A Spike Time-Dependent Online Learning Algorithm Derived From Biological Olfaction

A simple model for Behavioral Time Scale Synaptic Plasticity (BTSP) provides content addressable memory with binary synapses and one-shot learning

Recent experimental studies in the awake brain have identified a rule for synaptic plasticity that is instrumental for the instantaneous creation of memory traces in area CA1 of the mammalian brain: Behavioral Time scale Synaptic Plasticity. This one-shot learning rule differs in five essential aspects from previously considered plasticity mechanisms. We introduce a transparent model for the core function of this learning rule and establish a theory that enables a principled understanding of the system of memory traces that it creates. Theoretical predictions and numerical simulations show that our model is able to create a functionally powerful content-addressable memory without the need for high-resolution synaptic weights. Furthermore, it reproduces the repulsion effect of human memory, whereby traces for similar memory items are pulled apart to enable differential downstream processing. Altogether, our results create a link between synaptic plasticity in area CA1 of the hippocampus and its network function. They also provide a promising approach for implementing content-addressable memory with on-chip learning capability in highly energy-efficient crossbar arrays of memristors.

A Novel Calibration Scheme of Gas Sensor Array for a More Accurate Measurement Model of Mixed Gases

Gas sensor arrays (GSAs) usually encounter challenges due to the cross-contamination of mixed gases, leading to reduced accuracy in measuring gas mixtures. However, with the advent of artificial intelligence, there is a promising avenue for addressing this issue effectively. In pursuit of more accurate mixed gas measurements, we proposed a measurement model leveraging neural networks. Our approach involved employing the encoder of an autoencoder network (AEN) to extract features from experimental data, while fully connected layers were utilized for predicting concentrations of mixed gases. To refine the neural network parameters, we employed a variational autoencoder to generate additional data resembling the distribution of experimental data. Subsequently, we designed a domain difference maximum entropy technique to identify optimal concentration points for the calibration data. These calibration points were instrumental in training the fully connected layers, enhancing the model's accuracy. During practical usage, with the AEN configuration fixed, the model can be fine-tuned by using a small subset of test points across large-scale GSA deployments. Simulation and practical measurement results demonstrated the efficacy of our proposed measurement model, boasting high accuracy, with confidence intervals for relative errors of the four gas measurements below 3% at the 95% confidence level. Besides, the calibration scheme reduced the number of test points compared with traditional methods, reducing the cost of labor and equipment.

A Bio-Inspired Spiking Neural Network with Few-Shot Class-Incremental Learning for Gas Recognition

The sensitivity and selectivity profiles of gas sensors are always changed by sensor drifting, sensor aging, and the surroundings (e.g., temperature and humidity changes), which lead to a serious decline in gas recognition accuracy or even invalidation. To address this issue, the practical solution is to retrain the network to maintain performance, leveraging its rapid, incremental online learning capacity. In this paper, we develop a bio-inspired spiking neural network (SNN) to recognize nine types of flammable and toxic gases, which supports few-shot class-incremental learning, and can be retrained quickly with a new gas at a low accuracy cost. Compared with gas recognition approaches such as support vector machine (SVM), k-nearest neighbor (KNN), principal component analysis (PCA) +SVM, PCA+KNN, and artificial neural network (ANN), our network achieves the highest accuracy of 98.75% in five-fold cross-validation for identifying nine types of gases, each with five different concentrations. In particular, the proposed network has a 5.09% higher accuracy than that of other gas recognition algorithms, which validates its robustness and effectiveness for real-life fire scenarios.

Gas Recognition in E-Nose System: A Review

Gas recognition is essential in an electronic nose (E-nose) system, which is responsible for recognizing multivariate responses obtained by gas sensors in various applications. Over the past decades, classical gas recognition approaches such as principal component analysis (PCA) have been widely applied in E-nose systems. In recent years, artificial neural network (ANN) has revolutionized the field of E-nose, especially spiking neural network (SNN). In this paper, we investigate recent gas recognition methods for E-nose, and compare and analyze them in terms of algorithms and hardware implementations. We find each classical gas recognition method has a relatively fixed framework and a few parameters, which makes it easy to be designed and perform well with limited gas samples, but weak in multi-gas recognition under noise. While ANN-based methods obtain better recognition accuracy with flexible architectures and lots of parameters. However, some ANNs are too complex to be implemented in portable E-nose systems, such as deep convolutional neural networks (CNNs). In contrast, SNN-based gas recognition methods achieve satisfying accuracy and recognize more types of gases, and could be implemented with energy-efficient hardware, which makes them a promising candidate in multi-gas identification.

A Sparsity-Driven Backpropagation-Less Learning Framework Using Populations of Spiking Growth Transform Neurons

Growth-transform (GT) neurons and their population models allow for independent control over the spiking statistics and the transient population dynamics while optimizing a physically plausible distributed energy functional involving continuous-valued neural variables. In this paper we describe a backpropagation-less learning approach to train a network of spiking GT neurons by enforcing sparsity constraints on the overall network spiking activity. The key features of the model and the proposed learning framework are: (a) spike responses are generated as a result of constraint violation and hence can be viewed as Lagrangian parameters; (b) the optimal parameters for a given task can be learned using neurally relevant local learning rules and in an online manner; (c) the network optimizes itself to encode the solution with as few spikes as possible (sparsity); (d) the network optimizes itself to operate at a solution with the maximum dynamic range and away from saturation; and (e) the framework is flexible enough to incorporate additional structural and connectivity constraints on the network. As a result, the proposed formulation is attractive for designing neuromorphic tinyML systems that are constrained in energy, resources, and network structure. In this paper, we show how the approach could be used for unsupervised and supervised learning such that minimizing a training error is equivalent to minimizing the overall spiking activity across the network. We then build on this framework to implement three different multi-layer spiking network architectures with progressively increasing flexibility in training and consequently, sparsity. We demonstrate the applicability of the proposed algorithm for resource-efficient learning using a publicly available machine olfaction dataset with unique challenges like sensor drift and a wide range of stimulus concentrations. In all of these case studies we show that a GT network trained using the proposed learning approach is able to minimize the network-level spiking activity while producing classification accuracy that are comparable to standard approaches on the same dataset.

A Systematic Framework for Olfactory Bulb Signal Transformations

We describe an integrated theory of olfactory systems operation that incorporates experimental findings across scales, stages, and methods of analysis into a common framework. In particular, we consider the multiple stages of olfactory signal processing as a collective system, in which each stage samples selectively from its antecedents. We propose that, following the signal conditioning operations of the nasal epithelium and glomerular-layer circuitry, the plastic external plexiform layer of the olfactory bulb effects a process of category learning-the basis for extracting meaningful, quasi-discrete odor representations from the metric space of undifferentiated olfactory quality. Moreover, this early categorization process also resolves the foundational problem of how odors of interest can be recognized in the presence of strong competitive interference from simultaneously encountered background odorants. This problem is fundamentally constraining on early-stage olfactory encoding strategies and must be resolved if these strategies and their underlying mechanisms are to be understood. Multiscale general theories of olfactory systems operation are essential in order to leverage the analytical advantages of engineered approaches together with our expanding capacity to interrogate biological systems.

In vivo odorant input induces spike timing-dependent plasticity of glutamatergic synapses in developing zebrafish olfactory bulb

Early odorant experience and neural activity are essential for refining developing neural connections. Although neural activity-induced synaptic plasticity is one of the most important cellular mechanisms underlying the refinement of neural circuits, whether and how natural odorant experience induces long-term plasticity in the olfactory bulb remains unknown. In vivo perforated whole-cell recording from mitral cells (MCs) in larval zebrafish showed that odorant experience induced persistent modification of developing olfactory bulb circuits via spike timing-dependent plasticity (STDP). Repetitive odorant stimuli paired with postsynaptic spiking in a critical time window (pre-post, positive timing) resulted in persistent enhancement of glutamatergic inputs from olfactory sensory neurons, but long-term depression within the opposite time window (post-pre, negative timing). Furthermore, spike-timing-dependent potentiation (tLTP) in STDP induced by repetitive odorant stimulation had similar cellular processes to those of electrical stimulation-induced tLTP. Finally, odorant input induced STDP required the activation of postsynaptic N-methyl-d-aspartate receptors (NMDARs). Thus, the NMDAR is likely to be a postsynaptic coincidence detector responsible for the sensory experience-dependent refinement of developing connections.

Bio-Inspired Strategies for Improving the Selectivity and Sensitivity of Artificial Noses: A Review

Artificial noses are broad-spectrum multisensors dedicated to the detection of volatile organic compounds (VOCs). Despite great recent progress, they still suffer from a lack of sensitivity and selectivity. We will review, in a systemic way, the biomimetic strategies for improving these performance criteria, including the design of sensing materials, their immobilization on the sensing surface, the sampling of VOCs, the choice of a transduction method, and the data processing. This reflection could help address new applications in domains where high-performance artificial noses are required such as public security and safety, environment, industry, or healthcare.

Rapid online learning and robust recall in a neuromorphic olfactory circuit

We present a neural algorithm for the rapid online learning and identification of odourant samples under noise, based on the architecture of the mammalian olfactory bulb and implemented on the Intel Loihi neuromorphic system. As with biological olfaction, the spike timing-based algorithm utilizes distributed, event-driven computations and rapid (one-shot) online learning. Spike timing-dependent plasticity rules operate iteratively over sequential gamma-frequency packets to construct odour representations from the activity of chemosensor arrays mounted in a wind tunnel. Learned odourants then are reliably identified despite strong destructive interference. Noise resistance is further enhanced by neuromodulation and contextual priming. Lifelong learning capabilities are enabled by adult neurogenesis. The algorithm is applicable to any signal identification problem in which high-dimensional signals are embedded in unknown backgrounds.