Multilayer Batch Learning Growing Neural Gas for Learning Multiscale Topologies

Hierarchical topological structure learning methods are expected to be developed in the field of data mining for extracting multiscale topological structures from an unknown dataset. However, most methods require user-defined parameters, and it is difficult for users to determine these parameters and effectively utilize the method. In this paper, we propose a new parameter-less hierarchical topological structure learning method based on growing neural gas (GNG). First, we propose batch learning GNG (BL-GNG) to improve the learning convergence and reduce the user-designed parameters in GNG. BL-GNG uses an objective function based on fuzzy C-means to improve the learning convergence. Next, we propose multilayer BL-GNG (MBL-GNG), which is a parameter-less unsupervised learning algorithm based on hierarchical topological structure learning. In MBL-GNG, the input data of each layer uses parent nodes to learn more abstract topological structures from the dataset. Furthermore, MBL-GNG can automatically determine the number of nodes and layers according to the data distribution. Finally, we conducted several experiments to evaluate our proposed method by comparing it with other hierarchical approaches and discuss the effectiveness of our proposed method.

Continual learning classification method with single-label memory cells based on the intelligent mechanism of the biological immune system

The traditional batch learning classification methods need to obtain all kinds of data once before training. This makes them unable to recognize the data from the unseen types and cannot continuously enhance their classification ability through learning the testing data in the testing process, because they lack continual learning ability. Inspired by the continual learning mechanism of the biological immune system (BIS), this paper proposed a continual learning classification method with single-label memory cells (S-CLCM). The type of testing data is identified by memory cells, and the data type from unseen types is determined by an affinity threshold. New memory cells are cultivated continuously by learning the testing data to enhance S-CLCM’s classification ability gradually. Every memory cell has the same size and a unique type. It becomes a standard batch learning classification method or a standard clustering method under certain conditions. Take the experiments on twenty benchmark datasets to estimate its classification performance and possible superiority. Results show S-CLCM has good performance when it becomes a standard batch learning classification method, and S-CLCM is superior to the other classical classification algorithms when the data from unseen types or new labeled data appear during the testing process. It can improve the classification accuracy by up to 33%, and by at least 14%.

The State of Lifelong Learning in Service Robots:

Service robots are appearing more and more in our daily life. The development of service robots combines multiple fields of research, from object perception to object manipulation. The state-of-the-art continues to improve to make a proper coupling between object perception and manipulation. This coupling is necessary for service robots not only to perform various tasks in a reasonable amount of time but also to continually adapt to new environments and safely interact with non-expert human users. Nowadays, robots are able to recognize various objects, and quickly plan a collision-free trajectory to grasp a target object in predefined settings. Besides, in most of the cases, there is a reliance on large amounts of training data. Therefore, the knowledge of such robots is fixed after the training phase, and any changes in the environment require complicated, time-consuming, and expensive robot re-programming by human experts. Therefore, these approaches are still too rigid for real-life applications in unstructured environments, where a significant portion of the environment is unknown and cannot be directly sensed or controlled. In such environments, no matter how extensive the training data used for batch learning, a robot will always face new objects. Therefore, apart from batch learning, the robot should be able to continually learn about new object categories and grasp affordances from very few training examples on-site. Moreover, apart from robot self-learning, non-expert users could interactively guide the process of experience acquisition by teaching new concepts, or by correcting insufficient or erroneous concepts. In this way, the robot will constantly learn how to help humans in everyday tasks by gaining more and more experiences without the need for re-programming. In this paper, we review a set of previously published works and discuss advances in service robots from object perception to complex object manipulation and shed light on the current challenges and bottlenecks.

Time-series trend of pandemic SARS-CoV-2 variants visualized using batch-learning self-organizing map for oligonucleotide compositions

To confront the global threat of coronavirus disease 2019, a massive number of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) genome sequences have been decoded, with the results promptly released through the GISAID database. Based on variant types, eight clades have already been defined in GISAID, but the diversity can be far greater. Owing to the explosive increase in available sequences, it is important to develop new technologies that can easily grasp the whole picture of the big-sequence data and support efficient knowledge discovery. An ability to efficiently clarify the detailed time-series changes in genome-wide mutation patterns will enable us to promptly identify and characterize dangerous variants that rapidly increase their population frequency. Here, we collectively analyzed over 150,000 SARS-CoV-2 genomes to understand their overall features and time-dependent changes using a batch-learning self-organizing map (BLSOM) for oligonucleotide composition, which is an unsupervised machine learning method. BLSOM can separate clades defined by GISAID with high precision, and each clade is subdivided into clusters, which shows a differential increase/decrease pattern based on geographic region and time. This allowed us to identify prevalent strains in each region and to show the commonality and diversity of the prevalent strains. Comprehensive characterization of the oligonucleotide composition of SARS-CoV-2 and elucidation of time-series trends of the population frequency of variants can clarify the viral adaptation processes after invasion into the human population and the time-dependent trend of prevalent epidemic strains across various regions, such as continents.