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

2022 ◽  
Vol 13 (1) ◽  
pp. 1-23
Christoffer Löffler ◽  
Luca Reeb ◽  
Daniel Dzibela ◽  
Robert Marzilger ◽  
Nicolas Witt ◽  

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.

2021 ◽  
Vol 4 ◽  
Linmei Hu ◽  
Mengmei Zhang ◽  
Shaohua Li ◽  
Jinghan Shi ◽  
Chuan Shi ◽  

Knowledge Graphs (KGs) such as Freebase and YAGO have been widely adopted in a variety of NLP tasks. Representation learning of Knowledge Graphs (KGs) aims to map entities and relationships into a continuous low-dimensional vector space. Conventional KG embedding methods (such as TransE and ConvE) utilize only KG triplets and thus suffer from structure sparsity. Some recent works address this issue by incorporating auxiliary texts of entities, typically entity descriptions. However, these methods usually focus only on local consecutive word sequences, but seldom explicitly use global word co-occurrence information in a corpus. In this paper, we propose to model the whole auxiliary text corpus with a graph and present an end-to-end text-graph enhanced KG embedding model, named Teger. Specifically, we model the auxiliary texts with a heterogeneous entity-word graph (called text-graph), which entails both local and global semantic relationships among entities and words. We then apply graph convolutional networks to learn informative entity embeddings that aggregate high-order neighborhood information. These embeddings are further integrated with the KG triplet embeddings via a gating mechanism, thus enriching the KG representations and alleviating the inherent structure sparsity. Experiments on benchmark datasets show that our method significantly outperforms several state-of-the-art methods.

Cancers ◽  
2021 ◽  
Vol 13 (9) ◽  
pp. 2111
Bo-Wei Zhao ◽  
Zhu-Hong You ◽  
Lun Hu ◽  
Zhen-Hao Guo ◽  
Lei Wang ◽  

Identification of drug-target interactions (DTIs) is a significant step in the drug discovery or repositioning process. Compared with the time-consuming and labor-intensive in vivo experimental methods, the computational models can provide high-quality DTI candidates in an instant. In this study, we propose a novel method called LGDTI to predict DTIs based on large-scale graph representation learning. LGDTI can capture the local and global structural information of the graph. Specifically, the first-order neighbor information of nodes can be aggregated by the graph convolutional network (GCN); on the other hand, the high-order neighbor information of nodes can be learned by the graph embedding method called DeepWalk. Finally, the two kinds of feature are fed into the random forest classifier to train and predict potential DTIs. The results show that our method obtained area under the receiver operating characteristic curve (AUROC) of 0.9455 and area under the precision-recall curve (AUPR) of 0.9491 under 5-fold cross-validation. Moreover, we compare the presented method with some existing state-of-the-art methods. These results imply that LGDTI can efficiently and robustly capture undiscovered DTIs. Moreover, the proposed model is expected to bring new inspiration and provide novel perspectives to relevant researchers.

Nicolo Botteghi ◽  
Ruben Obbink ◽  
Daan Geijs ◽  
Mannes Poel ◽  
Beril Sirmacek ◽  

Shengsheng Qian ◽  
Jun Hu ◽  
Quan Fang ◽  
Changsheng Xu

In this article, we focus on fake news detection task and aim to automatically identify the fake news from vast amount of social media posts. To date, many approaches have been proposed to detect fake news, which includes traditional learning methods and deep learning-based models. However, there are three existing challenges: (i) How to represent social media posts effectively, since the post content is various and highly complicated; (ii) how to propose a data-driven method to increase the flexibility of the model to deal with the samples in different contexts and news backgrounds; and (iii) how to fully utilize the additional auxiliary information (the background knowledge and multi-modal information) of posts for better representation learning. To tackle the above challenges, we propose a novel Knowledge-aware Multi-modal Adaptive Graph Convolutional Networks (KMAGCN) to capture the semantic representations by jointly modeling the textual information, knowledge concepts, and visual information into a unified framework for fake news detection. We model posts as graphs and use a knowledge-aware multi-modal adaptive graph learning principal for the effective feature learning. Compared with existing methods, the proposed KMAGCN addresses challenges from three aspects: (1) It models posts as graphs to capture the non-consecutive and long-range semantic relations; (2) it proposes a novel adaptive graph convolutional network to handle the variability of graph data; and (3) it leverages textual information, knowledge concepts and visual information jointly for model learning. We have conducted extensive experiments on three public real-world datasets and superior results demonstrate the effectiveness of KMAGCN compared with other state-of-the-art algorithms.

Sophia Bano ◽  
Francisco Vasconcelos ◽  
Emmanuel Vander Poorten ◽  
Tom Vercauteren ◽  
Sebastien Ourselin ◽  

Abstract Purpose Fetoscopic laser photocoagulation is a minimally invasive surgery for the treatment of twin-to-twin transfusion syndrome (TTTS). By using a lens/fibre-optic scope, inserted into the amniotic cavity, the abnormal placental vascular anastomoses are identified and ablated to regulate blood flow to both fetuses. Limited field-of-view, occlusions due to fetus presence and low visibility make it difficult to identify all vascular anastomoses. Automatic computer-assisted techniques may provide better understanding of the anatomical structure during surgery for risk-free laser photocoagulation and may facilitate in improving mosaics from fetoscopic videos. Methods We propose FetNet, a combined convolutional neural network (CNN) and long short-term memory (LSTM) recurrent neural network architecture for the spatio-temporal identification of fetoscopic events. We adapt an existing CNN architecture for spatial feature extraction and integrated it with the LSTM network for end-to-end spatio-temporal inference. We introduce differential learning rates during the model training to effectively utilising the pre-trained CNN weights. This may support computer-assisted interventions (CAI) during fetoscopic laser photocoagulation. Results We perform quantitative evaluation of our method using 7 in vivo fetoscopic videos captured from different human TTTS cases. The total duration of these videos was 5551 s (138,780 frames). To test the robustness of the proposed approach, we perform 7-fold cross-validation where each video is treated as a hold-out or test set and training is performed using the remaining videos. Conclusion FetNet achieved superior performance compared to the existing CNN-based methods and provided improved inference because of the spatio-temporal information modelling. Online testing of FetNet, using a Tesla V100-DGXS-32GB GPU, achieved a frame rate of 114 fps. These results show that our method could potentially provide a real-time solution for CAI and automating occlusion and photocoagulation identification during fetoscopic procedures.

Jay Ryan U. Roldan ◽  
Dejan Milutinović ◽  
Zhi Li ◽  
Jacob Rosen

In this paper, we propose a quantitative approach based on identifying hand trajectory dissimilarities through the use of a multidimensional scaling (MDS) analysis. A high-rate motion capture system is used to gather three-dimensional (3D) trajectory data of healthy and stroke-impacted hemiparetic subjects. The mutual dissimilarity between any two trajectories is measured by the area between them. This area is used as a dissimilarity variable to create an MDS map. The map reveals a structure for measuring the difference and variability of individual trajectories and their groups. The results suggest that the recovery of hemiparetic subjects can be quantified by comparing the difference and variability of their individual MDS map points to the points from the cluster of healthy subject trajectories. Within the MDS map, we can identify fully recovered patients, those who are only functionally recovered, and those who are either in an early phase of, or are nonresponsive to the therapy.

Hongfei Xu ◽  
Deyi Xiong ◽  
Josef van Genabith ◽  
Qiuhui Liu

Existing Neural Machine Translation (NMT) systems are generally trained on a large amount of sentence-level parallel data, and during prediction sentences are independently translated, ignoring cross-sentence contextual information. This leads to inconsistency between translated sentences. In order to address this issue, context-aware models have been proposed. However, document-level parallel data constitutes only a small part of the parallel data available, and many approaches build context-aware models based on a pre-trained frozen sentence-level translation model in a two-step training manner. The computational cost of these approaches is usually high. In this paper, we propose to make the most of layers pre-trained on sentence-level data in contextual representation learning, reusing representations from the sentence-level Transformer and significantly reducing the cost of incorporating contexts in translation. We find that representations from shallow layers of a pre-trained sentence-level encoder play a vital role in source context encoding, and propose to perform source context encoding upon weighted combinations of pre-trained encoder layers' outputs. Instead of separately performing source context and input encoding, we propose to iteratively and jointly encode the source input and its contexts and to generate input-aware context representations with a cross-attention layer and a gating mechanism, which resets irrelevant information in context encoding. Our context-aware Transformer model outperforms the recent CADec [Voita et al., 2019c] on the English-Russian subtitle data and is about twice as fast in training and decoding.

2021 ◽  
Vol 2021 ◽  
pp. 1-10
Jifeng Guo ◽  
Zhiqi Pang ◽  
Wenbo Sun ◽  
Shi Li ◽  
Yu Chen

Active learning aims to select the most valuable unlabelled samples for annotation. In this paper, we propose a redundancy removal adversarial active learning (RRAAL) method based on norm online uncertainty indicator, which selects samples based on their distribution, uncertainty, and redundancy. RRAAL includes a representation generator, state discriminator, and redundancy removal module (RRM). The purpose of the representation generator is to learn the feature representation of a sample, and the state discriminator predicts the state of the feature vector after concatenation. We added a sample discriminator to the representation generator to improve the representation learning ability of the generator and designed a norm online uncertainty indicator (Norm-OUI) to provide a more accurate uncertainty score for the state discriminator. In addition, we designed an RRM based on a greedy algorithm to reduce the number of redundant samples in the labelled pool. The experimental results on four datasets show that the state discriminator, Norm-OUI, and RRM can improve the performance of RRAAL, and RRAAL outperforms the previous state-of-the-art active learning methods.

2020 ◽  
Jing Qian ◽  
Gangmin Li ◽  
Katie Atkinson ◽  
Yong Yue

Knowledge representation learning (KRL) aims at encoding components of a knowledge graph (KG) into a low-dimensional continuous space, which has brought considerable successes in applying deep learning to graph embedding. Most famous KGs contain only positive instances for space efficiency. Typical KRL techniques, especially translational distance-based models, are trained through discriminating positive and negative samples. Thus, negative sampling is unquestionably a non-trivial step in KG embedding. The quality of generated negative samples can directly influence the performance of final knowledge representations in downstream tasks, such as link prediction and triple classification. This review summarizes current negative sampling methods in KRL and we categorize them into three sorts, fixed distribution-based, generative adversarial net (GAN)-based and cluster sampling. Based on this categorization we discuss the most prevalent existing approaches and their characteristics.

Sign in / Sign up

Export Citation Format

Share Document