Comparison of Pre-Trained CNNs for Audio Classification Using Transfer Learning

The paper investigates retraining options and the performance of pre-trained Convolutional Neural Networks (CNNs) for sound classification. CNNs were initially designed for image classification and recognition, and, at a second phase, they extended towards sound classification. Transfer learning is a promising paradigm, retraining already trained networks upon different datasets. We selected three ‘Image’- and two ‘Sound’-trained CNNs, namely, GoogLeNet, SqueezeNet, ShuffleNet, VGGish, and YAMNet, and applied transfer learning. We explored the influence of key retraining parameters, including the optimizer, the mini-batch size, the learning rate, and the number of epochs, on the classification accuracy and the processing time needed in terms of sound preprocessing for the preparation of the scalograms and spectrograms as well as CNN training. The UrbanSound8K, ESC-10, and Air Compressor open sound datasets were employed. Using a two-fold criterion based on classification accuracy and time needed, we selected the ‘champion’ transfer-learning parameter combinations, discussed the consistency of the classification results, and explored possible benefits from fusing the classification estimations. The Sound CNNs achieved better classification accuracy, reaching an average of 96.4% for UrbanSound8K, 91.25% for ESC-10, and 100% for the Air Compressor dataset.

Model-based event-triggered adaptive formation control for underactuated surface vehicles via the minimal learning parameter technique

This article presents an adaptive neural formation control algorithm for underactuated surface vehicles by the model-based event-triggered method. In the algorithm, the leader–follower structure is employed to construct the formation model. Meanwhile, two new coordinate variables are introduced to avoid the possible singularity problem that exists in the polar coordinate system. Furthermore, the event-triggered mechanism is developed by constructing the adaptive model in a concise form. Related state variables and control parameters are required to be updated only at the event-triggered instants. Thus, the communication load between the controller and the actuator could be effectively reduced. Besides, for merits of the radial basis function neural network and the minimal learning parameter techniques, only two adaptive parameters are employed to compensate for the effects of the model uncertainties and the external disturbances. With the Lyapunov theory, all signals in the closed-loop system are proved to be semi-global uniformly ultimately bounded. Finally, numerical simulations are conducted to illustrate the effectiveness and feasibility of the proposed algorithm.

SyNC, a Computationally Extensive and Realistic Neural Net to Identify Relative Impacts of Synaptopathy Mechanisms on Glutamatergic Neurons and Their Networks in Autism and Complex Neurological Disorders

Synaptic function and experience-dependent plasticity across multiple synapses are dependent on the types of neurons interacting as well as the intricate mechanisms that operate at the molecular level of the synapse. To understand the complexity of information processing at synaptic networks will rely in part on effective computational models. Such models should also evaluate disruptions to synaptic function by multiple mechanisms. By co-development of algorithms alongside hardware, real time analysis metrics can be co-prioritized along with biological complexity. The hippocampus is implicated in autism spectrum disorders (ASD) and within this region glutamatergic neurons constitute 90% of the neurons integral to the functioning of neuronal networks. Here we generate a computational model referred to as ASD interrogator (ASDint) and corresponding hardware to enable in silicon analysis of multiple ASD mechanisms affecting glutamatergic neuron synapses. The hardware architecture Synaptic Neuronal Circuit, SyNC, is a novel GPU accelerator or neural net, that extends discovery by acting as a biologically relevant realistic neuron synapse in real time. Co-developed ASDint and SyNC expand spiking neural network models of plasticity to comparative analysis of retrograde messengers. The SyNC model is realized in an ASIC architecture, which enables the ability to compute increasingly complex scenarios without sacrificing area efficiency of the model. Here we apply the ASDint model to analyse neuronal circuitry dysfunctions associated with autism spectral disorder (ASD) synaptopathies and their effects on the synaptic learning parameter and demonstrate SyNC on an ideal ASDint scenario. Our work highlights the value of secondary pathways in regard to evaluating complex ASD synaptopathy mechanisms. By comparing the degree of variation in the synaptic learning parameter to the response obtained from simulations of the ideal scenario we determine the potency and time of the effect of a particular evaluated mechanism. Hence simulations of such scenarios in even a small neuronal network now allows us to identify relative impacts of changed parameters and their effect on synaptic function. Based on this, we can estimate the minimum fraction of a neuron exhibiting a particular dysfunction scenario required to lead to complete failure of a neural network to coordinate pre-synaptic and post-synaptic outputs.

Toward a More Accurate 3D Atlas of C. elegans Neurons

Determining cell identity in volumetric images of tagged neuronal nuclei is an ongoing challenge in contemporary neuroscience. Frequently, cell identity is determined by aligning and matching tags to an "atlas" of labeled neuronal positions and other identifying characteristics. Previous analyses of such C. elegans datasets have been hampered by the limited accuracy of such atlases, especially for neurons present in the ventral nerve cord, and also by time-consuming manual elements of the alignment process. We present a novel automated alignment method for sparse and incomplete point clouds of the sort resulting from typical C. elegans fluorescence microscopy datasets. This method involves a tunable learning parameter and a kernel that enforces biologically realistic deformation. We also present a pipeline for creating alignment atlases from datasets of the recently developed NeuroPAL transgene. In combination, these advances allow us to label neurons in volumetric images with confidence much higher than previous methods. We release, to the best of our knowledge, the most complete C. elegans 3D positional neuron atlas, encapsulating positional variability derived from 7 animals, for the purposes of cell-type identity prediction for myriad applications (e.g., imaging neuronal activity, gene expression, and cell-fate).

Neural Network-Based Finite-Time Fault-Tolerant Control for Spacecraft without Unwinding

In this paper, we focus on solving the problems of inertia-free attitude tracking control for spacecraft subject to external disturbance, unknown inertial parameters, and actuator faults. The robust control architecture is designed by using the rotation matrix and neural networks. In the presence of external disturbance and parametric uncertainties, a fault-tolerant control (FTC) scheme synthesized with the minimum-learning-parameter (MLP) algorithm is proposed to improve the reliability of the system when unknown actuator faults occur. These methods are developed based on backstepping to ensure that finite-time convergence is achievable for the entire closed-loop system states with low computational complexity. The validity and advantage of the designed controllers are highlighted by using Lyapunov-based analysis. Finally, the simulation results demonstrate the satisfactory performance of the developed controllers.

State space constrained iterative learning control for 3DOF robotic manipulator

In this paper, the trajectory tracking problem of a nonlinear robotic system with 3DOFs under the control signal obtained through nonlinearly constrained state spaceIterative Learning Control (ILC) methods is considered. The focus of this paper is the analysis of different control system parameters on the convergence rate of two constrained state space ILCalgorithms: Bounded Error Algorithm (BEAILC) and Constrained Output algorithm (COILC), as well as the comparison between these two algorithms through simulations. The obtained results have shown that COILC algorithm converges faster than BEAILC algorithm when compared with the same learning and feedback parameters, due to lower trajectory restrictions. Also, it has been shown that an increase in feedback gains can decrease the number of iteration terminations during the learning process, thus allowing for more of the trajectory error information to be learned from during the single iteration. Moreover, simulations have shown that the decrease in learning parameter values will increase the number of iterations required to obtain the desired tracking accuracy.

Intelligent random noise modeling by the improved variational autoencoding method and its application to data augmentation

Deep-learning methods facilitate the development of seismic data processing methods; however, they also offer some challenges. The primary challenges are the lack of labeled samples for training, due to heterogeneity in seismic data, expensive acquisition apparatus, and data confidentiality. These problems limit the acquisition of high-quality training data. To solve this problem, we have developed variational autoencoding (VAE) to generate synthetic noise for data augmentation; however, the simplified Kullback-Leibler (KL) distance definition and parameter learning result in the outputs of the original VAE being blurry. To optimize VAE for simulating random desert noise and improve its simulation capability, here we have developed an improved VAE based on KL redefinition and learning parameter replacement. Specifically, we (1) build a training set containing desert random noise samples, (2) redefine the KL distance calculated between two Gaussian mixture densities (rather than two simple Gaussians) because the KL distance plays an important role in the learning accuracy of VAE, and (3) use [Formula: see text] rather than [Formula: see text] to improve the learning efficiency. Statistical analysis indicates that the simulated random noise is statistically indistinguishable from real noise, indicating that our improved VAE is suitable for noise modeling. We also trained a denoising convolutional neural network (DnCNN) using the simulated noise. Data augmentation conducted using the simulated noise improved the effect of DnCNN, proving that our method contributes to data augmentation.