Hebbian Research Articles

Unsupervised end-to-end training with a self-defined target

Abstract Designing algorithms for versatile AI hardware that can learn on the edge using both labeled and unlabeled data is challenging. Deep end-to-end training methods incorporating phases of self-supervised and supervised learning are accurate and adaptable to input data but self-supervised learning requires even more computational and memory resources than supervised learning, too high for current embedded hardware. Conversely, unsupervised layer-by-layer training, such as Hebbian learning, is more compatible with existing hardware but doesn't integrate well with supervised learning. 
To address this, we propose a method enabling networks or hardware designed for end-to-end supervised learning to also perform high-performance unsupervised learning by adding two simple elements to the output layer: Winner-Take-All (WTA) selectivity and homeostasis regularization. These mechanisms introduce a 'self-defined target' for unlabeled data, allowing purely unsupervised training for both fully-connected and convolutional layers using backpropagation or equilibrium propagation on datasets like MNIST (up to 99.2%), Fashion-MNIST (up to 90.3%), and SVHN (up to 81.5%).
We extend this method to semi-supervised learning, adjusting targets based on data type, achieving 96.6% accuracy with only 600 labeled MNIST samples in a multi-layer perceptron. Our results show that this approach can effectively enable networks and hardware initially dedicated to supervised learning to also perform unsupervised learning, adapting to varying availability of labeled data.

Lateral cell polarization drives organization of epithelia in sea anemone embryos and embryonic cell aggregates

One of the first organizing processes during animal development is the assembly of embryonic cells into epithelia. Common features unite epithelialization across select bilaterians, however, we know less about the molecular and cellular mechanisms that drive epithelial emergence in early branching nonbilaterians. In sea anemones, epithelia emerge both during embryonic development and after cell aggregation of dissociated tissues. Although adhesion is required to keep cells together, it is not clear whether cell polarization plays a role as epithelia emerge from disordered aggregates. Here, we use the embryos of the sea anemone Nematostella vectensis to investigate the evolutionary origins of epithelial development. We demonstrate that lateral cell polarization is essential for epithelial organization in both embryos and aggregates. With disrupted lateral polarization, cell contact in the aggregate is not sufficient to trigger epithelialization and further tissue development. Specifically, knockdown of the conserved lateral polarity and tumor suppressor protein Lethal giant larvae (Lgl) disrupts epithelia in developing embryos and impairs the capacity of dissociated cells to epithelialize from aggregates. In contrast to other systems, cells in Nematostella lgl knockdown embryos do not undergo excessive proliferation. Cells with reduced Lgl levels lose their columnar shape and proper positioning of their mitotic spindles and basal bodies. Due to misoriented divisions and aberrant shapes, cells arrange nonuniformly without forming a monolayer. Together our data show that, in Nematostella, Lgl drives epithelialization in embryos and cell aggregates through its effect on cell shape and organelle localization.

A flexible porous polyimide/copper composite film toward high-mass-loading anodes in lithium-ion batteries

Conventional metal foil current collectors are essential for electronic conduction in lithium-ion batteries, but the planar structure limits the fast transporting of electron and the power density. Herein, a light, conductive, self-extinguishing, and flexible porous polyimide/copper composite film (PIF/Cu) derived from a polyimide foam was fabricated via hot pressing and electroless plating methods. The honeycomb structure was illustrated in PIF/Cu under a 3D X-ray tomography microscope, revealing the successful construction of a 3D conductive network. PIF/Cu-30 possessed an electronic conductivity of as high as 3.6 × 105 S m−1 at a bulk density of 0.26 g cm−3 and exhibited exceptional structural stability even after successive and harsh mechanical loads. Graphite with a mass load of 12 mg cm−2 was loaded into the PIF/Cu-30 to prepare the anode for cell assembly. The cells using PIF/Cu-30 achieved a better performance in capacity retention (98.2 %@60 cycles) than those using traditional Cu foils (55.1 %@60 cycles), which resulted from the fast transport of electrons and Li-ion through the 3D conductive networks in the porous PIF/Cu-30.

Sensory experience steers representational drift in mouse visual cortex

Representational drift—the gradual continuous change of neuronal representations—has been observed across many brain areas. It is unclear whether drift is caused by synaptic plasticity elicited by sensory experience, or by the intrinsic volatility of synapses. Here, using chronic two-photon calcium imaging in primary visual cortex of female mice, we find that the preferred stimulus orientation of individual neurons slowly drifts over the course of weeks. By using cylinder lens goggles to limit visual experience to a narrow range of orientations, we show that the direction of drift, but not its magnitude, is biased by the statistics of visual input. A network model suggests that drift of preferred orientation largely results from synaptic volatility, which under normal visual conditions is counteracted by experience-driven Hebbian mechanisms, stabilizing preferred orientation. Under deprivation conditions these Hebbian mechanisms enable adaptation. Thus, Hebbian synaptic plasticity steers drift to match the statistics of the environment.

Integration of rate and phase codes by hippocampal cell-assemblies supports flexible encoding of spatiotemporal context.

Spatial information is encoded by location-dependent hippocampal place cell firing rates and sub-second, rhythmic entrainment of spike times. These rate and temporal codes have primarily been characterized in low-dimensional environments under limited cognitive demands; but how is coding configured in complex environments when individual place cells signal several locations, individual locations contribute to multiple routes and functional demands vary? Quantifying CA1 population dynamics of male rats during a decision-making task, here we show that the phase of individual place cells' spikes relative to the local theta rhythm shifts to differentiate activity in different place fields. Theta phase coding also disambiguates repeated visits to the same location during different routes, particularly preceding spatial decisions. Using unsupervised detection of cell assemblies alongside theoretical simulation, we show that integrating rate and phase coding mechanisms dynamically recruits units to different assemblies, generating spiking sequences that disambiguate episodes of experience and multiplexing spatial information with cognitive context.

Bilateral Alignment of Receptive Fields in the Olfactory Cortex.

Each olfactory cortical hemisphere receives ipsilateral odor information directly from the olfactory bulb and contralateral information indirectly from the other cortical hemisphere. Since neural projections to the olfactory cortex are disordered and non-topographic, spatial information cannot be used to align projections from the two sides like in the visual cortex. Therefore, how bilateral information is integrated in individual cortical neurons is unknown. We have found, in mice, that the odor responses of individual neurons to selective stimulation of each of the two nostrils are significantly correlated, such that odor identity decoding optimized with information arriving from one nostril transfers very well to the other side. Nevertheless, these aligned responses are asymmetric enough to allow decoding of stimulus laterality. Computational analysis shows that such matched odor tuning is incompatible with purely random connections but is explained readily by Hebbian plasticity structuring bilateral connectivity. Our data reveal that despite the distributed and fragmented sensory representation in the olfactory cortex, odor information across the two hemispheres is highly coordinated.Significance statement Like other sense organs, animals typically have two nostrils, but how odor information from the two sides is combined to build bilateral olfactory representations remains largely unknown. Grimaud et al. find that the responses of neurons in the olfactory cortex in awake mice to odors presented separately to the ipsilateral or contralateral nostril are significantly correlated, beyond chance. Such aligned responses could arise from Hebbian plasticity in interhemispheric connections that relies on common odor experiences across the two nostrils. While responses are correlated, the remaining asymmetries in responses to the two nostrils allowed decoding of stimulus laterality. This study points to unexpected order in an olfactory circuit and prompts future work on how olfactory experience can shape interhemispheric information integration.

Sensitivity of Capacity Fade in Vanadium Redox Flow Battery to Electrolyte Impurity Content.

The gradual capacity decrease of vanadium redox flow battery (VRFB) over long-term charge-discharge cycling is determined by electrolyte degradation. While it was initially believed that this degradation was solely caused by crossover, recent research suggests that oxidative imbalance induced by hydrogen evolution reaction (HER) also plays a significant role. In this work by using vanadium pentoxides with different impurities content, we prepared three grades of vanadium electrolyte. By measuring electrochemical properties on carbon felt electrode in three-electrode cell and VRFB membrane-electrode assembly we evaluate the influence of impurity content on battery polarization and rate of side reactions which is indicated by the increase of average oxidation state (AOS) during charge-discharge tests and varies from 0.061 to 0.027 day-1 for electrolytes made from 99.1 and 99.9 wt % V2O5. We found that increase of AOS correlates with the increase of open-circuit voltage of VRFB in the discharged state ranging from 9.6 to 14.9 mV day-1 for highest and lowest electrolyte purity levels, respectively. While AOS increase is significant, it does not solely determine capacity fade. It is demonstrated that the presence of vanadium crossover decreases capacity fade, i. e. levels the contribution of side reactions on capacity drop.

Long sequence Hopfield memory*

Abstract Sequence memory is an essential attribute of natural and artificial intelligence that enables agents to encode, store, and retrieve complex sequences of stimuli and actions. Computational models of sequence memory have been proposed where recurrent Hopfield-like neural networks are trained with temporally asymmetric Hebbian rules. However, these networks suffer from limited sequence capacity (maximal length of the stored sequence) due to interference between the memories. Inspired by recent work on Dense Associative Memories, we expand the sequence capacity of these models by introducing a nonlinear interaction term, enhancing separation between the patterns. We derive novel scaling laws for sequence capacity with respect to network size, significantly outperforming existing scaling laws for models based on traditional Hopfield networks, and verify these theoretical results with numerical simulation. Moreover, we introduce a generalized pseudoinverse rule to recall sequences of highly correlated patterns. Finally, we extend this model to store sequences with variable timing between states’ transitions and describe a biologically-plausible implementation, with connections to motor neuroscience.

Autonomous Battery Optimization by Deploying Distributed Experiments and Simulations

AbstractNon‐trivial relationships link individual materials properties to device‐level performance. Device optimization therefore calls for new automation approaches beyond the laboratory bench with tight integration of different research methods. This study demonstrates a Materials Acceleration Platform (MAP) in the field of battery research based on the problem‐agnostic Fast INtention‐Agnostic LEarning Server (FINALES) framework, which integrates simulations and physical experiments while leaving the active control of the hardware and software resources executing experiments or simulations with the partners running the respective units. This decentralization of control is a distinctive feature of MAPs using the FINALES framework. The connected capabilities entail the formulation and characterization of electrolytes, cell assembly and testing, early lifetime prediction, and ontology‐mapped data storage provided by institutions distributed across Europe. The infrastructure is used to optimize the ionic conductivity of electrolytes and the End Of Life (EOL) of lithium‐ion coin cells by varying the electrolyte formulation. Trends in ionic conductivity are rediscovered and the effect of the electrolyte formulation on the EOL is investigated. Further, the capability of this MAP to bridge diverse research modalities, scales, and institutions enabling system‐level investigations under asynchronous conditions while handling concurrent workflows on the material‐ and system‐level is shown, demonstrating true intention‐agnosticism.

Stress granules formation in HEI-OC1 auditory cells and in H4 human neuroglioma cells secondary to cisplatin exposure

Abstract Stress granules (SGs) are highly dynamic micromolecular membraneless condensates that generate in cells subjected to stress. Formed from pools of untranslating messenger ribonucleoproteins (RNP), SGs dynamics constitute vital processes essential for cell survival. Here, we investigate whether established cytotoxic agents, such as the platinum-based chemotherapeutic agent cisplatin and the aminoglycoside antibiotic gentamicin, elicit SG formation in the House Ear Institute-Organ of Corti-1 (HEI-OC1) auditory cell line, H4 human neuroglioma cells and HEK-293T human embryonic kidney cells. Cells were treated with cisplatin or gentamicin for specific durations at designated concentrations. SG formation was assessed using immunocytochemistry and live cell imaging. Levels of essential proteins involved in SG assembly were evaluated using immunoblotting. We observed cisplatin-associated SG assembly in HEI-OC1 and H4 cells via confocal microscopy through antibody colabeling of G3BP1 with PABP or Caprin1. While maintaining an unchanged pattern of expression of main constituent SG proteins, cisplatin-related SGs in H4 cells persisted for at least 12 h after drug removal. Cells subjected to gentamicin exposure did not exhibit SGs. Our findings offer insights into subcellular mechanisms related to cisplatin-associated cytotoxicity, highlighting the need for future studies to further investigate this stress-response mechanism.

MiR-326 overexpression inhibits colorectal cancer cell growth and proteasome activity by targeting PNO1: unveiling a novel therapeutic intervention strategy

Proteasome inhibition emerges as a promising strategy for cancer prevention. PNO1, pivotal for colorectal cancer (CRC) progression, is involved in proteasome assembly in Saccharomyces cerevisiae. Hence, we aimed to explore the role of PNO1 in proteasome assembly and its up- and down-streams in CRC. Here, we demonstrated that PNO1 knockdown suppressed CRC cells growth, proteasome activities and assembly, as well as CDKN1B/p27Kip1 (p27) degradation. Moreover, p27 knockdown partially attenuated the inhibition of HCT116 cells growth by PNO1 knockdown. The up-stream studies of PNO1 identified miR-326 as a candidate miRNA directly targeting to CDS-region of PNO1 and its overexpression significantly down-regulated PNO1 protein expression, resulting in suppression of cell growth, decrease of proteasome activities and assembly, as well as increasing the stability of p27 in CRC cells. These findings indicated that miR-326 overexpression can suppress CRC cell growth, acting as an endogenous proteasome inhibitor by targeting PNO1.

Artificial potential field-empowered dynamic holographic optical tweezers for particle-array assembly and transformation

Owing to the ability to parallel manipulate micro-objects, dynamic holographic optical tweezers (HOTs) are widely used for assembly and patterning of particles or cells. However, for simultaneous control of large-scale targets, potential collisions could lead to defects in the formed patterns. Herein we introduce the artificial potential field (APF) to develop dynamic HOTs that enable collision-avoidance micro-manipulation. By eliminating collision risks among particles, this method can maximize the degree of parallelism in multi-particle transport, and it permits the implementation of the Hungarian algorithm for matching the particles with their target sites in a minimal pathway. In proof-of-concept experiments, we employ APF-empowered dynamic HOTs to achieve direct assembly of a defect-free 8 × 8 array of microbeads, which starts from random initial positions. We further demonstrate successive flexible transformations of a 7 × 7 microbead array, by regulating its tilt angle and inter-particle spacing distances with a minimalist path. We anticipate that the proposed method will become a versatile tool to open up new possibilities for parallel optical micromanipulation tasks in a variety of fields.

Functional networks of inhibitory neurons orchestrate synchrony in the hippocampus.

Inhibitory interneurons are pivotal components of cortical circuits. Beyond providing inhibition, they have been proposed to coordinate the firing of excitatory neurons within cell assemblies. While the roles of specific interneuron subtypes have been extensively studied, their influence on pyramidal cell synchrony in vivo remains elusive. Employing an all-optical approach in mice, we simultaneously recorded hippocampal interneurons and pyramidal cells and probed the network influence of individual interneurons using optogenetics. We demonstrate that CA1 interneurons form a functionally interconnected network that promotes synchrony through disinhibition during awake immobility, while preserving endogenous cell assemblies. Our network model underscores the importance of both cell assemblies and dense, unspecific interneuron connectivity in explaining our experimental findings, suggesting that interneurons may operate not only via division of labor but also through concerted activity.

Homeostatic regulation of brain activity: from endogenous mechanisms to homeostatic nanomachines.

After the initial concepts of the constancy of the internal milieu or homeostasis, put forward by Claude Bernard and Walter Cannon, homeostasis emerged as a mechanism to control oscillations of biologically meaningful variables within narrow physiological ranges. This is a primary need in the central nervous system that is continuously subjected to a multitude of stimuli from the internal and external environments that affect its function and structure, allowing to adapt the individual to the ever-changing daily conditions. Preserving physiological levels of activity despite disturbances that could either depress neural computation or excessively stimulate neural activity is fundamental, and failure of these homeostatic mechanisms can lead to brain diseases. In this review, we cover the role and main mechanisms of homeostatic plasticity involving the regulation of excitability and synaptic strength from the single neuron to the network level. We analyze the relationships between homeostatic and Hebbian plasticity and the conditions under which the preservation of the excitatory/inhibitory balance fails, triggering epileptogenesis and eventually epilepsy. Several therapeutic strategies to cure epilepsy have been designed to strengthen homeostasis when endogenous homeostatic plasticity mechanisms have become insufficient or ineffective to contrast hyperactivity. We describe "on demand" gene therapy strategies including optogenetics, chemogenetics, and chemo-optogenetics, and particularly focus on new closed loop sensor-actuator strategies mimicking homeostatic plasticity that can be endogenously expressed to strengthen the homeostatic defenses against brain diseases.

Controlling Cell Interactions with DNA Directed Assembly.

The creation of complex cellular environments is critical to mimicking tissue environments that will play a critical role in next-generation tissue engineering, stem cell programming, and therapeutic screening. To address this growing need, techniques capable of manipulating cell-cell and cell-material interactions are required that span single-cell to 3D tissue architectures. DNA programmed assembly and placement of cells present a powerful technique for the bottom-up synthesis of living microtissues for probing key questions in cell-cell and cell-material-driven behaviors through its refined control over placement and architecture. This review examines the current state of the art in the programming of cellular interactions with DNA and its applications spanning tissue model building, fundamental cellular biology, and cell manipulation for measurements across a host of applications.

Prototype Analysis in Hopfield Networks With Hebbian Learning.

We discuss prototype formation in the Hopfield network. Typically, Hebbian learning with highly correlated states leads to degraded memory performance. We show that this type of learning can lead to prototype formation, where unlearned states emerge as representatives of large correlated subsets of states, alleviating capacity woes. This process has similarities to prototype learning in human cognition. We provide a substantial literature review of prototype learning in associative memories, covering contributions from psychology, statistical physics, and computer science. We analyze prototype formation from a theoretical perspective and derive a stability condition for these states based on the number of examples of the prototype presented for learning, the noise in those examples, and the number of nonexample states presented. The stability condition is used to construct a probability of stability for a prototype state as the factors of stability change. We also note similarities to traditional network analysis, allowing us to find a prototype capacity. We corroborate these expectations of prototype formation with experiments using a simple Hopfield network with standard Hebbian learning. We extend our experiments to a Hopfield network trained on data with multiple prototypes and find the network is capable of stabilizing multiple prototypes concurrently. We measure the basins of attraction of the multiple prototype states, finding attractor strength grows with the number of examples and the agreement of examples. We link the stability and dominance of prototype states to the energy profile of these states, particularly when comparing the profile shape to target states or other spurious states.

Spike-based computational primitives to reproduce neural dynamics in the parietal cortex during motor preparation

Abstract Biologically plausible spiking neural network models of sensory cortices can be instrumental in understanding and validating their principles of computation.
 Models based on Cortical Computational Primitives (CCPs), such as Hebbian plasticity and Winner-Take-All (WTA) networks, have already been successful in this approach.
 However, the specific nature and roles of CCPs in sensorimotor cortices during cognitive tasks are yet to be fully deciphered.
 The evolution of motor intention in the Posterior Parietal Cortex (PPC) before arm-reaching movements is a well-suited cognitive process to assess the effectiveness of different CCPs.
 To this end, we propose a biologically plausible model composed of heterogeneous spiking neurons which implements and combines multiple CCPs, such as multi-timescale learning and soft WTA modules.
 By training the model to replicate the dynamics of in-vivo recordings from non-human primates, we show how it is effective in generating meaningful representations from unbalanced input data, and in faithfully reproducing the transition from motor planning to action selection.
 Our findings elucidate the importance of distributing spike-based plasticity across multi-timescales, and provide an explanation for the role of different CCPs in models of frontoparietal cortical networks for performing multisensory integration to efficiently inform action execution.

P-Block Aluminum Single-Atom Catalyst for Electrocatalytic CO2 Reduction with High Intrinsic Activity.

Atomically dispersed transition metal sites on nitrogen-doped carbon catalysts hold great potential for the electrochemical CO2 reduction reaction (CO2RR) to CO due to their encouraging selectivity. However, their intrinsic activity is restricted by the hurdle of the high energy barrier of either *COOH formation or *CO desorption due to the scaling relationship. Herein, we discover a p-block aluminum single-atom catalyst (Al-NC) featuring an Al-N4 site that enables disentangling this hurdle, which endows a moderate reaction kinetic barrier for *COOH formation and *CO desorption, as validated by in situ attenuated total reflection infrared spectroscopy and theoretical simulations. As a result, the developed Al-NC shows a CO Faradaic efficiency (FECO) of up to 98.76% at -0.65 V vs RHE and an intrinsic catalytic turnover frequency of 3.60 s-1 at -0.99 V vs RHE, exceeding those of the state-of-the-art Ni-NC and Fe-NC counterparts. Moreover, it also delivers a partial CO current of 309 mA·cm-2 at 93.65% FECO and 605 mA at >85% FECO in a flow cell and membrane electrode assembly (MEA), respectively. Strikingly, when using low-concentration CO2 (30%) as the feedstock, this catalyst can still deliver a partial CO current of 240 mA at >80% FECO in the MEA. Considering the earth-abundant character of the Al element and the high intrinsic activity of the Al-NC catalyst, it is a promising alternative to today's transition metal-based single-atom catalysts.

Synaptic weight dynamics underlying memory consolidation: Implications for learning rules, circuit organization, and circuit function

Systems consolidation is a common feature of learning and memory systems, in which a long-term memory initially stored in one brain region becomes persistently stored in another region. We studied the dynamics of systems consolidation in simple circuit architectures with two sites of plasticity, one in an early-learning and one in a late-learning brain area. We show that the synaptic dynamics of the circuit during consolidation of an analog memory can be understood as a temporal integration process, by which transient changes in activity driven by plasticity in the early-learning area are accumulated into persistent synaptic changes at the late-learning site. This simple principle naturally leads to a speed-accuracy tradeoff in systems consolidation and provides insight into how the circuit mitigates the stability-plasticity dilemma of storing new memories while preserving core features of older ones. Furthermore, it imposes two constraints on the circuit. First, the plasticity rule at the late-learning site must stably support a continuum of possible outputs for a given input. We show that this is readily achieved by heterosynaptic but not standard Hebbian rules. Second, to turn off the consolidation process and prevent erroneous changes at the late-learning site, neural activity in the early-learning area must be reset to its baseline activity. We provide two biologically plausible implementations for this reset that propose functional roles in stabilizing consolidation for core elements of the cerebellar circuit.

Cortical development in the structural model and free energy minimization.

A model of neocortical development invoking Friston's Free Energy Principle is applied within the Structural Model of Barbas etal. and the associated functional interpretation advanced by Tucker and Luu. Evolution of a neural field with Hebbian and anti-Hebbian plasticity, maximizing synchrony and minimizing axonal length by apoptotic selection, leads to paired connection systems with mirror symmetry, interacting via Markov blankets along their line of reflection. Applied to development along the radial lines of development in the Structural Model, a primary Markov blanket emerges between the centrifugal synaptic flux in layers 2,3 and 5,6, versus the centripetal flow in layer 4, and axonal orientations in layer 4 give rise to the differing shape and movement sensitivities characteristic of neurons of dorsal and ventral neocortex. Prediction error minimization along the primary blanket integrates limbic and subcortical networks with the neocortex. Synaptic flux bypassing the blanket triggers the arousal response to surprising stimuli, enabling subsequent adaptation. As development progresses ubiquitous mirror systems separated by Markov blankets and enclosed blankets-within-blankets arise throughout neocortex, creating the typical order and response characteristics of columnar and noncolumnar cortex.