Incremental Learning Based on Data Translation and Knowledge Distillation

Recently, deep convolutional neural networks (DCNNs) have achieved remarkable results in image classification tasks. Despite convolutional networks’ great successes, their training process relies on a large amount of data prepared in advance, which is often challenging in real-world applications, such as streaming data and concept drift. For this reason, incremental learning (continual learning) has attracted increasing attention from scholars. However, incremental learning is associated with the challenge of catastrophic forgetting: the performance on previous tasks drastically degrades after learning a new task. In this paper, we propose a new strategy to alleviate catastrophic forgetting when neural networks are trained in continual domains. Specifically, two components are applied: data translation based on transfer learning and knowledge distillation. The former translates a portion of new data to reconstruct the partial data distribution of the old domain. The latter uses an old model as a teacher to guide a new model. The experimental results on three datasets have shown that our work can effectively alleviate catastrophic forgetting by a combination of the two methods aforementioned.

An Elevator Button Recognition Method Combining YOLOv5 and OCR

Fast recognition of elevator buttons is a key step for service robots toride elevators automatically. Although there are some studies in this field, noneof them can achieve real-time application due to problems such as recognitionspeed and algorithm complexity. Elevator button recognition is a comprehensiveproblem. Not only does it need to detect the position of multiple buttonsat the same time, but also needs to accurately identify the characters on eachbutton. The latest version 5 of you only look once algorithm (YOLOv5) hasthe fastest reasoning speed and can be used for detecting multiple objects inreal-time. The advantages ofYOLOv5 make it an ideal choice for detecting theposition of multiple buttons in an elevator, but it’s not good at specific wordrecognition. Optical character recognition (OCR) is a well-known techniquefor character recognition. This paper innovatively improved the YOLOv5network, integrated OCR technology, and applied them to the elevator buttonrecognition process. First, we changed the detection scale in the YOLOv5network and only maintained the detection scales of 40 ∗ 40 and 80 ∗ 80, thusimproving the overall object detection speed. Then, we put a modified OCRbranch after the YOLOv5 network to identify the numbers on the buttons.Finally, we verified this method on different datasets and compared it withother typical methods. The results show that the average recall and precisionof this method are 81.2% and 92.4%. Compared with others, the accuracyof this method has reached a very high level, but the recognition speed hasreached 0.056 s, which is far higher than other methods.

Revolutionizing the Life Sciences by Developing a Holographic Digital Mannequin

1.The need to develop a holographic digital mannequin Life processes,including high intelligence,self-organization,and homeostasis,are characterized by the biological organism in the form of self-renewal,self-replication and self-regulation,metabolism,self-repair,and self-reproduction,which are all processes of multisystem coordinated movement[1].Research in the field of life sciences is not limited to the use of advanced observational methods to reveal microscopic structures at the subcellular or molecular level.Discoveries based on these methods alone cannot characterize the dynamic processes of life at the microscopic and molecular level[2].

Combat data shift in few-shot learning with knowledge graph

Many few-shot learning approaches have been designed under the meta-learning framework, which learns from a variety of learning tasks and generalizes to new tasks. These meta-learning approaches achieve the expected performance in the scenario where all samples are drawn from the same distributions (i.i.d. observations). However, in real-world applications, few-shot learning paradigm often suffers from data shift, i.e., samples in different tasks, even in the same task, could be drawn from various data distributions. Most existing few-shot learning approaches are not designed with the consideration of data shift, and thus show downgraded performance when data distribution shifts. However, it is non-trivial to address the data shift problem in few-shot learning, due to the limited number of labeled samples in each task. Targeting at addressing this problem, we propose a novel metric-based meta-learning framework to extract task-specific representations and task-shared representations with the help of knowledge graph. The data shift within/between tasks can thus be combated by the combination of task-shared and task-specific representations. The proposed model is evaluated on popular benchmarks and two constructed new challenging datasets. The evaluation results demonstrate its remarkable performance.