arXiv subjects

Roni Vardi

Publications and source records attributed to Roni Vardi.

At least 19 recordsLinked to original sources

arXiv2023

The mechanism underlying successful deep learning

Deep architectures consist of tens or hundreds of convolutional layers (CLs) that terminate with a few fully connected (FC) layers and an output layer representing the possible labels of a complex classification task. According to the existing deep learning (DL) rationale, the first CL reveals localized features from the raw data, whereas the subsequent layers progressively extract higher-level features required for refined classification. This article presents an efficient three-phase procedure for quantifying the mechanism underlying successful DL. First, a deep architecture is trained to maximize the success rate (SR). Next, the weights of the first several CLs are fixed and only the concatenated new FC layer connected to the output is trained, resulting in SRs that progress with the layers. Finally, the trained FC weights are silenced, except for those emerging from a single filter, enabling the quantification of the functionality of this filter using a correlation matrix between input labels and averaged output fields, hence a well-defined set of quantifiable features is obtained. Each filter essentially selects a single output label independent of the input label, which seems to prevent high SRs; however, it counterintuitively identifies a small subset of possible output labels. This feature is an essential part of the underlying DL mechanism and is progressively sharpened with layers, resulting in enhanced signal-to-noise ratios and SRs. Quantitatively, this mechanism is exemplified by the VGG-16, VGG-6, and AVGG-16. The proposed mechanism underlying DL provides an accurate tool for identifying each filter's quality and is expected to direct additional procedures to improve the SR, computational complexity, and latency of DL.

cs.CV↗

arXiv2023

Enhancing the accuracies by performing pooling decisions adjacent to the output layer

Learning classification tasks of (2^nx2^n) inputs typically consist of \le n (2x2) max-pooling (MP) operators along the entire feedforward deep architecture. Here we show, using the CIFAR-10 database, that pooling decisions adjacent to the last convolutional layer significantly enhance accuracies. In particular, average accuracies of the advanced-VGG with m layers (A-VGGm) architectures are 0.936, 0.940, 0.954, 0.955, and 0.955 for m=6, 8, 14, 13, and 16, respectively. The results indicate A-VGG8s' accuracy is superior to VGG16s', and that the accuracies of A-VGG13 and A-VGG16 are equal, and comparable to that of Wide-ResNet16. In addition, replacing the three fully connected (FC) layers with one FC layer, A-VGG6 and A-VGG14, or with several linear activation FC layers, yielded similar accuracies. These significantly enhanced accuracies stem from training the most influential input-output routes, in comparison to the inferior routes selected following multiple MP decisions along the deep architecture. In addition, accuracies are sensitive to the order of the non-commutative MP and average pooling operators adjacent to the output layer, varying the number and location of training routes. The results call for the reexamination of previously proposed deep architectures and their accuracies by utilizing the proposed pooling strategy adjacent to the output layer.

cs.CV↗

arXiv2022

Brain inspired neuronal silencing mechanism to enable reliable sequence identification

Real-time sequence identification is a core use-case of artificial neural networks (ANNs), ranging from recognizing temporal events to identifying verification codes. Existing methods apply recurrent neural networks, which suffer from training difficulties; however, performing this function without feedback loops remains a challenge. Here, we present an experimental neuronal long-term plasticity mechanism for high-precision feedforward sequence identification networks (ID-nets) without feedback loops, wherein input objects have a given order and timing. This mechanism temporarily silences neurons following their recent spiking activity. Therefore, transitory objects act on different dynamically created feedforward sub-networks. ID-nets are demonstrated to reliably identify 10 handwritten digit sequences, and are generalized to deep convolutional ANNs with continuous activation nodes trained on image sequences. Counterintuitively, their classification performance, even with a limited number of training examples, is high for sequences but low for individual objects. ID-nets are also implemented for writer-dependent recognition, and suggested as a cryptographic tool for encrypted authentication. The presented mechanism opens new horizons for advanced ANN algorithms.

physics.bio-ph↗

arXiv2022

Long anisotropic absolute refractory periods with rapid rise-times to reliable responsiveness

Refractoriness is a fundamental property of excitable elements, such as neurons, indicating the probability for re-excitation in a given time-lag, and is typically linked to the neuronal hyperpolarization following an evoked spike. Here we measured the refractory periods (RPs) in neuronal cultures and observed that an average anisotropic absolute RP could exceed 10 milliseconds and its tail 20 milliseconds, independent of a large stimulation frequency range. It is an order of magnitude longer than anticipated and comparable with the decaying membrane potential timescale. It is followed by a sharp rise-time (relative RP) of merely ~1 millisecond to complete responsiveness. Extracellular stimulations result in longer absolute RPs than solely intracellular ones, and a pair of extracellular stimulations from two different routes exhibits distinct absolute RPs, depending on their order. Our results indicate that a neuron is an accurate excitable element, where the diverse RPs cannot be attributed solely to the soma and imply fast mutual interactions between different stimulation routes and dendrites. Further elucidation of neuronal computational capabilities and their interplay with adaptation mechanisms is warranted.

q-bio.NC↗

arXiv2021

Significant anisotropic neuronal refractory period plasticity

Refractory periods are an unavoidable feature of excitable elements, resulting in necessary time-lags for re-excitation. Herein, we measure neuronal absolute refractory periods (ARPs) in synaptic blocked neuronal cultures. In so doing, we show that their duration can be significantly extended by dozens of milliseconds using preceding evoked spikes generated by extracellular stimulations. The ARP increases with the frequency of preceding stimulations, and saturates at the intermittent phase of the neuronal response latency, where a short relative refractory period might appear. Nevertheless, preceding stimulations via a different extracellular route does not affect the ARP. It is also found to be independent of preceding intracellular stimulations. All these features strongly suggest that the anisotropic ARPs originate in neuronal dendrites. The results demonstrate the fast and significant plasticity of the neuronal ARP, depending on the firing activity of its connecting neurons, which is expected to affect network dynamics.

q-bio.NC↗

arXiv2020

Brain experiments imply adaptation mechanisms which outperform common AI learning algorithms

Attempting to imitate the brain functionalities, researchers have bridged between neuroscience and artificial intelligence for decades; however, experimental neuroscience has not directly advanced the field of machine learning. Here, using neuronal cultures, we demonstrate that increased training frequency accelerates the neuronal adaptation processes. This mechanism was implemented on artificial neural networks, where a local learning step-size increases for coherent consecutive learning steps and tested on a simple dataset of handwritten digits, MNIST. Based on our online learning results with a few handwriting examples, success rates for brain-inspired algorithms substantially outperform the commonly used machine learning algorithms. We speculate this emerging bridge from slow brain function to machine learning will promote ultrafast decision making under limited examples, which is the reality in many aspects of human activity, robotic control, and network optimization.

q-bio.NC↗

arXiv2017

Vitality of Neural Networks under Reoccurring Catastrophic Failures

Catastrophic failures are complete and sudden collapses in the activity of large networks such as economics, electrical power grids and computer networks, which typically require a manual recovery process. Here we experimentally show that excitatory neural networks are governed by a non-Poissonian reoccurrence of catastrophic failures, where their repetition time follows a multimodal distribution characterized by a few tenths of a second and tens of seconds timescales. The mechanism underlying the termination and reappearance of network activity is quantitatively shown here to be associated with nodal time-dependent features, neuronal plasticity, where hyperactive nodes damage the response capability of their neighbors. It presents a complementary mechanism for the emergence of Poissonian catastrophic failures from damage conductivity. The effect that hyperactive nodes degenerate their neighbors represents a type of local competition which is a common feature in the dynamics of real-world complex networks, whereas their spontaneous recoveries represent a vitality which enhances reliable functionality.

q-bio.NC↗

arXiv2017

Simultaneous multi-patch-clamp and extracellular-array recordings: Single neuron reflects network activity

The increasing number of recording electrodes enhances the capability of capturing the network's cooperative activity, however, using too many monitors might alter the properties of the measured neural network and induce noise. Using a technique that merges simultaneous multi-patch-clamp and multi-electrode array recordings of neural networks in-vitro, we show that the membrane potential of a single neuron is a reliable and super-sensitive probe for monitoring such cooperative activities and their detailed rhythms. Specifically, the membrane potential and the spiking activity of a single neuron are either highly correlated or highly anti-correlated with the time-dependent macroscopic activity of the entire network. This surprising observation also sheds light on the cooperative origin of neuronal burst in cultured networks. Our findings present an alternative flexible approach to the technique based on a massive tiling of networks by large-scale arrays of electrodes to monitor their activity.

q-bio.NC↗

arXiv2017

Fast Reversible Learning based on Neurons functioning as Anisotropic Multiplex Hubs

Neural networks are composed of neurons and synapses, which are responsible for learning in a slow adaptive dynamical process. Here we experimentally show that neurons act like independent anisotropic multiplex hubs, which relay and mute incoming signals following their input directions. Theoretically, the observed information routing enriches the computational capabilities of neurons by allowing, for instance, equalization among different information routes in the network, as well as high-frequency transmission of complex time-dependent signals constructed via several parallel routes. In addition, this kind of hubs adaptively eliminate very noisy neurons from the dynamics of the network, preventing masking of information transmission. The timescales for these features are several seconds at most, as opposed to the imprint of information by the synaptic plasticity, a process which exceeds minutes. Results open the horizon to the understanding of fast and adaptive learning realities in higher cognitive functionalities of the brain.

q-bio.NC↗

arXiv2016

Mimicking Collective Firing Patterns of Hundreds of Connected Neurons using a Single-Neuron Experiment

The experimental study of neural networks requires simultaneous measurements of a massive number of neurons, while monitoring properties of the connectivity, synaptic strengths and delays. Current technological barriers make such a mission unachievable. In addition, as a result of the enormous number of required measurements, the estimated network parameters would differ from the original ones. Here we present a versatile experimental technique, which enables the study of recurrent neural networks activity while being capable of dictating the network connectivity and synaptic strengths. This method is based on the observation that the response of neurons depends solely on their recent stimulations, a short-term memory. It allows a long-term scheme of stimulation and recording of a single neuron, to mimic simultaneous activity measurements of neurons in a recurrent network. Utilization of this technique demonstrates the spontaneous emergence of cooperative synchronous oscillations, in particular the coexistence of fast Gamma and slow Delta oscillations, and opens the horizon for the experimental study of other cooperative phenomena within large-scale neural networks.

q-bio.NC↗

arXiv2015

Broadband Macroscopic Cortical Oscillations Emerge from Intrinsic Neuronal Response Failures

Broadband spontaneous macroscopic neural oscillations are rhythmic cortical firing which were extensively examined during the last century, however, their possible origination is still controversial. In this work we show how macroscopic oscillations emerge in solely excitatory random networks and without topological constraints. We experimentally and theoretically show that these oscillations stem from the counterintuitive underlying mechanism - the intrinsic stochastic neuronal response failures. These neuronal response failures, which are characterized by short-term memory, lead to cooperation among neurons, resulting in sub- or several- Hertz macroscopic oscillations which coexist with high frequency gamma oscillations. A quantitative interplay between the statistical network properties and the emerging oscillations is supported by simulations of large networks based on single-neuron in-vitro experiments and a Langevin equation describing the network dynamics. Results call for the examination of these oscillations in the presence of inhibition and external drives.

q-bio.NC↗

arXiv2015

Neuronal Response Impedance Mechanism Implementing Cooperative Networks with Low Firing Rates and Microseconds Precision

Realizations of low firing rates in neural networks usually require globally balanced distributions among excitatory and inhibitory links, while feasibility of temporal coding is limited by neuronal millisecond precision. We show that cooperation, governing global network features, emerges through nodal properties, as opposed to link distributions. Using in vitro and in vivo experiments we demonstrate microsecond precision of neuronal response timings under low stimulation frequencies, whereas moderate frequencies result in a chaotic neuronal phase characterized by degraded precision. Above a critical stimulation frequency, which varies among neurons, response failures were found to emerge stochastically such that the neuron functions as a low pass filter, saturating the average inter-spike-interval. This intrinsic neuronal response impedance mechanism leads to cooperation on a network level, such that firing rates are suppressed towards the lowest neuronal critical frequency simultaneously with neuronal microsecond precision. Our findings open up opportunities of controlling global features of network dynamics through few nodes with extreme properties.

q-bio.NC↗

arXiv2014

Chaotic and non-chaotic phases in experimental responses of a single neuron

Consistency and predictability of brain functionalities depend on reproducible activity of a single neuron. We identify a reproducible non-chaotic neuronal phase where deviations between concave response latency profiles of a single neuron do not increase with the number of stimulations. A chaotic neuronal phase emerges at a transition to convex latency profiles which diverge exponentially, indicating irreproducible response timings. Our findings are supported by a quantitative mathematical framework and found robust to periodic and random stimulation patterns. In addition, these results put a bound on the neuronal temporal resolution which can be enhanced below a millisecond using neuronal chains.

q-bio.NC↗

arXiv2014

Computational paradigm for dynamic logic-gates in neuronal activity

In 1943 McCulloch and Pitts suggested that the brain is composed of reliable logic-gates similar to the logic at the core of today's computers. This framework had a limited impact on neuroscience, since neurons exhibit far richer dynamics. Here we propose a new experimentally corroborated paradigm in which the truth tables of the brain's logic-gates are time dependent, i.e. dynamic logicgates (DLGs). The truth tables of the DLGs depend on the history of their activity and the stimulation frequencies of their input neurons. Our experimental results are based on a procedure where conditioned stimulations were enforced on circuits of neurons embedded within a large-scale network of cortical cells in-vitro. We demonstrate that the underlying biological mechanism is the unavoidable increase of neuronal response latencies to ongoing stimulations, which imposes a nonuniform gradual stretching of network delays. The limited experimental results are confirmed and extended by simulations and theoretical arguments based on identical neurons with a fixed increase of the neuronal response latency per evoked spike. We anticipate our results to lead to better understanding of the suitability of this computational paradigm to account for the brain's functionalities and will require the development of new systematic mathematical methods beyond the methods developed for traditional Boolean algebra.

q-bio.QM↗

arXiv2014

Error correction and fast detectors implemented by ultra-fast neuronal plasticity

We experimentally show that the neuron functions as a precise time-integrator, where the accumulated changes in neuronal response latencies, under complex and random stimulation patterns, are solely a function of a global quantity, the average time-lag between stimulations. In contrast, momentary leaps in the neuronal response latency follow trends of consecutive stimulations, indicating ultra-fast neuronal plasticity. On a circuit level, this ultra-fast neuronal plasticity phenomenon implements error-correction mechanisms and fast detectors for misplaced stimulations. Additionally, at moderate/high stimulation rates this phenomenon destabilizes/stabilizes a periodic neuronal activity disrupted by misplaced stimulations.

q-bio.NC↗

arXiv2013

Sudden synchrony leaps accompanied by frequency multiplications in neuronal activity

A classical view of neural coding relies on temporal firing synchrony among functional groups of neurons; however the underlying mechanism remains an enigma. Here we experimentally demonstrate a mechanism where time-lags among neuronal spiking leap from several tens of milliseconds to nearly zero-lag synchrony. It also allows sudden leaps out of synchrony, hence forming short epochs of synchrony. Our results are based on an experimental procedure where conditioned stimulations were enforced on circuits of neurons embedded within a large-scale network of cortical cells in vitro and are corroborated by simulations of neuronal populations. The underlying biological mechanisms are the unavoidable increase of the neuronal response latency to ongoing stimulations and temporal or spatial summation required to generate evoked spikes. These sudden leaps in and out of synchrony may be accompanied by multiplications of the neuronal firing frequency, hence offering reliable information-bearing indicators which may bridge between the two principal neuronal coding paradigms.

q-bio.NC↗

arXiv2013

An experimental evidence-based computational paradigm for new logic-gates in neuronal activity

We propose a new experimentally corroborated paradigm in which the functionality of the brain's logic-gates depends on the history of their activity, e.g. an OR-gate that turns into a XOR-gate over time. Our results are based on an experimental procedure where conditioned stimulations were enforced on circuits of neurons embedded within a large-scale network of cortical cells in-vitro. The underlying biological mechanism is the unavoidable increase of neuronal response latency to ongoing stimulations, which imposes a non-uniform gradual stretching of network delays.

q-bio.NC↗

arXiv2012

Synchronization with mismatched synaptic delays: A unique role of elastic neuronal latency

We show that the unavoidable increase in neuronal response latency to ongoing stimulation serves as a nonuniform gradual stretching of neuronal circuit delay loops and emerges as an essential mechanism in the formation of various types of neuronal timers. Synchronization emerges as a transient phenomenon without predefined precise matched synaptic delays. These findings are described in an experimental procedure where conditioned stimulations were enforced on a circuit of neurons embedded within a large-scale network of cortical cells in-vitro, and are corroborated by neuronal simulations. They evidence a new cortical timescale based on tens of microseconds stretching of neuronal circuit delay loops per spike, and with realistic delays of a few milliseconds, synchronization emerges for a finite fraction of neuronal circuit delays.

q-bio.NC↗