Deep Siamese Metric Learning: A Highly Scalable Approach to Searching Unordered Sets of Trajectories

This work proposes metric learning for fast similarity-based scene retrieval of unstructured ensembles of trajectory data from large databases. We present a novel representation learning approach using Siamese Metric Learning that approximates a distance preserving low-dimensional representation and that learns to estimate reasonable solutions to the assignment problem. To this end, we employ a Temporal Convolutional Network architecture that we extend with a gating mechanism to enable learning from sparse data, leading to solutions to the assignment problem exhibiting varying degrees of sparsity. Our experimental results on professional soccer tracking data provides insights on learned features and embeddings, as well as on generalization, sensitivity, and network architectural considerations. Our low approximation errors for learned representations and the interactive performance with retrieval times several magnitudes smaller shows that we outperform previous state of the art.

Iterative Compilation Optimization Based on Metric Learning and Collaborative Filtering

Pass selection and phase ordering are two critical compiler auto-tuning problems. Traditional heuristic methods cannot effectively address these NP-hard problems especially given the increasing number of compiler passes and diverse hardware architectures. Recent research efforts have attempted to address these problems through machine learning. However, the large search space of candidate pass sequences, the large numbers of redundant and irrelevant features, and the lack of training program instances make it difficult to learn models well. Several methods have tried to use expert knowledge to simplify the problems, such as using only the compiler passes or subsequences in the standard levels (e.g., -O1, -O2, and -O3) provided by compiler designers. However, these methods ignore other useful compiler passes that are not contained in the standard levels. Principal component analysis (PCA) and exploratory factor analysis (EFA) have been utilized to reduce the redundancy of feature data. However, these unsupervised methods retain all the information irrelevant to the performance of compilation optimization, which may mislead the subsequent model learning. To solve these problems, we propose a compiler pass selection and phase ordering approach, called Iterative Compilation based on Metric learning and Collaborative filtering (ICMC) . First, we propose a data-driven method to construct pass subsequences according to the observed collaborative interactions and dependency among passes on a given program set. Therefore, we can make use of all available compiler passes and prune the search space. Then, a supervised metric learning method is utilized to retain useful feature information for compilation optimization while removing both the irrelevant and the redundant information. Based on the learned similarity metric, a neighborhood-based collaborative filtering method is employed to iteratively recommend a few superior compiler passes for each target program. Last, an iterative data enhancement method is designed to alleviate the problem of lacking training program instances and to enhance the performance of iterative pass recommendations. The experimental results using the LLVM compiler on all 32 cBench programs show the following: (1) ICMC significantly outperforms several state-of-the-art compiler phase ordering methods, (2) it performs the same or better than the standard level -O3 on all the test programs, and (3) it can reach an average performance speedup of 1.20 (up to 1.46) compared with the standard level -O3.

HyperSoRec: Exploiting Hyperbolic User and Item Representations with Multiple Aspects for Social-aware Recommendation

Social recommendation has achieved great success in many domains including e-commerce and location-based social networks. Existing methods usually explore the user-item interactions or user-user connections to predict users’ preference behaviors. However, they usually learn both user and item representations in Euclidean space, which has large limitations for exploring the latent hierarchical property in the data. In this article, we study a novel problem of hyperbolic social recommendation, where we aim to learn the compact but strong representations for both users and items. Meanwhile, this work also addresses two critical domain-issues, which are under-explored. First, users often make trade-offs with multiple underlying aspect factors to make decisions during their interactions with items. Second, users generally build connections with others in terms of different aspects, which produces different influences with aspects in social network. To this end, we propose a novel graph neural network (GNN) framework with multiple aspect learning, namely, HyperSoRec. Specifically, we first embed all users, items, and aspects into hyperbolic space with superior representations to ensure their hierarchical properties. Then, we adapt a GNN with novel multi-aspect message-passing-receiving mechanism to capture different influences among users. Next, to characterize the multi-aspect interactions of users on items, we propose an adaptive hyperbolic metric learning method by introducing learnable interactive relations among different aspects. Finally, we utilize the hyperbolic translational distance to measure the plausibility in each user-item pair for recommendation. Experimental results on two public datasets clearly demonstrate that our HyperSoRec not only achieves significant improvement for recommendation performance but also shows better representation ability in hyperbolic space with strong robustness and reliability.

Sparse Representation-Based Discriminative Metric Learning for Brain MRI Image Retrieval

Magnetic resonance imaging (MRI) can have a good diagnostic function for important organs and parts of the body. MRI technology has become a common and important disease detection technology. At the same time, medical imaging data is increasing at an explosive rate. Retrieving similar medical images from a huge database is of great significance to doctors’ auxiliary diagnosis and treatment. In this paper, combining the advantages of sparse representation and metric learning, a sparse representation-based discriminative metric learning (SRDML) approach is proposed for medical image retrieval of brain MRI. The SRDML approach uses a sparse representation framework to learn robust feature representation of brain MRI, and uses metric learning to project new features into the metric space with matching discrimination. In such a metric space, the optimal similarity measure is obtained by using the local constraints of atoms and the pairwise constraints of coding coefficients, so that the distance between similar images is less than the given threshold, and the distance between dissimilar images is greater than another given threshold. The experiments are designed and tested on the brain MRI dataset created by Chang. Experimental results show that the SRDML approach can obtain satisfactory retrieval performance and achieve accurate brain MRI image retrieval.

Detection of Children’s Molars Based on Noise Filter

Pit and fissure sealant is for children aged seven to twelve years to prevent molars from becoming caries. In this paper, we propose a new detection framework to identify whether children need pit and fissure sealing. We divide the framework into two parts: molar detection and molar classification. According to the characteristics of teeth, we propose to use the clustering method to filter the bounding box in the object detection part. In the region divided by clustering, we only keep one detection frame in the same category. In the classification part, we propose a noise filtering layer based on wavelet transform for feature extraction. We map the training samples to another space in the training process based on metric learning to increase the distance between categories and improve the accuracy of classification.

Dissecting Cellular Heterogeneity Based on Network Denoising of scRNA-seq Using Local Scaling Self-Diffusion

Identifying the phenotypes and interactions of various cells is the primary objective in cellular heterogeneity dissection. A key step of this methodology is to perform unsupervised clustering, which, however, often suffers challenges of the high level of noise, as well as redundant information. To overcome the limitations, we proposed self-diffusion on local scaling affinity (LSSD) to enhance cell similarities’ metric learning for dissecting cellular heterogeneity. Local scaling infers the self-tuning of cell-to-cell distances that are used to construct cell affinity. Our approach implements the self-diffusion process by propagating the affinity matrices to further improve the cell similarities for the downstream clustering analysis. To demonstrate the effectiveness and usefulness, we applied LSSD on two simulated and four real scRNA-seq datasets. Comparing with other single-cell clustering methods, our approach demonstrates much better clustering performance, and cell types identified on colorectal tumors reveal strongly biological interpretability.

Similar Pair-free Partial Label Metric Learning

Partial label learning is a new weak- ly supervised learning framework. In this frame- work, the real category label of a training sample is usually concealed in a set of candidate labels, which will lead to lower accuracy of learning al- gorithms compared with traditional strong super- vised cases. Recently, it has been found that met- ric learning technology can be used to improve the accuracy of partial label learning algorithm- s. However, because it is difficult to ascertain similar pairs from training samples, at present there are few metric learning algorithms for par- tial label learning framework. In view of this, this paper proposes a similar pair-free partial la- bel metric learning algorithm. The main idea of the algorithm is to define two probability distri- butions on the training samples, i.e., the proba- bility distribution determined by the distance of sample pairs and the probability distribution de- termined by the similarity of candidate label set of sample pairs, and then the metric matrix is ob- tained via minimizing the KL divergence of the two probability distributions. The experimental results on several real-world partial label dataset- s show that the proposed algorithm can improve the accuracy of k-nearest neighbor partial label learning algorithm (PL-KNN) better than the ex- isting partial label metric learning algorithms, up to 8 percentage points.

Pedestrian Reidentification Based on Feature Fusion and Metric Learning

The pedestrian recognition in public environment is influenced by the pedestrian environment and the dynamic characteristic boundary factors, so it is easy to produce the tracking error. In order to improve the ability of pedestrian re-identification in public environment, we need to carry out feature fusion and metric learning, and propose pedestrian re-identification based on feature fusion and metric learning. The geometric grid area model of pedestrian recognition in public environment is constructed, the method of fuzzy dynamic feature segmentation is used to reconstruct the dynamic boundary feature point of pedestrian recognition in public environment, the method of bottom-up modeling is used to design the dynamic area grid model of pedestrian recognition in public environment, the design of dynamic area grid model is three-dimensional grid area, the grayscale pixel set of pedestrian recognition dynamic constraint under public environment is extracted, the boundary feature fusion is carried out according to the distribution intensity of grayscale, the image fusion and enhancement information processing of pedestrian recognition under public environment, and the method of 3D dynamic constraint is used to realize the local motion planning of pedestrian recognition under public environment, and the recognition feature fusion and learning of pedestrian recognition under public environment is realized according to the result of contour segmentation. The simulation results show that the method is used for pedestrian recognition again in public environment, and the fuzzy judgment ability of pedestrian dynamic edge features is strong, which makes the error controlled below 10 mm, and the fluctuation of pedestrian recognition again is more stable, the recognition accuracy is higher and the robustness is better.