Incremental Group-Level Popularity Prediction in Online Social Networks

Predicting the popularity of web contents in online social networks is essential for many applications. However, existing works are usually under non-incremental settings. In other words, they have to rebuild models from scratch when new data occurs, which are inefficient in big data environments. It leads to an urgent need for incremental prediction, which can update previous results with new data and conduct prediction incrementally. Moreover, the promising direction of group-level popularity prediction has not been well treated, which explores fine-grained information while keeping a low cost. To this end, we identify the problem of incremental group-level popularity prediction, and propose a novel model IGPP to address it. We first predict the group-level popularity incrementally by exploiting the incremental CANDECOMP/PARAFCAC (CP) tensor decomposition algorithm. Then, to reduce the cumulative error by incremental prediction, we propose three strategies to restart the CP decomposition. To the best of our knowledge, this is the first work that identifies and solves the problem of incremental group-level popularity prediction. Extensive experimental results show significant improvements of the IGPP method over other works both in the prediction accuracy and the efficiency.

Measuring and Modeling Group Dynamics in Open-Source Software Development: A Tensor Decomposition Approach

Many open-source software projects depend on a few core developers, who take over both the bulk of coordination and programming tasks. They are supported by peripheral developers, who contribute either via discussions or programming tasks, often for a limited time. It is unclear what role these peripheral developers play in the programming and communication efforts, as well as the temporary task-related sub-groups in the projects. We mine code-repository data and mailing-list discussions to model the relationships and contributions of developers in a social network and devise a method to analyze the temporal collaboration structures in communication and programming, learning about the strength and stability of social sub-groups in open-source software projects. Our method uses multi-modal social networks on a series of time windows. Previous work has reduced the network structure representing developer collaboration to networks with only one type of interaction, which impedes the simultaneous analysis of more than one type of interaction. We use both communication and version-control data of open-source software projects and model different types of interaction over time. To demonstrate the practicability of our measurement and analysis method, we investigate 10 substantial and popular open-source software projects and show that, if sub-groups evolve, modeling these sub-groups helps predict the future evolution of interaction levels of programmers and groups of developers. Our method allows maintainers and other stakeholders of open-source software projects to assess instabilities and organizational changes in developer interaction and can be applied to different use cases in organizational analysis, such as understanding the dynamics of a specific incident or discussion.

Compression of Deep Learning Models for Text: A Survey

In recent years, the fields of natural language processing (NLP) and information retrieval (IR) have made tremendous progress thanks to deep learning models like Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs) and Long Short-Term Memory (LSTMs) networks, and Transformer [ 121 ] based models like Bidirectional Encoder Representations from Transformers (BERT) [ 24 ], Generative Pre-training Transformer (GPT-2) [ 95 ], Multi-task Deep Neural Network (MT-DNN) [ 74 ], Extra-Long Network (XLNet) [ 135 ], Text-to-text transfer transformer (T5) [ 96 ], T-NLG [ 99 ], and GShard [ 64 ]. But these models are humongous in size. On the other hand, real-world applications demand small model size, low response times, and low computational power wattage. In this survey, we discuss six different types of methods (Pruning, Quantization, Knowledge Distillation (KD), Parameter Sharing, Tensor Decomposition, and Sub-quadratic Transformer-based methods) for compression of such models to enable their deployment in real industry NLP projects. Given the critical need of building applications with efficient and small models, and the large amount of recently published work in this area, we believe that this survey organizes the plethora of work done by the “deep learning for NLP” community in the past few years and presents it as a coherent story.

Tensor-Based Recursive Least-Squares Adaptive Algorithms with Low-Complexity and High Robustness Features

The recently proposed tensor-based recursive least-squares dichotomous coordinate descent algorithm, namely RLS-DCD-T, was designed for the identification of multilinear forms. In this context, a high-dimensional system identification problem can be efficiently addressed (gaining in terms of both performance and complexity), based on tensor decomposition and modeling. In this paper, following the framework of the RLS-DCD-T, we propose a regularized version of this algorithm, where the regularization terms are incorporated within the cost functions. Furthermore, the optimal regularization parameters are derived, aiming to attenuate the effects of the system noise. Simulation results support the performance features of the proposed algorithm, especially in terms of its robustness in noisy environments.

FDA-MIMO for target localization via multi-pulse tensor decomposition

By combining multiple input multiple output (MIMO) technology and multiple matched filters with frequency diverse array (FDA), FDA-MIMO radar can be used to achieve two-dimensional target localization with range and angle. In this paper, we propose two FDA-MIMO multi-pulse target localization methods based on tensor decomposition. Based on the canonical polyadic decomposition theory, the signal models of CPD-DP-FDA with double-pulse and CPD-SP-FDA with stepped frequency pulses are established. By analyzing the signal processing procedures of the two schemes, the indicator beampattern used for target localization is obtained. The parameter estimation accuracy of the proposed method is investigated in single target and multiple targets scenarios, and the proposed method is compared with the traditional double-pulse method. The results show that the target localization method based on tensor decomposition can effectively solve the problem of multi-target indication ambiguity. The target positioning effect can be further improved by combining stepped frequency pulses. The derivation of Cramer–Rao Lower Bound (CRLB) demonstrates the superiority of the method.

Robust Barron-Loss Tucker Tensor Decomposition

<p>In this work, we propose a new formulation for low-rank tensor approximation, with tunable outlier-robustness, and present a unified algorithmic solution framework. This formulation relies on a new generalized robust loss function (Barron loss), which encompasses several well-known loss-functions with variable outlier resistance. The robustness of the proposed framework is corroborated by the presented numerical studies on synthetic and real data.</p>

Robust Barron-Loss Tucker Tensor Decomposition

<p>In this work, we propose a new formulation for low-rank tensor approximation, with tunable outlier-robustness, and present a unified algorithmic solution framework. This formulation relies on a new generalized robust loss function (Barron loss), which encompasses several well-known loss-functions with variable outlier resistance. The robustness of the proposed framework is corroborated by the presented numerical studies on synthetic and real data.</p>