Tool wear is a critical factor affecting the quality and efficiency of machining. In-process tool wear prediction is of great importance for unmanned factories. However, its industrial application faces many challenges, especially in terms of complicated cutting conditions and limited accuracy. In this article, a novel approach of physics-informed liquid neural networks (PILNNs) is proposed for tool wear prediction under varying cutting conditions. The model extracts multidimensional features related to tool wear from the acceleration signals and inputs them into a multilayer perceptual machine. Physical constraints are constructed through a tool wear model to provide physical priori knowledge for the neural networks. Liquid time-constant neurons are constructed by the neural circuit policies method to build the liquid neural networks (LNNs). In this case, the PILNNs merge both physical laws and LNN structures, which significantly enhances the generalization performance towards tool wear prediction. The performance of the proposed model is evaluated by the data obtained under varying cutting conditions, as well as the comparison against traditional physics-informed neural networks and other machine learning models through the same dataset. The results illustrate that the proposed model shows high accuracy and stability in terms of tool wear prediction under varying cutting conditions, which is also superior to the traditional machine learning models.

Topic modeling is a powerful unsupervised tool for knowledge discovery.However, existing work struggles with generating limited-quality topics that are uninformative and incoherent, which hinders interpretable insights from managing textual data.In this paper, we improve the original variational autoencoder framework by incorporating contextual and graph information to address the above issues.First, the encoder utilizes topic fusion techniques to combine contextual and bag-of-words information well, and meanwhile exploits the constraints of topic alignment and topic sharpening to generate informative topics.Second, we develop a simple word co-occurrence graph information fusion strategy that efficiently increases topic coherence.On three benchmark datasets, our new framework generates more coherent and diverse topics compared to various baselines, and achieves strong performance on both automatic and manual evaluations.We make our code available for reproduction 1 .

Aiming at mining high quality topics by accumulating and utilizing semantic knowledge for a stream of documents, lifelong topic modeling (LTM) has attracted more and more attentions recently. However, the permutation of topics may change over time, resulting in asemantic misalignmentbetween the topic representations of document chunks across the stream. Such a misalignment deteriorates the model performances of various downstream tasks, while it has been overlooked by the existing lifelong topic models. Towards addressing the misalignment of semantics, we formulate LTM as a problem of non-negative matrix tri-factorization (NMTF) and propose a consolidation framework (i.e., NMTF-LTM) to enforce an alignment in a mapped topic space. In addition, a distributed parallel algorithm, namely PNMTF-LTM, is developed to meet the real-time requirement for large-scale stream processing. Empirical results show that our method can not only obtain a superior alignment of semantics without loss of topic quality, but also achieve effective speedup when deployed to a high performance computing cluster.

Self-supervised learning has shown very promising results for monocular depth estimation. Scene structure and local details both are significant clues for high-quality depth estimation. Recent works suffer from the lack of explicit modeling of scene structure and proper handling of details information, which leads to a performance bottleneck and blurry artefacts in predicted results. In this paper, we propose the Channel-wise Attention-based Depth Estimation Network (CADepth-Net) with two effective contributions: 1) The structure perception module employs the self-attention mechanism to capture long-range dependencies and aggregates discriminative features in channel dimensions, explicitly enhances the perception of scene structure, obtains the better scene understanding and rich feature representation. 2) The detail emphasis module re-calibrates channel-wise feature maps and selectively emphasizes the informative features, aiming to highlight crucial local details information and fuse different level features more efficiently, resulting in more precise and sharper depth prediction. Furthermore, the extensive experiments validate the effectiveness of our method and show that our model achieves the state-of-the-art results on the KITTI benchmark and Make3D datasets.

Shadow removal from a single image is a challenging problem, whose solution is proposed in this paper using 3D intensity surface modeling. Due to the high-order textural content in the original images, a direct modeling of the intensity surface of shadow image is difficult. In this paper, image decomposition technology is used as an edge-preserving filter to remove the textural detail while keeping the local-smoothness pattern of image intensity surface. Using 3D modeling, a proper intensity surface of illumination in shadow region can be obtained based on that corresponding to the same texture in the non-shadow one. Thus, the intensity surface of shadow region can be compensated with a respective shadow-removal. Experimental results demonstrate the effectiveness of the proposed approach in the aspect of single-image shadow removal. In contrast to the alternative methods, it is not limited by additional assumptions or conditions; moreover, it can deal with the non-uniform and curved surface shadows, and is applicable to the shadow regions consisting of different types of textures.