Target spike patterns enable efficient and biologically plausible learning for complex temporal tasks

Recurrent spiking neural networks (RSNN) in the brain learn to perform a wide range of perceptual, cognitive and motor tasks very efficiently in terms of energy consumption and their training requires very few examples. This motivates the search for biologically inspired learning rules for RSNNs, aiming to improve our understanding of brain computation and the efficiency of artificial intelligence. Several spiking models and learning rules have been proposed, but it remains a challenge to design RSNNs whose learning relies on biologically plausible mechanisms and are capable of solving complex temporal tasks. In this paper, we derive a learning rule, local to the synapse, from a simple mathematical principle, the maximization of the likelihood for the network to solve a specific task. We propose a novel target-based learning scheme in which the learning rule derived from likelihood maximization is used to mimic a specific spatio-temporal spike pattern that encodes the solution to complex temporal tasks. This method makes the learning extremely rapid and precise, outperforming state of the art algorithms for RSNNs. While error-based approaches, (e.g. e-prop) trial after trial optimize the internal sequence of spikes in order to progressively minimize the MSE we assume that a signal randomly projected from an external origin (e.g. from other brain areas) directly defines the target sequence. This facilitates the learning procedure since the network is trained from the beginning to reproduce the desired internal sequence. We propose two versions of our learning rule: spike-dependent and voltage-dependent. We find that the latter provides remarkable benefits in terms of learning speed and robustness to noise. We demonstrate the capacity of our model to tackle several problems like learning multidimensional trajectories and solving the classical temporal XOR benchmark. Finally, we show that an online approximation of the gradient ascent, in addition to guaranteeing complete locality in time and space, allows learning after very few presentations of the target output. Our model can be applied to different types of biological neurons. The analytically derived plasticity learning rule is specific to each neuron model and can produce a theoretical prediction for experimental validation.

Download Full-text

Kinematic Analysis of Lower Limb Joint Asymmetry during Gait in People with Multiple Sclerosis

Symmetry ◽

10.3390/sym13040598 ◽

2021 ◽

Vol 13 (4) ◽

pp. 598

Author(s):

Massimiliano Pau ◽

Bruno Leban ◽

Michela Deidda ◽

Federica Putzolu ◽

Micaela Porta ◽

...

Keyword(s):

Multiple Sclerosis ◽

Lower Limb ◽

Sagittal Plane ◽

Cross Sectional Study ◽

Camera Motion ◽

Cross Sectional ◽

Temporal Parameters ◽

Wide Range ◽

Edss Score ◽

Spatio Temporal

The majority of people with Multiple Sclerosis (pwMS), report lower limb motor dysfunctions, which may relevantly affect postural control, gait and a wide range of activities of daily living. While it is quite common to observe a different impact of the disease on the two limbs (i.e., one of them is more affected), less clear are the effects of such asymmetry on gait performance. The present retrospective cross-sectional study aimed to characterize the magnitude of interlimb asymmetry in pwMS, particularly as regards the joint kinematics, using parameters derived from angle-angle diagrams. To this end, we analyzed gait patterns of 101 pwMS (55 women, 46 men, mean age 46.3, average Expanded Disability Status Scale (EDSS) score 3.5, range 1–6.5) and 81 unaffected individuals age- and sex-matched who underwent 3D computerized gait analysis carried out using an eight-camera motion capture system. Spatio-temporal parameters and kinematics in the sagittal plane at hip, knee and ankle joints were considered for the analysis. The angular trends of left and right sides were processed to build synchronized angle–angle diagrams (cyclograms) for each joint, and symmetry was assessed by computing several geometrical features such as area, orientation and Trend Symmetry. Based on cyclogram orientation and Trend Symmetry, the results show that pwMS exhibit significantly greater asymmetry in all three joints with respect to unaffected individuals. In particular, orientation values were as follows: 5.1 of pwMS vs. 1.6 of unaffected individuals at hip joint, 7.0 vs. 1.5 at knee and 6.4 vs. 3.0 at ankle (p < 0.001 in all cases), while for Trend Symmetry we obtained at hip 1.7 of pwMS vs. 0.3 of unaffected individuals, 4.2 vs. 0.5 at knee and 8.5 vs. 1.5 at ankle (p < 0.001 in all cases). Moreover, the same parameters were sensitive enough to discriminate individuals of different disability levels. With few exceptions, all the calculated symmetry parameters were found significantly correlated with the main spatio-temporal parameters of gait and the EDSS score. In particular, large correlations were detected between Trend Symmetry and gait speed (with rho values in the range of –0.58 to –0.63 depending on the considered joint, p < 0.001) and between Trend Symmetry and EDSS score (rho = 0.62 to 0.69, p < 0.001). Such results suggest not only that MS is associated with significantly marked interlimb asymmetry during gait but also that such asymmetry worsens as the disease progresses and that it has a relevant impact on gait performances.

Download Full-text

Effects of memristive synapse radiation interactions on learning in spiking neural networks

SN Applied Sciences ◽

10.1007/s42452-021-04553-0 ◽

2021 ◽

Vol 3 (5) ◽

Author(s):

Sumedha Gandharava Dahl ◽

Robert C. Ivans ◽

Kurtis D. Cantley

Keyword(s):

Action Potentials ◽

Radiation Effects ◽

Synaptic Weight ◽

Learning Rule ◽

Spike Timing ◽

Spiking Neural Network ◽

Synaptic Action ◽

Drift Model ◽

Spatio Temporal ◽

Post Synaptic Neuron

AbstractThis study uses advanced modeling and simulation to explore the effects of external events such as radiation interactions on the synaptic devices in an electronic spiking neural network. Specifically, the networks are trained using the spike-timing-dependent plasticity (STDP) learning rule to recognize spatio-temporal patterns (STPs) representing 25 and 100-pixel characters. Memristive synapses based on a TiO2 non-linear drift model designed in Verilog-A are utilized, with STDP learning behavior achieved through bi-phasic pre- and post-synaptic action potentials. The models are modified to include experimentally observed state-altering and ionizing radiation effects on the device. It is found that radiation interactions tend to make the connection between afferents stronger by increasing the conductance of synapses overall, subsequently distorting the STDP learning curve. In the absence of consistent STPs, these effects accumulate over time and make the synaptic weight evolutions unstable. With STPs at lower flux intensities, the network can recover and relearn with constant training. However, higher flux can overwhelm the leaky integrate-and-fire post-synaptic neuron circuits and reduce stability of the network.

Download Full-text

Forest path condition monitoring based on crowd-based trajectory data analysis

Journal of Ambient Intelligence and Smart Environments ◽

10.3233/ais-200586 ◽

2020 ◽

pp. 1-18

Author(s):

Francisco Arcas-Tunez ◽

Fernando Terroso-Saenz

Keyword(s):

Information Acquisition ◽

Road Network ◽

Road Networks ◽

Trajectory Data ◽

Country Road ◽

Wide Range ◽

Sport Activities ◽

Path Condition ◽

Spatio Temporal ◽

The City

The development of Road Information Acquisition Systems (RIASs) based on the Mobile Crowdsensing (MCS) paradigm has been widely studied for the last years. In that sense, most of the existing MCS-based RIASs focus on urban road networks and assume a car-based scenario. However, there exist a scarcity of approaches that pay attention to rural and country road networks. In that sense, forest paths are used for a wide range of recreational and sport activities by many different people and they can be also affected by different problems or obstacles blocking them. As a result, this work introduces SAMARITAN, a framework for rural-road network monitoring based on MCS. SAMARITAN analyzes the spatio-temporal trajectories from cyclists extracted from the fitness application Strava so as to uncover potential obstacles in a target road network. The framework has been evaluated in a real-world network of forest paths in the city of Cieza (Spain) showing quite promising results.

Download Full-text

Predicting Human Mobility with Reinforcement-Learning-Based Long-Term Periodicity Modeling

ACM Transactions on Intelligent Systems and Technology ◽

10.1145/3469860 ◽

2021 ◽

Vol 12 (6) ◽

pp. 1-23

Author(s):

Shuo Tao ◽

Jingang Jiang ◽

Defu Lian ◽

Kai Zheng ◽

Enhong Chen

Keyword(s):

Reinforcement Learning ◽

Human Mobility ◽

Recurrent Network ◽

Mobility Prediction ◽

Learning Framework ◽

Temporal Features ◽

Wide Range ◽

Spatio Temporal ◽

Historical Trajectory

Mobility prediction plays an important role in a wide range of location-based applications and services. However, there are three problems in the existing literature: (1) explicit high-order interactions of spatio-temporal features are not systemically modeled; (2) most existing algorithms place attention mechanisms on top of recurrent network, so they can not allow for full parallelism and are inferior to self-attention for capturing long-range dependence; (3) most literature does not make good use of long-term historical information and do not effectively model the long-term periodicity of users. To this end, we propose MoveNet and RLMoveNet. MoveNet is a self-attention-based sequential model, predicting each user’s next destination based on her most recent visits and historical trajectory. MoveNet first introduces a cross-based learning framework for modeling feature interactions. With self-attention on both the most recent visits and historical trajectory, MoveNet can use an attention mechanism to capture the user’s long-term regularity in a more efficient way. Based on MoveNet, to model long-term periodicity more effectively, we add the reinforcement learning layer and named RLMoveNet. RLMoveNet regards the human mobility prediction as a reinforcement learning problem, using the reinforcement learning layer as the regularization part to drive the model to pay attention to the behavior with periodic actions, which can help us make the algorithm more effective. We evaluate both of them with three real-world mobility datasets. MoveNet outperforms the state-of-the-art mobility predictor by around 10% in terms of accuracy, and simultaneously achieves faster convergence and over 4x training speedup. Moreover, RLMoveNet achieves higher prediction accuracy than MoveNet, which proves that modeling periodicity explicitly from the perspective of reinforcement learning is more effective.

Download Full-text

Abstract 189: Activation of Voltage-Dependent Anion Channel 2 Suppresses Ca 2+ -Induced Cardiac Arrhythmia

Circulation Research ◽

10.1161/res.111.suppl_1.a189 ◽

2012 ◽

Vol 111 (suppl_1) ◽

Author(s):

Jaunian Chen ◽

Johann Schredelseker ◽

Hirohito Shimizu ◽

Jie Huang ◽

Kui Lu ◽

...

Keyword(s):

Outer Membrane ◽

Cardiac Arrhythmia ◽

Critical Role ◽

Anion Channel ◽

Cardiac Contraction ◽

Voltage Dependent Anion Channel ◽

Rescue Effect ◽

Cardiac Fibrillation ◽

Wide Range ◽

Voltage Dependent

Abnormal Ca2+ handling in cardiac muscle cells is associated with a wide range of human cardiac diseases, including heart failure and cardiac arrhythmias. Zebrafish tremblor (tre) mutant embryos manifest unsynchronized cardiac contractions due to a Ca2+ extrusion defect in cardiomyocytes and thus are used as an animal model for aberrant Ca2+ homeostasis-induced cardiac arrhythmia. To further dissect molecular mechanisms regulating cardiac Ca2+ homeostasis, we conducted a chemical suppressor screen on tre and found that efsevin, a synthetic compound, potently suppresses cardiac fibrillation and restores rhythmic cardiac contractions in tre embryos. In addition, the treatment with efsevin blocks the propagation of arrhythmogenic Ca2+ waves and accelerates the decay phase of Ca2+ sparks in adult murine cardiomyocytes under Ca2+ overload conditions, demonstrating that efsevin modulates Ca2+ handling in both embryonic and adult cardiac tissues. Through a biochemical pulldown assay, we identified a direct interaction between efsevin and VDAC2, a mitochondrial outer membrane voltage dependent anion channel. Overexpression of VDAC2 restores synchronized cardiac contraction in tre and knocking down VDAC2 activity abolishes the rescue effect of efsevin on tre, suggesting that efsevin modulates cardiac Ca2+ homeostasis by potentiating VDAC2 activity. We further showed that enhancing mitochondria Ca2+ uptake by overexpressing MICU or MCU suppresses cardiac fibrillation in tre just like VDAC2 does. Interestingly, this suppressive effect is absent in tre/vdac2 double deficient embryos and co-expression of VDAC2 and MICU or MCU results in synergistic rescue effect on tre, indicating a critical role for mitochondria in regulating cardiac Ca2+ handling and rhythmicity and suggesting that VDAC2 functions as a gate for transporting Ca2+ across the outer membrane. Taken together, our findings identify efsevin as a potent pharmacological tool to modulate cardiac Ca2+ handling, suggest a critical role of mitochondria in the control of cardiac rhythmicity and establish VDAC2 as a modulator of cardiac Ca2+ handling and a potential therapeutic target for the treatment of arrhythmias.

Download Full-text

Voltage-gated transient currents in bovine adrenal fasciculata cells. I. T-type Ca2+ current.

The Journal of General Physiology ◽

10.1085/jgp.102.2.217 ◽

1993 ◽

Vol 102 (2) ◽

pp. 217-237 ◽

Cited By ~ 30

Author(s):

B Mlinar ◽

B A Biagi ◽

J J Enyeart

Keyword(s):

Time Constant ◽

Low Voltage ◽

Zona Fasciculata ◽

Patch Clamp Technique ◽

Time Constants ◽

Voltage Gated ◽

Wide Range ◽

Boltzmann Function ◽

Voltage Dependent ◽

Ca2 Channels

The whole cell version of the patch clamp technique was used to identify and characterize voltage-gated Ca2+ channels in enzymatically dissociated bovine adrenal zona fasciculata (AZF) cells. The great majority of cells (84 of 86) expressed only low voltage-activated, rapidly inactivating Ca2+ current with properties of T-type Ca2+ current described in other cells. Voltage-dependent activation of this current was fit by a Boltzmann function raised to an integer power of 4 with a midpoint at -17 mV. Independent estimates of the single channel gating charge obtained from the activation curve and using the "limiting logarithmic potential sensitivity" were 8.1 and 6.8 elementary charges, respectively. Inactivation was a steep function of voltage with a v1/2 of -49.9 mV and a slope factor K of 3.73 mV. The expression of a single Ca2+ channel subtype by AZF cells allowed the voltage-dependent gating and kinetic properties of T current to be studied over a wide range of potentials. Analysis of the gating kinetics of this Ca2+ current indicate that T channel activation, inactivation, deactivation (closing), and reactivation (recovery from inactivation) each include voltage-independent transitions that become rate limiting at extreme voltages. Ca2+ current activated with voltage-dependent sigmoidal kinetics that were described by an m4 model. The activation time constant varied exponentially at test potentials between -30 and +10 mV, approaching a voltage-independent minimum of 1.6 ms. The inactivation time constant (tau i) also decreased exponentially to a minimum of 18.3 ms at potentials positive to 0 mV. T channel closing (deactivation) was faster at more negative voltages; the deactivation time constant (tau d) decreased from 8.14 +/- 0.7 to 0.48 +/- 0.1 ms at potentials between -40 and -150 mV. T channels inactivated by depolarization returned to the closed state along pathways that included two voltage-dependent time constants. tau rec-s ranged from 8.11 to 4.80 s when the recovery potential was varied from -50 to -90 mV, while tau rec-f decreased from 1.01 to 0.372 s. At potentials negative to -70 mV, both time constants approached minimum values. The low voltage-activated Ca2+ current in AZF cells was blocked by the T channel selective antagonist Ni2+ with an IC50 of 20 microM. At similar concentrations, Ni2+ also blocked cortisol secretion stimulated by adrenocorticotropic hormone. Our results indicate that bovine AZF cells are distinctive among secretory cells in expressing primarily or exclusively T-type Ca2+ channels.(ABSTRACT TRUNCATED AT 400 WORDS)

Download Full-text

Models of Respiratory Rhythm Generation in the Pre-Bötzinger Complex. I. Bursting Pacemaker Neurons

Journal of Neurophysiology ◽

10.1152/jn.1999.82.1.382 ◽

1999 ◽

Vol 82 (1) ◽

pp. 382-397 ◽

Cited By ~ 316

Author(s):

Robert J. Butera ◽

John Rinzel ◽

Jeffrey C. Smith

Keyword(s):

Membrane Potential ◽

Respiratory Rhythm ◽

Rhythm Generation ◽

Minimal Models ◽

Dynamic Features ◽

Bursting Neurons ◽

Wide Range ◽

Bötzinger Complex ◽

Voltage Dependent ◽

Respiratory Rhythm Generation

A network of oscillatory bursting neurons with excitatory coupling is hypothesized to define the primary kernel for respiratory rhythm generation in the pre-Bötzinger complex (pre-BötC) in mammals. Two minimal models of these neurons are proposed. In model 1, bursting arises via fast activation and slow inactivation of a persistent Na+ current I NaP-h. In model 2, bursting arises via a fast-activating persistent Na+ current INaP and slow activation of a K+ current IKS. In both models, action potentials are generated via fast Na+ and K+currents. The two models have few differences in parameters to facilitate a rigorous comparison of the two different burst-generating mechanisms. Both models are consistent with many of the dynamic features of electrophysiological recordings from pre-BötC oscillatory bursting neurons in vitro, including voltage-dependent activity modes (silence, bursting, and beating), a voltage-dependent burst frequency that can vary from 0.05 to >1 Hz, and a decaying spike frequency during bursting. These results are robust and persist across a wide range of parameter values for both models. However, the dynamics of model 1 are more consistent with experimental data in that the burst duration decreases as the baseline membrane potential is depolarized and the model has a relatively flat membrane potential trajectory during the interburst interval. We propose several experimental tests to demonstrate the validity of either model and to differentiate between the two mechanisms.

Download Full-text

The Repeat-In-Toxin Family Member TosA Mediates Adherence of Uropathogenic Escherichia coli and Survival during Bacteremia

Infection and Immunity ◽

10.1128/iai.05713-11 ◽

2011 ◽

Vol 80 (2) ◽

pp. 493-505 ◽

Cited By ~ 31

Author(s):

Patrick D. Vigil ◽

Travis J. Wiles ◽

Michael D. Engstrom ◽

Lev Prasov ◽

Matthew A. Mulvey ◽

...

Keyword(s):

Escherichia Coli ◽

Urinary Tract ◽

Virulence Factors ◽

Upper Urinary Tract ◽

Host Cells ◽

Content Type ◽

E Coli ◽

Wide Range ◽

Novel Target ◽

Tract Infections

ABSTRACTUropathogenicEscherichia coli(UPEC) is responsible for the majority of uncomplicated urinary tract infections (UTI) and represents the most common bacterial infection in adults. UPEC utilizes a wide range of virulence factors to colonize the host, including the novel repeat-in-toxin (RTX) protein TosA, which is specifically expressed in the host urinary tract and contributes significantly to the virulence and survival of UPEC.tosA, found in strains within the B2 phylogenetic subgroup ofE. coli, serves as a marker for strains that also contain a large number of well-characterized UPEC virulence factors. The presence oftosAin anE. coliisolate predicts successful colonization of the murine model of ascending UTI, regardless of the source of the isolate. Here, a detailed analysis of the function oftosArevealed that this gene is transcriptionally linked to genes encoding a conserved type 1 secretion system similar to other RTX family members. TosA localized to the cell surface and was found to mediate (i) adherence to host cells derived from the upper urinary tract and (ii) survival in disseminated infections and (iii) to enhance lethality during sepsis (as assessed in two different animal models of infection). An experimental vaccine, using purified TosA, protected vaccinated animals against urosepsis. From this work, it was concluded that TosA belongs to a novel group of RTX proteins that mediate adherence and host damage during UTI and urosepsis and could be a novel target for the development of therapeutics to treat ascending UTIs.

Download Full-text

Rainfall spatio-temporal correlation and intermittency structure from micro-γ to meso-β scale in the Netherlands

Journal of Hydrometeorology ◽

10.1175/jhm-d-20-0311.1 ◽

2021 ◽

Author(s):

Thomas C. van Leth ◽

Hidde Leijnse ◽

Aart Overeem ◽

Remko Uijlenhoet

Keyword(s):

The Netherlands ◽

Spatial Scales ◽

Rainfall Variability ◽

Temporal Structure ◽

Temporal Correlation ◽

Exponential Model ◽

Rain Gauges ◽

Wide Range ◽

Spatio Temporal ◽

Two Parameter

AbstractWe investigate the spatio-temporal structure of rainfall at spatial scales from 7m to over 200 km in the Netherlands. We used data from two networks of laser disdrometers with complementary interstation distances in two Dutch cities (comprising five and six disdrometers, respectively) and a Dutch nationwide network of 31 automatic rain gauges. The smallest aggregation interval for which raindrop size distributions were collected by the disdrometers was 30 s, while the automatic rain gauges provided 10-min rainfall sums. This study aims to supplement other micro-γ investigations (usually performed in the context of spatial rainfall variability within a weather radar pixel) with new data, while characterizing the correlation structure across an extended range of scales. To quantify the spatio-temporal variability, we employ a two-parameter exponential model fitted to the spatial correlograms and characterize the parameters of the model as a function of the temporal aggregation interval. This widely used method allows for a meaningful comparison with seven other studies across contrasting climatic settings all around the world. We also separately analyzed the intermittency of the rainfall observations. We show that a single parameterization, consisting of a two-parameter exponential spatial model as a function of interstation distance combined with a power-law model for decorrelation distance as a function of aggregation interval, can coherently describe rainfall variability (both spatial correlation and intermittency) across a wide range of scales. Limiting the range of scales to those typically found in micro-γ variability studies (including four of the seven studies to which we compare our results) skews the parameterization and reduces its applicability to larger scales.

Download Full-text

A Survey on Spatio-temporal Data Analytics Systems

ACM Computing Surveys ◽

10.1145/3507904 ◽

2022 ◽

Author(s):

Md Mahbub Alam ◽

Luis Torgo ◽

Albert Bifet

Keyword(s):

Programming Languages ◽

Spatial Data ◽

Data Analytics ◽

Spatial Databases ◽

Location Based Services ◽

Temporal Data ◽

Wide Range ◽

Data Volume ◽

Spatio Temporal ◽

Gis Software

Due to the surge of spatio-temporal data volume, the popularity of location-based services and applications, and the importance of extracted knowledge from spatio-temporal data to solve a wide range of real-world problems, a plethora of research and development work has been done in the area of spatial and spatio-temporal data analytics in the past decade. The main goal of existing works was to develop algorithms and technologies to capture, store, manage, analyze, and visualize spatial or spatio-temporal data. The researchers have contributed either by adding spatio-temporal support with existing systems, by developing a new system from scratch, or by implementing algorithms for processing spatio-temporal data. The existing ecosystem of spatial and spatio-temporal data analytics systems can be categorized into three groups, (1) spatial databases (SQL and NoSQL), (2) big spatial data processing infrastructures, and (3) programming languages and GIS software. Since existing surveys mostly investigated infrastructures for processing big spatial data, this survey has explored the whole ecosystem of spatial and spatio-temporal analytics. This survey also portrays the importance and future of spatial and spatio-temporal data analytics.

Download Full-text