Multi-Scale Batch-Learning Growing Neural Gas Efficiently for Dynamic Data Distributions

Abstract

Growing neural gas (GNG) has many applications, including topology preservation, feature extraction, dynamic adaptation, clustering, and dimensionality reduction. These methods have broad applicability in extracting the topological structure of 3D point clouds, enabling unsupervised motion estimation, and depicting objects within a scene. Furthermore, multi-scale batch-learning GNG (MS-BL-GNG) has improved learning convergence. However, it is only implemented on static or stationary datasets, and adapting to dynamic data remains difficult. Similarly, the learning rate cannot be increased if new nodes are added to the existing network after accumulating errors in the sampling data. Next, we propose a new growth approach that, when applied to MS-BL-GNG, significantly increases the learning speed and adaptability of dynamic data distribution input patterns. This method immediately adds data samples as new nodes to existing networks. The probability of adding a new node is determined by the distance between the first, second, and third closest nodes. We applied our method for monitoring a moving object at its pace to demonstrate the usefulness of the proposed model. In addition, optimization methods are used such that processing can be performed in real-time.

Distributed Batch Learning of Growing Neural Gas for Quick and Efficient Clustering

“…Moreover, both methods add a node in each epoch or batch, thus reducing the learning speed. To accelerate network growth, Fernando et al [3] propose using a "add-if-silent strategy" method on top of the MS-BL-GNG method to insert nodes into the network in each iteration (FastMS-BL-GNG). With this implementation, FastMS-BL-GNG is able to operate in dynamic or non-stationary data distributions.…”

“…Therefore, we introduce batch learning with matrix computation to accelerate learning. There are various GNG batch learning methods [3,12,13]. One method is to use FCM to calculate the membership of each network node to each batch data [12].…”

“…We compare the performance of DBL-GNG with standard GNG, GNG-U [33], FCM-BL-GNG [12], MS-BL-GNG [14], and FastMS-BL-GNG [3]. The quantization error and total calculation time are shown in Table 4 and Table 5, respectively.…”

“…Recently, various types of unsupervised learning methods [1][2][3] have been applied to data mining tasks with the rapid progress of Society 5.0, Digital Twin, and cyber-physical systems [4][5][6]. For example, we need an online real-time processing system to deal with the huge size of 3D distance data measured by LiDAR in the case of self-driving cars.…”

“…It divides the dataset into multiple mini-batches for learning. Batch learning has also recently been applied to GNG to obtain topological maps from datasets [3,[12][13][14]. Multiple past studies have indeed shown that batch learning performs better than standard GNG in terms of learning stability and speed.…”