Towards Sensorimotor Coupling of a Spiking Neural Network and Deep Reinforcement Learning for Robotics Application.



Deep reinforcement learning augments the reinforcement learning framework and utilizes the powerful representation of deep neural networks. Recent works have demonstrated the great achievements of deep reinforcement learning in various domains including finance,medicine, healthcare, video games, robotics and computer vision.Deep neural network was started with multi-layer perceptron (1stgeneration) and developed to deep neural networks (2ndgeneration)and it is moving forward to spiking neural networks which are knownas3rdgeneration of neural networks. Spiking neural networks aim to bridge the gap between neuroscience and machine learning, using biologically-realistic models of neurons to carry out computation. In this thesis, we first provide a comprehensive review on both spiking neural networks and deep reinforcement learning with emphasis on robotic applications. Then we will demonstrate how to develop a robotics application for context-aware scene understanding to perform sensorimotor coupling. Our system contains two modules corresponding to scene understanding and robotic navigation. The first module is implemented as a spiking neural network to carry out semantic segmentation to understand the scene in front of the robot. The second module provides a high-level navigation command to robot, which is considered as an agent and implemented by online reinforcement learning. The module was implemented with biologically plausible local learning rule that allows the agent to adopt quickly to the environment. To benchmark our system, we have tested the first module on Oxford-IIIT Pet dataset and the second module on the custom-made Gym environment. Our experimental results have proven that our system is able present the competitive results with deep neural network in segmentation task and adopts quickly to the environment.In recent years, deep neural networks have achieved state-of-the-art performance in a variety of recognition and segmentation tasks in medical imaging including brain tumor segmentation. We investigate that segmenting a brain tumor is facing to the imbalanced data problem where the number of pixels belonging to the background class (non tumor pixel) is much larger than the number of pixels belonging to the foreground class (tumor pixel). To address this problem, we propose a multi-task network which is formed as a cascaded structure. Our model consists of two targets, i.e., (i) effectively differentiate the brain tumor regions and (ii) estimate the brain tumor mask. The first objective is performed by our proposed contextual brain tumor detection network, which plays a role of an attention gate and focuses on the region around brain tumor only while ignoring the far neighbor background which is less correlated to the tumor. The second objective is built upon a 3D atrous residual network and under an encode-decode network in order to effectively segment both large and small objects (brain tumor). Our 3D atrous residual network is designed with a skip connection to enables the gradient from the deep layers to be directly propagated to shallow layers, thus, features of different depths are preserved and used for refining each other. In order to incorporate larger contextual information from volume MRI data, our network utilizes the 3D atrous convolution with various kernel sizes, which enlarges the receptive field of filters. Our proposed network has been evaluated on various datasets including BRATS2015, BRATS2017 and BRATS2018 datasets with both validation set and testing set. Our performance has been benchmarked by both region-based metrics and surface-based metrics. We also have conducted comparisons against state-of-the-art approaches.


Yamazaki, K. (2020). “Towards Sensorimotor Coupling of a Spiking Neural Network and Deep Reinforcement Learning for Robotics Application.” Mechanical Engineering Undergraduate Honors Theses Retrieved from