Fast learning without synaptic plasticity in spiking neural networks.

Spiking neural networks are of high current interest, both from the perspective of modelling neural networks of the brain and for porting their fast learning capability and energy efficiency into neuromorphic hardware. But so far we have not been able to reproduce fast learning capabilities of the brain in spiking neural networks. Biological data suggest that a synergy of synaptic plasticity on a slow time scale with network dynamics on a faster time scale is responsible for fast learning capabilities of the brain. We show here that a suitable orchestration of this synergy between synaptic plasticity and network dynamics does in fact reproduce fast learning capabilities of generic recurrent networks of spiking neurons. This points to the important role of recurrent connections in spiking networks, since these are necessary for enabling salient network dynamics. We show more specifically that the proposed synergy enables synaptic weights to encode more general information such as priors and task structures, since moment-to-moment processing of new information can be delegated to the network dynamics.

An exact mapping from ReLU networks to spiking neural networks.

Deep spiking neural networks (SNNs) offer the promise of low-power artificial intelligence. However, training deep SNNs from scratch or converting deep artificial neural networks to SNNs without loss of performance has been a challenge. Here we propose an exact mapping from a network with Rectified Linear Units (ReLUs) to an SNN that fires exactly one spike per neuron. For our constructive proof, we assume that an arbitrary multi-layer ReLU network with or without convolutional layers, batch normalization and max pooling layers was trained to high performance on some training set. Furthermore, we assume that we have access to a representative example of input data used during training and to the exact parameters (weights and biases) of the trained ReLU network. The mapping from deep ReLU networks to SNNs causes zero percent drop in accuracy on CIFAR10, CIFAR100 and the ImageNet-like data sets Places365 and PASS. More generally our work shows that an arbitrary deep ReLU network can be replaced by an energy-efficient single-spike neural network without any loss of performance.

Healthcare-associated adverse events and readmission to the emergency departments within seven days after a first consultation.

Introduction: The use of emergency hospital service has become increasingly frequent with a rise of approximately 3.6%. in annual emergency department visits. The objective of this study was to describe the reasons for reconsultations to emergency departments and to identify the risk and protective factors of reconsultations linked to healthcare-associated adverse events. Materials and methods: A retrospective, descriptive, multicenter study was performed in the emergency department of Troyes Hospital and the Sainte Anne Army Training Hospital in Toulon, France from January 1 to December 31, 2019. Patients over 18 years of age who returned to the emergency department for a reconsultation within 7 days were included. Healthcare-associated adverse events in the univariate analysis (p < 0.10) were introduced into a multivariate logistic regression model. Model performance was examined using the Hosmer-Lemeshow test and calculated with c-statistic. Results: Weekend visits and performing radiology examinations were risk factors linked to healthcare associated adverse events. Biological examinations and the opinion of a specialist were protective factors. Discussion: Numerous studies have reported that a first consultation occurring on a weekend is a reconsultation risk factor for healthcare-associated adverse events, however, performing radiology examinations were subjected to confusion bias. Patients having radiology examinations due to trauma-related pathologies were more apt for a reconsultation. Conclusion: Our study supports the need for better emergency departments access to biological examinations and specialist second medical opinions. An appropriate patient to doctor ratio in hospital emergency departments may be necessary at all times.

Assessment of emergency physicians' performance in identifying shockable rhythm in out-of-hospital cardiac arrest: an observational simulation study.

BACKGROUND: Emergency physicians can use a manual or an automated defibrillator to provide defibrillation of patients who had out-of-hospital cardiac arrest (OHCA). Performance of emergency physicians in identifying shockable rhythm with a manual defibrillator has been poorly explored whereas that of automated defibrillators is well known (sensitivity 0.91-1.00, specificity 0.96-0.99). We conducted this study to estimate the sensitivity/specificity and speed of shock/no-shock decision-making by prehospital emergency physicians for shockable or non-shockable rhythm, and their preference for manual versus automated defibrillation. METHODS: We developed a web application that simulates a manual defibrillator (https://simul-shock.firebaseapp.com/). In 2019, all (262) emergency physicians of six French emergency medical services were invited to participate in a study in which 60 ECG rhythms from real OHCA recordings were successively presented to the physicians for determination of whether they would or would not administer a shock. Time to decision was recorded. Answers were compared with a gold standard (concordant answers of three experts). We report sensitivity for shockable rhythms (decision to shock) and specificity for non-shockable rhythms (decision not to shock). Physicians were also asked whether they preferred manual or automated defibrillation. RESULTS: Among 215 respondents, we were able to analyse results for 190 physicians. 57% of emergency physicians preferred manual defibrillation. Median (IQR) sensitivity for a shock delivery for shockable rhythm was 0.91 (0.81-1.00); median specificity for no-shock delivery for non-shockable rhythms was 0.91 (0.80-0.96). More precisely, sensitivities for shock delivery for ventricular tachycardia (VT) and coarse ventricular fibrillation (VF) were both 1.0 (1.0-1.0); sensitivity for fine VF was 0.6 (0.2-1). Specificity for not shocking a pulseless electrical activity (PEA) was 0.83 (0.72-0.86), and for asystole, specificity was 0.93 (0.86-1). Median speed of decision-making (in seconds) were: VT 2.0 (1.6-2.7), coarse VF 2.1 (1.7-2.9), asystole 2.4 (1.8-3.5), PEA 2.8 (2.0-4.2) and fine VF 2.8 (2.1-4.3). CONCLUSIONS: Global sensitivity and specificity were comparable with published automated external defibrillator studies. Shockable rhythms with the best clinical prognoses (VT and coarse VF) were very rapidly recognised with very good sensitivity. The decision-making for fine VF or asystole and PEA was less accurate.

Spike frequency adaptation supports network computations on temporally dispersed information.

For solving tasks such as recognizing a song, answering a question, or inverting a sequence of symbols, cortical microcircuits need to integrate and manipulate information that was dispersed over time during the preceding seconds. Creating biologically realistic models for the underlying computations, especially with spiking neurons and for behaviorally relevant integration time spans, is notoriously difficult. We examine the role of spike frequency adaptation in such computations and find that it has a surprisingly large impact. The inclusion of this well-known property of a substantial fraction of neurons in the neocortex - especially in higher areas of the human neocortex - moves the performance of spiking neural network models for computations on network inputs that are temporally dispersed from a fairly low level up to the performance level of the human brain.

A solution to the learning dilemma for recurrent networks of spiking neurons.

Recurrently connected networks of spiking neurons underlie the astounding information processing capabilities of the brain. Yet in spite of extensive research, how they can learn through synaptic plasticity to carry out complex network computations remains unclear. We argue that two pieces of this puzzle were provided by experimental data from neuroscience. A mathematical result tells us how these pieces need to be combined to enable biologically plausible online network learning through gradient descent, in particular deep reinforcement learning. This learning method-called e-prop-approaches the performance of backpropagation through time (BPTT), the best-known method for training recurrent neural networks in machine learning. In addition, it suggests a method for powerful on-chip learning in energy-efficient spike-based hardware for artificial intelligence.

Epidemiology of physician interventions in maritime environment by the Marseille Fire Brigade (BMPM) from 2005 to 2017.

BACKGROUND: Marseille is the second largest city in France. The Marseille Fire Brigade (BMPM) is the largestunity of the French Navy. This organization is in charge of rescue operations and medical intervention in theMarseille area. The aim of the study was to describe the epidemiology of interventions that required a physicianto be present that were performed by the BMPM between the years of 2005 to 2017. MATERIALS AND METHODS: The statistical office database of the BMPM and the medical interventions forms (FIM)acquired from the BMPM medical ambulances (SMUR) archives were analysed from the years 2005 to 2017. RESULTS: The BMPM performed a total of 2,375 interventions in the maritime environment between 2005and 2017. A physician was necessary for intervention a total of 186 times. The extraction and analysisreports of 107 medical intervention forms found the BMPM archives revealed a significant number ofinterventions (67%) in the southern bay of Marseille and Frioul, specifically from the If and Planier islands.The majority of interventions (77%) took place within the 300m band. The most common cause of medicalintervention was due to an accidental fall into the water, followed by boating (sailing and motor), and swimming.Drowning was the most common cause of mortality, consisting of 34% of all interventions. Divingaccidents represented 14% of interventions. Trauma affected 22% of the study population and 83% oftrauma patients were transported to the hospital under the supervision of a physician. CONCLUSIONS: Potential areas for improvement in the management of drowning victims are the use ofSzpilman's classification, sonography, and non-invasive ventilation. A recertification course for medicaleducation training of BMPM doctors on the management of diving accidents could help to optimize theinformation recorded on FIM. Accident prevention training should be continued and reinforced when itcomes to maritime activities.

Memory-Efficient Deep Learning on a SpiNNaker 2 Prototype.

The memory requirement of deep learning algorithms is considered incompatible with the memory restriction of energy-efficient hardware. A low memory footprint can be achieved by pruning obsolete connections or reducing the precision of connection strengths after the network has been trained. Yet, these techniques are not applicable to the case when neural networks have to be trained directly on hardware due to the hard memory constraints. Deep Rewiring (DEEP R) is a training algorithm which continuously rewires the network while preserving very sparse connectivity all along the training procedure. We apply DEEP R to a deep neural network implementation on a prototype chip of the 2nd generation SpiNNaker system. The local memory of a single core on this chip is limited to 64 KB and a deep network architecture is trained entirely within this constraint without the use of external memory. Throughout training, the proportion of active connections is limited to 1.3%. On the handwritten digits dataset MNIST, this extremely sparse network achieves 96.6% classification accuracy at convergence. Utilizing the multi-processor feature of the SpiNNaker system, we found very good scaling in terms of computation time, per-core memory consumption, and energy constraints. When compared to a X86 CPU implementation, neural network training on the SpiNNaker 2 prototype improves power and energy consumption by two orders of magnitude.

Slow feature analysis with spiking neurons and its application to audio stimuli.

Extracting invariant features in an unsupervised manner is crucial to perform complex computation such as object recognition, analyzing music or understanding speech. While various algorithms have been proposed to perform such a task, Slow Feature Analysis (SFA) uses time as a means of detecting those invariants, extracting the slowly time-varying components in the input signals. In this work, we address the question of how such an algorithm can be implemented by neurons, and apply it in the context of audio stimuli. We propose a projected gradient implementation of SFA that can be adapted to a Hebbian like learning rule dealing with biologically plausible neuron models. Furthermore, we show that a Spike-Timing Dependent Plasticity learning rule, shaped as a smoothed second derivative, implements SFA for spiking neurons. The theory is supported by numerical simulations, and to illustrate a simple use of SFA, we have applied it to auditory signals. We show that a single SFA neuron can learn to extract the tempo in sound recordings.