In this paper, we propose a novel pretraining-based encoder-decoder framework, which can generate the output sequence based on input in two-stage manner. For encoder of our model, encode into context representations using BERT. decoder, there are two stages first stage, use Transformer-based decoder to draft sequence. second mask each word and feed it BERT, then by combining representation generated predict refined for masked position. To best knowledge, approach is method applies BERT text...

Hierarchical text classification is an essential yet challenging subtask of multi-label with a taxonomic hierarchy. Existing methods have difficulties in modeling the hierarchical label structure global view. Furthermore, they cannot make full use mutual interactions between feature space and space. In this paper, we formulate hierarchy as directed graph introduce hierarchy-aware encoders for dependencies. Based on encoder, propose novel end-to-end model (HiAGM) two variants. A attention...

Training machine learning (ML) models with large datasets can incur significant resource contention on shared clusters. This training typically involves many iterations that continually improve the quality of model. Yet in exploratory settings, better be obtained faster by directing resources to jobs most potential for improvement. We describe SLAQ, a cluster scheduling system approximate ML aims maximize overall job quality.

Personalized outfit recommendation, which aims to recommend the outfits a given user according his/her preference, has gained increasing research attention due its economic value. Nevertheless, majority of existing methods mainly focus on improving recommendation effectiveness, while overlooking efficiency. Inspired by this, we devise novel bi-directional heterogeneous graph hashing scheme, called BiHGH, towards efficient personalized recommendation. In particular, this scheme consists three...

The rapidly growing size of data and complexity analytics present new challenges for large-scale processing systems. Modern systems keep partitions in memory pipelined operators, persist across stages with wide dependencies on disks fault tolerance. While can often scale well by splitting jobs into smaller tasks better parallelism, all-to-all transfer---called shuffle operations---become the scaling bottleneck when running many small multi-stage jobs. Our key observation is that this due to...

In this work, we focus on complex question semantic parsing and propose a novel Hierarchical Semantic Parsing (HSP) method, which utilizes the decompositionality of questions for parsing. Our model is designed within three-stage architecture based idea decomposition-integration. first stage, decomposer decomposes into sequence sub-questions. second design an information extractor to derive type predicate these questions. last integrate generated from previous stages generate logical form...

In this paper, we propose a novel pretraining-based encoder-decoder framework, which can generate the output sequence based on input in two-stage manner. For encoder of our model, encode into context representations using BERT. decoder, there are two stages first stage, use Transformer-based decoder to draft sequence. second mask each word and feed it BERT, then by combining representation generated predict refined for masked position. To best knowledge, approach is method applies BERT text...

Abstract The extended Kalman filter (EKF) is a widely used method in navigation applications. EKF suffers from noise covariance uncertainty, potentially causing it to perform poorly practice. This paper attempts suppress the unfavorable effect of uncertainty framework reinforcement learning. proposed state estimation algorithm combines and Q‐learning method, where adaptation strategy designed based on Q‐values, leading gradual improvement performance. resultant called (QLEKF), which less...

Multimodal dialog system has attracted increasing attention from both academia and industry over recent years. Although existing methods have achieved some progress, they are still confronted with challenges in the aspect of question understanding (i.e., user intention comprehension). In this paper, we present a relational graph-based context-aware scheme, which enhances comprehension local to global. Specifically, first utilize multiple attribute matrices as guidance information fully...

Partial person re-identification (ReID) aims to solve the problem of image spatial misalignment due occlusions or out-of-views. Despite significant progress through introduction additional information, such as human pose landmarks, mask maps, and partial ReID remains challenging noisy keypoints impressionable pedestrian representations. To address these issues, we propose a unified attribute-guided collaborative learning scheme for ReID. Specifically, introduce an adaptive threshold-guided...

Jamming is a big threat to the survival of radar system. Therefore, recognition jamming signal type part countermeasure. Recently, convolutional neural networks (CNNs) have shown their effectiveness in processing, including recognition. However, most existing CNN methods do not regard as complex value signal. In this study, complex-valued (CV-CNN) investigated fully explore inherent characteristics signal, and we find that can obtain better accuracy using method compared with real-valued...

Fine-grained Entity Typing (FET) has made great progress based on distant supervision but still suffers from label noise. Existing FET noise learning methods rely prediction distributions in an instance-independent manner, which causes the problem of confirmation bias. In this work, we propose a clustering-based loss correction framework named Feature Cluster Loss Correction (FCLC), to address these two problems. FCLC first train coarse backbone model as feature extractor and estimator. is...

The learning problem of ranking arises in many tasks, including the question answering, information retrieval, and movie recommendation. In these ordering answers, documents or movies returned is a critical aspect system. Recently, deep approaches have gained lot attention from research community industry for their ability to automatically learn optimal feature representation given task. We aim solve answer practical answering system with approaches. this paper, we define composite questions...

Training machine learning (ML) models with large datasets can incur significant resource contention on shared clusters. This training typically involves many iterations that continually improve the quality of model. Yet in exploratory settings, better be obtained faster by directing resources to jobs most potential for improvement. We describe SLAQ, a cluster scheduling system approximate ML aims maximize overall job quality. When allocating resources, SLAQ explores quality-runtime...

To solve the problem of false tracks generated by breakdowns and clutter in point-target tracking polar coordinates, a fusion algorithm based on converted measurement Kalman filter random matrix expansion is proposed. The (CMKF) transforms coordinate data target at current time into Cartesian coordinates without bias. Based linear measurements states, position extended group was predicted updated using matrix, its track drawn combining nearest neighbors to realize size, shape azimuth target....

A series of inorganic acid posttreatmented carbon nitride were synthesized to regulate the electronic structure, surface properties and then improve photocatalytic activity. Carbon displayed a thinner smaller...

Graph Convolutional Networks (GCNs) are powerful representation learning methods for non-Euclidean data. Compared with the Euclidean data, labeling data is more expensive. Meanwhile, most existing GCNs only utilize few labeled but ignore of unlabeled To address this issue, we design a novel end-to-end Iterative Feature Clustering (IFC-GCN) that enhances standard GCN an (IFC) module. The proposed IFC module constrains node features iteratively based on predicted pseudo labels and feature...