MC-QDSNN: Quantized Deep evolutionary SNN with Multi-Dendritic Compartment Neurons for Stress Detection using Physiological Signals

TL;DR

The proposed multicompartment leaky (MCLeaky) neuron-based spiking neural network (SNN) model and its quantized variant were benchmarked against state-of-the-art (SOTA) spiking LSTMs to perform human stress detection by comparing computing requirements, compute-latency, and real-world performances on freshly acquired unseen data with models that is acquired by employing neural architecture search (NAS).

Abstract

Long short-term memory (LSTM) has emerged as a definitive network for analyzing and inferring time series data. LSTM has the capability to extract spectral features and a mixture of temporal features. Due to this benefit, a similar feature extraction method is explored for the spiking counterparts targeting time-series data. Though LSTMs perform well in their spiking form, they tend to be compute and power intensive. Addressing this issue, this work proposes Multi-Compartment Leaky (MCLeaky) neuron as a viable alternative for efficient processing of time series data. The MCLeaky neuron, derived from the Leaky Integrate and Fire (LIF) neuron model, contains multiple memristive synapses interlinked to form a memory component, which emulates the human brain's Hippocampus region. The proposed MCLeaky neuron based Spiking Neural Network model and its quantized variant were benchmarked against state-of-the-art (SOTA) Spiking LSTMs to perform human stress detection, by comparing compute requirements, latency and real-world performances on unseen data with models derived through Neural Architecture Search (NAS). Results show that networks with MCLeaky activation neuron managed a superior accuracy of 98.8% to detect stress based on Electrodermal Activity (EDA) signals, better than any other investigated models, while using 20% less parameters on average. MCLeaky neuron was also tested for various signals including EDA Wrist and Chest, Temperature, ECG, and combinations of them. Quantized MCLeaky model was also derived and validated to forecast their performance on hardware architectures, which resulted in 91.84% accuracy. The neurons were evaluated for multiple modalities of data towards stress detection, which resulted in energy savings of 25.12x to 39.20x and EDP gains of 52.37x to 81.9x over ANNs, while offering a best accuracy of 98.8% when compared with the rest of the SOTA implementations.

Figures (15)

• Figure 1: Schematic representation of the Multi-Compartment Dendritic Neuron model.
• Figure 2: The two different versions of WESAD dataset Schmidt2018 study protocol is shown. The red boxes refer to transition state between two emotions where the subjects are made to self label the experienced emotion.
• Figure 3: A sample of Min-Max normalized EDA chest sensor data (in blue) and EDA wrist sensor data (in orange) for subject S2 picked from WESAD dataset versus time for the five labels - Amusement (Amuse), Meditation (Medit), Stress, Baseline (Base), and Transient. The green line represents the emotion label of that particular sample, which is denoted on the second y axis. Only some of the label definition are provided in WESAD, hence apart from these five, all others are referred to as "ignore" labels . The transient label represents the in-between phase of the recording emotions.
• Figure 4: Schematic representation of: (a) LSTM network, showcasing all gates, (b) Proposed MCLeaky neuron, consisting of multiple dendrite compartments and a single Soma block, (c) Highlighted Hippocampus location, that inspired the MCLeaky's design, (d) Multi-dendritic Compartment neuron model with a Soma and multiple dendrite compartments. Besides the membrane potential, which portrays the memory of the Soma, each of the compartment is enabled with decaying memory in the form of the dendritic potentials and recurrent connections with decay factors. (e) Illustration of MCLeaky neuron-based feed-forward SNN with recurrent connections. The built-in memory component in the MCLeaky neuron, induces sparsity for MC-SNN model, as compared to the LIF-based SNNs.
• Figure 5: System design approach: EDA signals from WESAD dataset are rate encoded and fed to the proposed NE-QDSNN along with the pre-defined Grammar. NE-QDSNN generates best choices of QDSNN along with stateful spiking neuron parameters. QDSNN model is further validated for an in-house data-acquisition unit built for EDA signal.