Pre-trained language models (PLMs) like BERT have made great progress in NLP. News articles usually contain rich textual information, and PLMs have the potentials to enhance news text modeling for various intelligent news applications like news recommendation and retrieval. However, most existing PLMs are in huge size with hundreds of millions of parameters. Many online news applications need to serve millions of users with low latency tolerance, which poses huge challenges to incorporating PLMs in these scenarios. Knowledge distillation techniques can compress a large PLM into a much smaller one and meanwhile keeps good performance. However, existing language models are pre-trained and distilled on general corpus like Wikipedia, which has some gaps with the news domain and may be suboptimal for news intelligence. In this paper, we propose NewsBERT, which can distill PLMs for efficient and effective news intelligence. In our approach, we design a teacher-student joint learning and distillation framework to collaboratively learn both teacher and student models, where the student model can learn from the learning experience of the teacher model. In addition, we propose a momentum distillation method by incorporating the gradients of teacher model into the update of student model to better transfer useful knowledge learned by the teacher model. Extensive experiments on two real-world datasets with three tasks show that NewsBERT can effectively improve the model performance in various intelligent news applications with much smaller models.
As hashing becomes an increasingly appealing technique for large-scale image retrieval, multi-label hashing is also attracting more attention for the ability to exploit multi-level semantic contents. In this paper, we propose a novel deep hashing method for scalable multi-label image search. Unlike existing approaches with conventional objectives such as contrast and triplet losses, we employ a rank list, rather than pairs or triplets, to provide sufficient global supervision information for all the samples. Specifically, a new rank-consistency objective is applied to align the similarity orders from two spaces, the original space and the hamming space. A powerful loss function is designed to penalize the samples whose semantic similarity and hamming distance are mismatched in two spaces. Besides, a multi-label softmax cross-entropy loss is presented to enhance the discriminative power with a concise formulation of the derivative function. In order to manipulate the neighborhood structure of the samples with different labels, we design a multi-label clustering loss to cluster the hashing vectors of the samples with the same labels by reducing the distances between the samples and their multiple corresponding class centers. The state-of-the-art experimental results achieved on three public multi-label datasets, MIRFLICKR-25K, IAPRTC12 and NUS-WIDE, demonstrate the effectiveness of the proposed method.
Representation learning for graphs enables the application of standard machine learning algorithms and data analysis tools to graph data. Replacing discrete unordered objects such as graph nodes by real-valued vectors is at the heart of many approaches to learning from graph data. Such vector representations, or embeddings, capture the discrete relationships in the original data by representing nodes as vectors in a high-dimensional space. In most applications graphs model the relationship between real-life objects and often nodes contain valuable meta-information about the original objects. While being a powerful machine learning tool, embeddings are not able to preserve such node attributes. We address this shortcoming and consider the problem of learning discrete node embeddings such that the coordinates of the node vector representations are graph nodes. This opens the door to designing interpretable machine learning algorithms for graphs as all attributes originally present in the nodes are preserved. We present a framework for coordinated local graph neighborhood sampling (COLOGNE) such that each node is represented by a fixed number of graph nodes, together with their attributes. Individual samples are coordinated and they preserve the similarity between node neighborhoods. We consider different notions of similarity for which we design scalable algorithms. We show theoretical results for all proposed algorithms. Experiments on benchmark graphs evaluate the quality of the designed embeddings and demonstrate how the proposed embeddings can be used in training interpretable machine learning algorithms for graph data.
The process of information fusion needs to deal with a large number of uncertain information with multi-source, heterogeneity, inaccuracy, unreliability, and incompleteness. In practical engineering applications, Dempster-Shafer evidence theory is widely used in multi-source information fusion owing to its effectiveness in data fusion. Information sources have an important impact on multi-source information fusion in an environment of complex, unstable, uncertain, and incomplete characteristics. To address multi-source information fusion problem, this paper considers the situation of uncertain information modeling from the closed world to the open world assumption and studies the generation of basic probability assignment (BPA) with incomplete information. In this paper, a new method is proposed to generate generalized basic probability assignment (GBPA) based on the triangular fuzzy number model under the open world assumption. The proposed method can not only be used in different complex environments simply and flexibly, but also have less information loss in information processing. Finally, a series of comprehensive experiments basing on the UCI data sets are used to verify the rationality and superiority of the proposed method.
Matching information across image and text modalities is a fundamental challenge for many applications that involve both vision and natural language processing. The objective is to find efficient similarity metrics to compare the similarity between visual and textual information. Existing approaches mainly match the local visual objects and the sentence words in a shared space with attention mechanisms. The matching performance is still limited because the similarity computation is based on simple comparisons of the matching features, ignoring the characteristics of their distribution in the data. In this paper, we address this limitation with an efficient learning objective that considers the discriminative feature distributions between the visual objects and sentence words. Specifically, we propose a novel Adversarial Discriminative Domain Regularization (ADDR) learning framework, beyond the paradigm metric learning objective, to construct a set of discriminative data domains within each image-text pairs. Our approach can generally improve the learning efficiency and the performance of existing metrics learning frameworks by regulating the distribution of the hidden space between the matching pairs. The experimental results show that this new approach significantly improves the overall performance of several popular cross-modal matching techniques (SCAN, VSRN, BFAN) on the MS-COCO and Flickr30K benchmarks.
Acronyms are the short forms of longer phrases and they are frequently used in writing, especially scholarly writing, to save space and facilitate the communication of information. As such, every text understanding tool should be capable of recognizing acronyms in text (i.e., acronym identification) and also finding their correct meaning (i.e., acronym disambiguation). As most of the prior works on these tasks are restricted to the biomedical domain and use unsupervised methods or models trained on limited datasets, they fail to perform well for scientific document understanding. To push forward research in this direction, we have organized two shared task for acronym identification and acronym disambiguation in scientific documents, named AI@SDU and AD@SDU, respectively. The two shared tasks have attracted 52 and 43 participants, respectively. While the submitted systems make substantial improvements compared to the existing baselines, there are still far from the human-level performance. This paper reviews the two shared tasks and the prominent participating systems for each of them.
In this paper, we introduce a new neural network (NN) structure, multi-mode reservoir computing (Multi-Mode RC). It inherits the dynamic mechanism of RC and processes the forward path and loss optimization of the NN using tensor as the underlying data format. Multi-Mode RC exhibits less complexity compared with conventional RC structures (e.g. single-mode RC) with comparable generalization performance. Furthermore, we introduce an alternating least square-based learning algorithm for Multi-Mode RC as well as conduct the associated theoretical analysis. The result can be utilized to guide the configuration of NN parameters to sufficiently circumvent over-fitting issues. As a key application, we consider the symbol detection task in multiple-input-multiple-output (MIMO) orthogonal-frequency-division-multiplexing (OFDM) systems with massive MIMO employed at the base stations (BSs). Thanks to the tensor structure of massive MIMO-OFDM signals, our online learning-based symbol detection method generalizes well in terms of bit error rate even using a limited online training set. Evaluation results suggest that the Multi-Mode RC-based learning framework can efficiently and effectively combat practical constraints of wireless systems (i.e. channel state information (CSI) errors and hardware non-linearity) to enable robust and adaptive learning-based communications over the air.
For effective human-robot teaming, it is importantfor the robots to be able to share their visual perceptionwith the human operators. In a harsh remote collaborationsetting, however, it is especially challenging to transfer a largeamount of sensory data over a low-bandwidth network in real-time, e.g., for the task of 3D shape reconstruction given 2Dcamera images. To reduce the burden of data transferring, datacompression techniques such as autoencoder can be utilized toobtain and transmit the data in terms of latent variables in acompact form. However, due to the low-bandwidth limitation orcommunication delay, some of the dimensions of latent variablescan be lost in transit, degenerating the reconstruction results.Moreover, in order to achieve faster transmission, an intentionalover compression can be used where only partial elements ofthe latent variables are used. To handle these incomplete datacases, we propose a method for imputation of latent variableswhose elements are partially lost or manually excluded. Toperform imputation with only some dimensions of variables,exploiting prior information of the category- or instance-levelis essential. In general, a prior distribution used in variationalautoencoders is achieved from all of the training datapointsregardless of their labels. This type of flattened prior makes itdifficult to perform imputation from the category- or instance-level distributions.
Machine learning is rapidly becoming one of the most important technology for malware traffic detection, since the continuous evolution of malware requires a constant adaptation and the ability to generalize. However, network traffic datasets are usually oversized and contain redundant and irrelevant information, and this may dramatically increase the computational cost and decrease the accuracy of most classifiers, with the risk to introduce further noise. We propose two novel dataset optimization strategies which exploit and combine several state-of-the-art approaches in order to achieve an effective optimization of the network traffic datasets used to train malware detectors. The first approach is a feature selection technique based on mutual information measures and sensibility enhancement. The second is a dimensional reduction technique based autoencoders. Both these approaches have been experimentally applied on the MTA-KDD'19 dataset, and the optimized results evaluated and compared using a Multi Layer Perceptron as machine learning model for malware detection.
Lidar odometry (LO) is a key technology in numerous reliable and accurate localization and mapping systems of autonomous driving. The state-of-the-art LO methods generally leverage geometric information to perform point cloud registration. Furthermore, obtaining point cloud semantic information which can describe the environment more abundantly will help for the registration. We present a novel semantic lidar odometry method based on self-designed parameterized semantic features (PSFs) to achieve low-drift ego-motion estimation for autonomous vehicle in realtime. We first use a convolutional neural network-based algorithm to obtain point-wise semantics from the input laser point cloud, and then use semantic labels to separate the road, building, traffic sign and pole-like point cloud and fit them separately to obtain corresponding PSFs. A fast PSF-based matching enable us to refine geometric features (GeFs) registration, reducing the impact of blurred submap surface on the accuracy of GeFs matching. Besides, we design an efficient method to accurately recognize and remove the dynamic objects while retaining static ones in the semantic point cloud, which are beneficial to further improve the accuracy of LO. We evaluated our method, namely PSF-LO, on the public dataset KITTI Odometry Benchmark and ranked #1 among semantic lidar methods with an average translation error of 0.82% in the test dataset at the time of writing.