Neural networks with siamese and triplet structures can discriminate between samples by explicitly learning the distance measure. But the dominant metric learning model based on deep neural networks (DNN) has issues with low energy efficiency and slow on-chip operating speed. Motivated by this, we put forth novel impulsive neural networks (INN) based deep metric learning approachs. The loss function is proposed based on the Euclidean distance between the membrane potentials of the neurons in the output layer of the sub-networks composed by Leaky integrate-and-fire (LIF) neurons. Furthermore, the back-propagation through time (BPTT) technique can be effectively implemented and utilized for network optimization through the surrogate functions. Experiments on multiple datasets show that the proposed model can learn the difference between images with performance very close to DNN-based models. At the same time, we conducted an intuitive energy consumption analysis of our model. The results show that compared with DNN-based models, the proposed model has higher energy efficiency. Besides, it only needs an extremely short time step T=10 to complete inference process, which makes the low-latency on-chip operation possible.

Several related studies have investigated the feasibility of combining bio-inspired models with speech recognition tasks. However, developing a transplantable speech frontend for an impulse-based model that can handle raw waveforms still poses several challenges. Therefore, this paper presents a novel impulse-based model for raw waveform-based speech recognition tasks. Our model comprises a learnable multi-input multi-channel time encoding machine (mMC-TEM) and a simple impulsive convolution neural network, which can learn directly from raw waveforms without any hand-crafted features. Our experiments demonstrate that our model achieved the highest test accuracy of 95.8% on the Google SC V1 dataset, outperforming the current state-of-the-art impulse-based models. Moreover, our model completes the inference process in only a short time step of T=5, while maintaining high performance compared to similar models.

This paper presents an impulsive-backpropagation neural network (IBNN) based learning algorithm for detecting Coronavirus Disease 2019 (COVID-19), by classifying chest computed tomography (CT) images. Inspired by the nerve impulses in brain networks, the IBNN algorithm consists of two parts: a multi-layered network of impulsive neurons and a gradient decent backpropagation mechanism. The effectiveness of the IBNN algorithm is validated on clinical COVID-19 database, and a classification accuracy of 98.19% is achieved. It is further demonstrated by comparative studies that the IBNN may outperform some other learning algorithms through the integration of nerve impulses and backpropagation. Considering the intricate attributes of the chest CT scan images, the IBNN algorithm also exhibits a potential capacity of pattern recognition on complicated samples.

In this article, we consider the design of dynamic transmission scheduling policies for the industrial network systems sharing scarce communication resources. Only a few subsystems can obtain channel access for information updates to close their control loops at each time step, which highlights the necessity of designing optimal transmission scheduling schemes to achieve a minimum average linear quadratic cost of the industrial network systems. We first propose a greedy state-error-dependent scheduling (SES) policy based on the one-step expected profit and discuss its stability employing the Lyapunov function method. After formulating the scheduling optimization as a Markov decision process problem and relaxing with a soft constraint, we develop a heuristic near-optimal solution that guarantees the optimality of certainty equivalent controllers, namely, Whittle’s index-inspired error-dependent scheduling (WIES). A stochastic stability condition of WIES is further given based on f-ergodicity. Due to low computational complexity and ease of implementation, the proposed schemes are suitable for large-scale heterogeneous industrial network systems. Finally, simulation results show that the proposed policies outperform the existing round-robin, holding-time-prioritized, and error-aware scheduling schemes.

In this article, we investigate the routing problem of packet networks through multiagent reinforcement learning (RL), which is a very challenging topic in distributed and autonomous networked systems. In specific, the routing problem is modeled as a networked multiagent partially observable Markov decision process (MDP). Since the MDP of a network node is not only affected by its neighboring nodes' policies but also the network traffic demand, it becomes a multitask learning problem. Inspired by recent success of RL and metalearning, we propose two novel model-free multiagent RL algorithms, named multiagent proximal policy optimization (MAPPO) and multiagent metaproximal policy optimization (meta-MAPPO), to optimize the network performances under fixed and time-varying traffic demand, respectively. A practicable distributed implementation framework is designed based on the separability of exploration and exploitation in training MAPPO. Compared with the existing routing optimization policies, our simulation results demonstrate the excellent performances of the proposed algorithms.

This paper proposes a multi-layer self-organizing impulse neural network. By adding an improved lateral inhibition mechanism and neuron model, the novel network can extract more information from the pictures, thus improving the expressive ability and elevating the recognition accuracy of the classification model. Besides, we replace the topology of the excitation layer and the inhibition layer with a recurrent connection structure. Several experiments have been carried out based on the proposed neural network and proved that the network can well complete the task of handwritten digit recognition. By tuning the network parameters, the recognition accuracy can reach 92.8%, which is at a high level in the same type of network. Besides, we also verified the robust performance of the network by randomly reducing the number of synapses and the number of neurons. It turns out that the network can still achieve high recognition accuracy after randomly discarding some neurons or synapses.

A new control strategy is presented for trajectory tracking of heterogeneous Unmanned Ground Vehicle (UGV)-Mobile Manipulators (MMS) system in this paper. The heterogenous UGV-MMS system is comprised of the UGV and the MMS. The main innovation of this paper is that the situation that the manipulator compensates the dynamic interactions produced by the UGV is considered, while the UGV doesn't compensate the dynamic interactions produced by the manipulator. In this situation, we do low calculation and the tracking precision is excellent. Firstly, a kinematic and dynamic controller is designed for the UGV to ensure the UGV to track the desired trajectory. Secondly, a control law based on the sliding mode control is designed for the manipulator to track the desired trajectory. Finally, the Lyapunov function approach and the simualtion results in MATLAB are used to demonstrate the effectiveness and feasibility of the proposed control approach of both the UGV and the manipulator.

The reinforcement learning is studied to solve the problem of multi-agent pursuit and evasion games in this article. The main problem of current reinforcement learning for multi-agents is the low learning efficiency of agents. An important factor leading to this problem is that the delay of the Q function is related to the environment changing. To solve this problem, a probabilistic distribution reward value is used to replace the Q function in the multi-agent depth deterministic policy gradient framework (hereinafter referred to as MADDPG). The distribution Bellman equation is proved to be convergent, and can be brought into the framework of reinforcement learning algorithm. The probabilistic distribution reward value is updated in the algorithm, so that the reward value can be more adaptive to the complex environment. In the same time, eliminating the delay of rewards improves the efficiency of the strategy and obtains a better pursuit-evasion results. The final simulation and experiment show that the multi-agent algorithm with distribution rewards achieves better results under the setting environment.

In robot golfing games, how to identify and track small balls is a critical step for scoring. Considering uneven illumination and intensity, irregular distribution, and blurred borders, an integrated detection method based on the random Hough transform and the Kalman filter is presented to improve the recognition accuracy, i.e., guaranteeing the detection accuracy and the stability in shooting. Experiment on the NAO robot is performed to show the effectiveness of the proposed detecting method. Both theoretical and experimental results suggest that the proposed recognition method can reduce the calculation time of the Hough transform and has a certain degree of robustness against uncertain environments.

Human-Robot Interaction (HRI) has been extensively investigated in academia, industries, and technology companies. How to teach a robot to write is recognized as a difficult task, involving perception, learning and control of the overall robotic system. This paper develops a robotic handwriting system on a NAO robot by using HRI and Q-learning. In the simulator setup, the human demonstrator first writes to form a desired writing content, with the coordinates of each stroke writing saved; then, by applying the Q-learning algorithm, the virtual NAO automatically learns to find the correct writing order and to optimize the strokes. With the learned knowledge, the real NAO uses its inverse kinematics for generating the joint values from the stroke trajectories, and thus can quickly replicate the letters or words written by the human demonstrator. It is shown that by using the proposed HRI-based handwriting system, the human can naturally and conveniently teach NAO to write many characters and NAO can also have an acceptable writing quality close to the human.

In this paper, a leader-following time-varying formation tracking control scheme is developed for heterogeneous multi-agent systems. The states of followers, i.e., unmanned aerial vehicles (UAVs), are controlled to keep a desired formation and track the state of the leader, i.e., unmanned ground vehicle (UGV). The dynamics of the UAVs can be transformed into a fourth-order linear system via a static feedback linearization scheme. A distributed consensus-based control algorithm is proposed for the heterogeneous UGV-UAVs systems with switching direct topologies, so the networked UAVs can track the trajectories of the UGV along a desired formation. Finally, the proposed control strategy is applied to solve the formation tracking problem of a heterogeneous system consisting of three UAVs and one UGV, and the theoretical results are validated by the simulation.

In this paper, an energy efficient transport protocol is proposed to enhance the traffic capacity of complex networks. The new algorithm is derived on the basis of router betweenness centrality, which gives an estimate of traffic handled by the node as router. As one bottleneck node will degrade the traffic performance of the whole network, the improved routing aims to bypass nodes with heavy load to diminish the congestion. Particularly, depending on a parameter that controls the optimization scale, the new transport protocol can achieve smaller average path length while guaranteeing the required packet generation rate. Comparing results of previous routing strategies, it is shown that the presented improved routing strategy performs more effectively.

Representative frequent pattern mining from a transaction dataset has been well studied in both the database and the data mining community for many years. One popular scenario is that if the input dataset contains private information, publishing representative patterns may pose great threats to individual’s privacy. In this paper, we study the subject of mining representative patterns under the differential privacy model. We propose a method that combines RPlocal with differential privacy to mine representative patterns. We analyze the breach of privacy in RPlocal, and utilize the differential privacy to protect the private information of transaction dataset. Through formal privacy analysis, we prove that our proposed algorithm satisfies ϵ -differential privacy. Extensive experimental results on real datasets reveal that our algorithm produces similar number of representative patterns compared to RPlocal.

In this paper, an optimal routing strategy is proposed to enhance the traffic capacity of complex networks. In order to avoid nodes overloading, the new algorithm is derived on the basis of generalized betweenness centrality which gives an estimate of traffic handled by the node for a route set. Since the nodes with large betweenness centrality are more susceptible to traffic congestion, the traffic can be improved, as our strategy, by redistributing traffic load from nodes with large betweenness centrality to nodes with small betweenness centrality in the proceeding of computing collective routing table. Particularly, depending on a parameter that controls the optimization scale, the new routing can not only enlarge traffic capacity of networks more, but also enhance traffic efficiency with smaller average path length. Comparing results of previous routing strategies, it is shown that the present improved routing performs more effectively.

In this paper, an improved routing strategy is proposed for enhancing the traffic capacity of scale-free networks. Instead of using the information of degree and betweenness centrality, the new algorithm is derived on the basis of the expanding betweenness centrality of nodes, which gives an estimate of the traffic handled by the vertex for a certain route set. Since the nodes with large betweenness centrality are more susceptible to traffic congestion, the traffic can be improved by redistributing traffic loads from nodes with large betweenness centrality to nodes with small betweenness centrality in the process of computing the collective routing table. Comparing with results of previous routing strategies, it is shown that the present improved routing performs more effectively.