S.-B. (FPU18/03735), Generalitat Valenciana through VALi+d Programme awarded to M.C.C.-G. (ACIF/2019/139), and Instituto de Salud Carlos III and cofounded by the European Social Fund (ESF) "Investing in your future" through the Miguel Servet Program (CP20/00120 [H.F.]; CP19/00149 [I.C.]). The authors have no conflicts of interest to disclose. N/A.

Early and late place cells during postnatal development of the hippocampus

A proportion of hippocampal CA1 neurons function as place cells from the onset of navigation, which are referred to as early place cells. It is not clear whether this subset of neurons is predisposed to become place cells during early stages, or if all neurons have this potential. Here, we longitudinally imaged the activity of CA1 neurons in developing male rats during navigation with both one-photon and two-photon microscopy. Our results suggested that a largely consistent population of cells functioned as early place cells, demonstrating higher spatial coding abilities across environments and a tendency to form more synchronous cell assemblies. Early place cells were present in both deep and superficial layers of CA1. Cells in the deep layer exhibited greater synchrony than those in the superficial layer during early ages. These results support the theory that an initial cognitive map is primarily shaped by a predetermined set of hippocampal cells.

Insights of BDAPbI4-Based Flexible Memristor for Artificial Synapses and In-Memory Computing.

Inspired by brain-like spiking computational frameworks, neuromorphic computing-brain-inspired computing for machine intelligence promises to realize artificial intelligence (AI) while reducing the energy requirements of computing platforms. In this work, we show the potential of advanced learnings of butane-1,4-diammonium based low-dimensional Dion-Jacobson hybrid perovskite (BDAPbI4) memristor devices in the realm of artificial synapses and neuromorphic computing. Memristors validate Hebbian learning rules with various spike-dependent plasticity within a 10 ± 2 ms time frame, reminiscent of the human brains under flat and bending conditions (∼5 mm radium). A high recognition accuracy of ∼94% of handwritten images from the MNIST database via an artificial neural network (ANN) is achieved with only 50 epochs. An efficient demonstration of second-order memristors and the Pavlovian dog experiment exhibit significant promise in expediting learning and memory consolidation. To showcase the in-memory computing potential, a flexible 4 × 4 crossbar array is designed with measured data retention up to ∼103 s along with 26 multilevel resistance states. The crossbar array is successfully programmed for the facile configurability of image "Z". In conclusion, the integration of supervised, unsupervised, and associative learning holds great promise across a spectrum of future technologies, ranging from the realm of spiking neural networks to neuromorphic computing, brain-machine interfaces, and adaptive control systems.

Organic-Inorganic Hybrid Solid Composite Electrolytes for High Energy Density Lithium Batteries: Combining Manufacturability, Conductivity, and Stability.

The deployment of solid and quasi-solid electrolytes in lithium metal batteries is envisioned to push their energy densities to even higher levels, in addition to providing enhanced safety. This article discusses a set of hybrid solid composite electrolytes which combine functional properties with electrode compatibility and manufacturability. Their anodic stability >5V versus Li+/Li and compatibility with lithium metal stem from the incorporated ionic liquid electrolyte, whereas the organic-inorganic hybrid host structure boosts their conductivity up to 2.7mScm-1 at room temperature. The absence of strong acids enables compatibility with porous NMC811 electrodes. Liquid precursor solutions can be readily impregnated into porous electrodes, facilitating cell assembly. Electrolytes containing TFSI- as the only anion have a superior compatibility toward high-voltage positive electrode materials, whereas electrolytes containing both FSI- and TFSI- have a better compatibility toward lithium metal. Using the former as catholyte and the latter as anolyte, NMC811/Li coin cells retain up to 100% of their initial capacity after 100 cycles (0.2C, 3.0-4.4V vs Li+/Li). Because of their unprecedented combination of functional properties, electrode compatibility, and manufacturability, these hybrid solid composite electrolytes are potential candidates for the further development of lithium metal battery technology.

Perancangan Model Jaringan Syaraf Tiruan untuk Memprediksi Penyakit Demam Berdarah Menggunakan Algoritma Hebb Rule

Dangeu dengue fever or what we often call dengue fever is a disease transmitted by the Aedes aegypti mosquito and caused by the dengue virus. This disease can potentially cause serious complications if it does not receive proper treatment. In this research, the author uses the application of artificial neural networks with the Hebb rule approach to predict the risk level of dengue fever. Predictions are made based on factors such as weather conditions, population density and historical case data that influence this disease. The Hebb rule is used in this research because of its ability to strengthen connections between neurons based on the input patterns they receive, so it is hoped that it can produce more accurate predictions. Test results show that this method has a fairly high level of accuracy in predicting the pattern of dengue fever cases in an area. This research indicates that the application of artificial neural networks with the Hebb rule can be an effective tool for related parties in taking preventive measures to minimize the number of dengue cases in the future.

Can human brain connectivity explain verbal working memory?

ABSTRACT The ability of humans to store spoken words in verbal working memory and build extensive vocabularies is believed to stem from evolutionary changes in cortical connectivity across primate species. However, the underlying neurobiological mechanisms remain unclear. Why can humans acquire vast vocabularies, while non-human primates cannot? This study addresses this question using brain-constrained neural networks that realize between-species differences in cortical connectivity. It investigates how these structural differences support the formation of neural representations for spoken words and the emergence of verbal working memory, crucial for human vocabulary building. We develop comparative models of frontotemporal and occipital cortices, reflecting human and non-human primate neuroanatomy. Using meanfield and spiking neural networks, we simulate auditory word recognition and examine verbal working memory function. The “human models”, characterized by denser inter-area connectivity in core language areas, produced larger cell assemblies than the “monkey models”, with specific topographies reflecting semantic properties of the represented words. Crucially, longer-lasting reverberant neural activity was observed in human versus monkey architectures, compatible with robust verbal working memory, a necessary condition for vocabulary building. Our findings offer insights into the structural basis of human-specific symbol learning and verbal working memory, shedding light on humans’ unique capacity for large vocabulary acquisition.

Diversifying Multi-Head Attention in the Transformer Model

Recent studies have shown that, due to redundancy, some heads of the Transformer model can be pruned without diminishing the efficiency of the model. In this paper, we propose a constrained optimization algorithm based on Hebbian learning, which trains specific layers in the Transformer architecture in order to enforce diversification between the different heads in the multi-head attention module. The diversification of the heads is achieved through a single-layer feed-forward neural network that is added to the Transformer architecture and is trained with the proposed algorithm. We utilize the algorithm in three different architectural variations of the baseline Transformer model. In addition to the diversification of the heads, the proposed methodology can be used to prune the heads that capture redundant information. Experiments on diverse NLP tasks, including machine translation, text summarization, question answering and large language modeling, show that our proposed approach consistently improves the performance of baseline Transformer models.

Hierarchical behavior control by a single class of interneurons

Animal behavior is organized into nested temporal patterns that span multiple timescales. This behavior hierarchy is believed to arise from a hierarchical neural architecture: Neurons near the top of the hierarchy are involved in planning, selecting, initiating, and maintaining motor programs, whereas those near the bottom of the hierarchy act in concert to produce fine spatiotemporal motor activity. In Caenorhabditis elegans, behavior on a long timescale emerges from ordered and flexible transitions between different behavioral states, such as forward, reversal, and turn. On a short timescale, different parts of the animal body coordinate fast rhythmic bending sequences to produce directional movements. Here, we show that Sublateral Anterior A (SAA), a class of interneurons that enable cross-communication between dorsal and ventral head motor neurons, play a dual role in shaping behavioral dynamics on different timescales. On a short timescale, SAA regulate and stabilize rhythmic bending activity during forward movements. On a long timescale, the same neurons suppress spontaneous reversals and facilitate reversal termination by inhibiting Ring Interneuron M (RIM), an integrating neuron that helps maintain a behavioral state. These results suggest that feedback from a lower-level cell assembly to a higher-level command center is essential for bridging behavioral dynamics at different levels.

Combining finite element analysis and reinforcement learning for optimal grip point planning of flexible components

Abstract Handling large flexible components is still a challenge in many industries. Examples include the handling of fibre-reinforced plastics or the assembly of industrial-scale electrolytic cells. Difficulties often arise in the design of suitable endeffectors. Inefficient gripping point design, i.e. the total number and positioning of grippers, can lead to increased stress and deflection of the component being handled. To counteract this, endeffectors are often oversized resulting in the use of more grippers than needed. Correspondingly, heavier moving masses imply longer handling times as well as higher energy consumption. This paper presents a process for planning and optimising gripping points for large flexible components. In addition to the shape of the component, actual dynamic loads of the handling path are also taken into account. The key element to the process is an optimisation algorithm based on reinforcement learning and trained using an finite element method (FEM) simulation. After computing a desirable starting configuration, the algorithm optimises the placement of gripper positions while aiming for a reduced total number. In addition, the optimisation has prescribed handling limits, such as physical and geometric constraints, that must not be exceeded. It was shown that the algorithm satisfactorily optimises the gripping points for dynamic loads for different materials and shapes. Furthermore, it has been shown that the computation of an initial configuration yields preferable results for simple components, yet requiring optimisation in the case of more complex shapes.

Adaptive Depth Estimation via SoftHebbLayer in a Hybrid ResNet-UNet Architecture

Abstract. Depth estimation is a key task in computer vision, critical for applications such as autonomous navigation and augmented reality. This paper introduces a novel hybrid neural network that combines ResNet and UNet architectures with a SoftHebbLayer, inspired by Hebbian learning principles, to improve depth estimation from RGB images. The ResNet backbone extracts robust hierarchical features, while the UNet decoder reconstructs fine-grained depth maps. The SoftHebbLayer dynamically adjusts feature connections based on co-activation, enhancing the models adaptability to diverse environments. This approach addresses common challenges in depth estimation, including poor generalization and computational inefficiency. We evaluated the model on the DIODE dataset, achieving strong results in Mean Squared Error (MSE), which is 0.0800 and Root Mean Squared Error (RMSE), which is 0.2805, demonstrating improved accuracy in both indoor and outdoor scenes. While the model excels in precision, further refinement is needed to reduce computational overhead and improve performance in challenging environments. This research paves the way for more efficient, adaptable depth estimation models, with potential applications in mobile robotics and real-time edge computing systems.

Metaplasticity and memory in multilevel recurrent feed-forward networks.

Network systems can exhibit memory effects in which the interactions between different pairs of nodes adapt in time, leading to the emergence of preferred connections, patterns, and subnetworks. To a first approximation, this memory can be modeled through a "plastic" Hebbian or homophily mechanism, in which edges get reinforced proportionally to the amount of information flowing through them. However, recent studies on glia-neuron networks have highlighted how memory can evolve due to more complex dynamics, including multilevel network structures and "metaplastic" effects that modulate reinforcement. Inspired by those systems, here we develop a simple and general model for the dynamics of an adaptive network with an additional metaplastic mechanism that varies the rate of Hebbian strengthening of its edge connections. The metaplastic term acts on a second network level in which edges are grouped together, simulating local, longer timescale effects. Specifically, we consider a biased random walk on a cyclic feed-forward network. The random walk chooses its steps according to the weights of the network edges. The weights evolve through a Hebbian mechanism modulated by a metaplastic reinforcement, biasing the walker to prefer edges that have been already explored. We study the dynamical emergence (memorization) of preferred paths and their retrieval and identify three regimes: one dominated by the Hebbian term, one in which the metareinforcement drives memory formation, and a balanced one. We show that, in the latter two regimes, metareinforcement allows the retrieval of a previously stored path even after the weights have been reset to zero to erase Hebbian memory.

The α/β hydrolase domain-containing protein 1 (ABHD1) acts as a lysolipid lipase and is involved in lipid droplet formation

Abstract Lipid droplets (LDs) are the major sites of lipid and energy homeostasis. However, few LD biogenesis proteins have been identified. Using model microalga Chlamydomonas, we show that ABHD1, an α/β-hydrolase domain-containing protein, is localized to the LD surface and stimulates LD formation through two actions: one enzymatic and one structural. The knockout mutants contained similar amounts of triacylglycerols (TAG) but their LDs showed a higher content of lyso-derivatives of betaine lipid diacylglyceryl-N,N,N-trimethylhomoserine (DGTS). Over-expression of ABHD1 increased LD abundance and boosted TAG content. Purified recombinant ABHD1 hydrolyzed lyso-DGTS, producing a free fatty acid and a glyceryltrimethylhomoserine. In vitro droplet-embedded vesicles showed ABHD1 promoted LD emergence. Taken together, these results identify ABHD1 as a new player in LD formation by its lipase activity on lyso-DGTS and by its distinct biophysical property. This study further suggests that lipases targeted to LDs and able to act on their polar lipid coat may be interesting tools to promote LD assembly in eukaryotic cells.

Synaptopodin: a key regulator of Hebbian plasticity.

Synaptopodin, an actin-associated protein found in a subset of dendritic spines in telencephalic neurons, has been described to influence both functional and morphological plasticity under various plasticity paradigms. Synaptopodin is necessary and sufficient for the formation of the spine apparatus, stacks of smooth endoplasmic reticulum cisternae. The spine apparatus is a calcium store that locally regulates calcium dynamics in response to different patterns of activity and is also thought to be a site for local protein synthesis. Synaptopodin is present in ~30% of telencephalic large dendritic spines in vivo and in vitro highlighting the heterogeneous microanatomy and molecular architecture of dendritic spines, an important but not well understood aspect of neuroplasticity. In recent years, it has become increasingly clear that synaptopodin is a formidable regulator of multiple mechanisms essential for learning and memory. In fact, synaptopodin appears to be the decisive factor that determines whether plasticity can occur, acting as a key regulator for synaptic changes. In this review, we summarize the current understanding of synaptopodin's role in various forms of Hebbian synaptic plasticity.

Highly flexible SCOF proton exchange membrane reinforced with PTFE to enhance fuel cell power density

Sulfonated covalent organic frameworks (SCOFs) facilitate rapid proton conduction through densely ordered sulfonic acid groups, however, the brittleness of COFs self-supporting membranes often makes potential difficulty in fuel cell assembly and limits their power density. Herein, a highly flexible SCOF proton exchange membrane is developed through in-situ growth of a continuous BD(SO3H)2–COF microphase within porous PTFE networks. The strong hydrogen bonding between PTFE and BD(SO3H)2–COF contributes to the defect-free morphology of the BD(SO3H)2/PTFE membrane. The reinforce of PTFE network makes the membrane extremely high flexibility, achieving an elongation at break of 124.4 % even with a remarkably high SCOF mass proportion of 90 wt% (BD(SO3H)2/PTFE-0.9). This allows the membrane to be folded repeatedly, even in dry state. The swelling ratio in water at 80 °C is effectively restricted to 8.6 %, even with a high ion exchange capacity of 3.6 mmol g−1 and a water uptake of 68.2 %. The densely ordered sulfonic acid groups in continuous BD(SO3H)2–COF microphase contribute to a high proton conductivity up to 249.2 mSꞏcm−1 at 80 °C, approximately 1.5 folds that of Nafion 212. As a result, the BD(SO3H)2/PTFE-0.9 membrane achieves a fuel cell power density of 1195.3 mWꞏcm−2 at 80 °C, along with a high open circuit voltage of 1.01 V, surpassing the-state-of-the-art COF-based proton exchange membranes. This work provides a novel strategy to fabricate COFs into flexible and size scalable membranes, enhancing the performance of fuel cells.