Machine Learning for 5G and Beyond:From ModelBased to Data-Driven Mobile Wireless Networks

During the past few decades,mobile wireless communications have experienced four generations of technological revolution,namely from 1 G to 4 G,and the deployment of the latest 5 G networks is expected to take place in 2019.One fundamental question is how we can push forward the development of mobile wireless communications while it has become an extremely complex and sophisticated system.We believe that the answer lies in the huge volumes of data produced by the network itself,and machine learning may become a key to exploit such information.In this paper,we elaborate why the conventional model-based paradigm,which has been widely proved useful in pre-5 G networks,can be less efficient or even less practical in the future 5 G and beyond mobile networks.Then,we explain how the data-driven paradigm,using state-of-the-art machine learning techniques,can become a promising solution.At last,we provide a typical use case of the data-driven paradigm,i.e.,proactive load balancing,in which online learning is utilized to adjust cell configurations in advance to avoid burst congestion caused by rapid traffic changes.

CNN-RNN based method for license plate recognition

Achieving good recognition results for License plates is challenging due to multiple adverse factors. For instance, in Malaysia, where private vehicle （e.g., cars） have numbers with dark background, while public vehicle （taxis/cabs） have numbers with white background. To reduce the complexity of the problem, we propose to classify the above two types of images such that one can choose an appropriate method to achieve better results. Therefore, in this work, we explore the combination of Convolutional Neural Networks （CNN） and Recurrent Neural Networks namely, BLSTM （Bi-Directional Long Short Term Memory）, for recognition. The CNN has been used for feature extraction as it has high discriminative ability, at the same time, BLSTM has the ability to extract context information based on the past information. For classification, we propose Dense Cluster based Voting （DCV）, which separates foreground and background for successful classification of private and public. Experimental results on live data given by MIMOS, which is funded by Malaysian Government and the standard dataset UCSD show that the proposed classification outperforms the existing methods. In addition, the recognition results show that the recognition performance improves significantly after classification compared to before classification.

What is Discussed about COVID-19:A Multi-Modal Framework for Analyzing Microblogs from Sina Weibo without Human Labeling

Starting from late 2019,the new coronavirus disease(COVID-19)has become a global crisis.With the development of online social media,people prefer to express their opinions and discuss the latest news online.We have witnessed the positive influence of online social media,which helped citizens and governments track the development of this pandemic in time.It is necessary to apply artificial intelligence(AI)techniques to online social media and automatically discover and track public opinions posted online.In this paper,we take Sina Weibo,the most widely used online social media in China,for analysis and experiments.We collect multi-modal microblogs about COVID-19 from 2020/1/1 to 2020/3/31 with a web crawler,including texts and images posted by users.In order to effectively discover what is being discussed about COVID-19 without human labeling,we propose a unified multi-modal framework,including an unsupervised short-text topic model to discover and track bursty topics,and a self-supervised model to learn image features so that we can retrieve related images about COVID-19.Experimental results have shown the effectiveness and superiority of the proposed models,and also have shown the considerable application prospects for analyzing and tracking public opinions about COVID-19.

Model gradient: unified model and policy learning in model-based reinforcement learning

Model-based reinforcement learning is a promising direction to improve the sample efficiency of reinforcement learning with learning a model of the environment.Previous model learning methods aim at fitting the transition data,and commonly employ a supervised learning approach to minimize the distance between the predicted state and the real state.The supervised model learning methods,however,diverge from the ultimate goal of model learning,i.e.,optimizing the learned-in-the-model policy.In this work,we investigate how model learning and policy learning can share the same objective of maximizing the expected return in the real environment.We find model learning towards this objective can result in a target of enhancing the similarity between the gradient on generated data and the gradient on the real data.We thus derive the gradient of the model from this target and propose the Model Gradient algorithm(MG)to integrate this novel model learning approach with policy-gradient-based policy optimization.We conduct experiments on multiple locomotion control tasks and find that MG can not only achieve high sample efficiency but also lead to better convergence performance compared to traditional model-based reinforcement learning approaches.

Example based painting generation

We present an approach for generating paintings on photographic images with the style encoded by the example paintings and adopt representative brushes extracted from the example paintings as the painting primitives. Our system first divides the given photographic image into several regions on which we synthesize a grounding layer with texture patches extracted from the example paintings. Then, we paint those regions using brushes stochastically chosen from the brush library, with further brush color and shape perturbations. The brush direction is determined by a direction field either constructed by a convenient user interactive manner or synthesized from the examples. Our approach offers flexible and intuitive user control over the painting process and style.

Pairwise tagging framework for end-to-end emotion-cause pair extraction

Emotion-cause pair extraction(ECPE)aims to extract all the pairs of emotions and corresponding causes in a document.It generally contains three subtasks,emotions extraction,causes extraction,and causal relations detection between emotions and causes.Existing works adopt pipelined approaches or multi-task learning to address the ECPE task.However,the pipelined approaches easily suffer from error propagation in real-world scenarios.Typical multi-task learning cannot optimize all tasks globally and may lead to suboptimal extraction results.To address these issues,we propose a novel framework,Pairwise Tagging Framework(PTF),tackling the complete emotion-cause pair extraction in one unified tagging task.Unlike prior works,PTF innovatively transforms all subtasks of ECPE,i.e.,emotions extraction,causes extraction,and causal relations detection between emotions and causes,into one unified clause-pair tagging task.Through this unified tagging task,we can optimize the ECPE task globally and extract more accurate emotion-cause pairs.To validate the feasibility and effectiveness of PTF,we design an end-to-end PTF-based neural network and conduct experiments on the ECPE benchmark dataset.The experimental results show that our method outperforms pipelined approaches significantly and typical multi-task learning approaches.

CA-DTS:A Distributed and Collaborative Task Scheduling Algorithm for Edge Computing Enabled Intelligent Road Network

Edge computing enabled Intelligent Road Network(EC-IRN)provides powerful and convenient computing services for vehicles and roadside sensing devices.The continuous emergence of transportation applications has caused a huge burden on roadside units(RSUs)equipped with edge servers in the Intelligent Road Network(IRN).Collaborative task scheduling among RSUs is an effective way to solve this problem.However,it is challenging to achieve collaborative scheduling among different RSUs in a completely decentralized environment.In this paper,we first model the interactions involved in task scheduling among distributed RSUs as a Markov game.Given that multi-agent deep reinforcement learning(MADRL)is a promising approach for the Markov game in decision optimization,we propose a collaborative task scheduling algorithm based on MADRL for EC-IRN,named CA-DTS,aiming to minimize the long-term average delay of tasks.To reduce the training costs caused by trial-and-error,CA-DTS specially designs a reward function and utilizes the distributed deployment and collective training architecture of counterfactual multi-agent policy gradient(COMA).To improve the stability of performance in large-scale environments,CA-DTS takes advantage of the action semantics network(ASN)to facilitate cooperation among multiple RSUs.The evaluation results of both the testbed and simulation demonstrate the effectiveness of our proposed algorithm.Compared with the baselines,CA-DTS can achieve convergence about 35%faster,and obtain average task delay that is lower by approximately 9.4%,9.8%,and 6.7%,in different scenarios with varying numbers of RSUs,service types,and task arrival rates,respectively.

Minimal Gated Unit for Recurrent Neural Networks

Recurrent neural networks （RNN） have been very successful in handling sequence data. However, understanding RNN and finding the best practices for RNN learning is a difficult task, partly because there are many competing and complex hidden units, such as the long short-term memory （LSTM） and the gated recurrent unit （GRU）. We propose a gated unit for RNN, named as minimal gated unit （MCU）, since it only contains one gate, which is a minimal design among all gated hidden units. The design of MCU benefits from evaluation results on LSTM and GRU in the literature. Experiments on various sequence data show that MCU has comparable accuracy with GRU, but has a simpler structure, fewer parameters, and faster training. Hence, MGU is suitable in RNN＇s applications. Its simple architecture also means that it is easier to evaluate and tune, and in principle it is easier to study MGU＇s properties theoretically and empirically.

Software Defect Detection with ROCUS

Software defect detection aims to automatically identify defective software modules for efficient software test in order to improve the quality of a software system. Although many machine learning methods have been successfully applied to the task, most of them fail to consider two practical yet important issues in software defect detection. First, it is rather difficult to collect a large amount of labeled training data for learning a well-performing model; second, in a software system there are usually much fewer defective modules than defect-free modules, so learning would have to be conducted over an imbalanced data set. In this paper~ we address these two practical issues simultaneously by proposing a novel semi-supervised learning approach named Rocus. This method exploits the abundant unlabeled examples to improve the detection accuracy, as well as employs under-sampling to tackle the class-imbalance problem in the learning process. Experimental results of real-world software defect detection tasks show that Rocus is effective for software defect detection. Its performance is better than a semi-supervised learning method that ignores the class-imbalance nature of the task and a class-imbalance learning method that does not make effective use of unlabeled data.

Image Region Selection and Ensemble for Face Recognition

In this paper, a novel framework for face recognition, namely Selective Ensemble of Image Regions （SEIR）, is proposed. In this framework, all possible regions in the face image are regarded as a certain kind of features. There are two main steps in SEIR： the first step is to automatically select several regions from all possible candidates; the second step is to construct classifier ensemble from the selected regions. An implementation of SEIR based on multiple eigenspaces, namely SEME, is also proposed in this paper. SEME is analyzed and compared with eigenface, PCA ＋ LDA, eigenfeature, and eigenface ＋ eigenfeature through experiments. The experimental results show that SEME achieves the best performance.

Structural diversity for decision tree ensemble learning

Decision trees are a kind of off-the-shelf predictive models, and they have been successfully used as the base learners in ensemble learning. To construct a strong classi- fier ensemble, the individual classifiers should be accurate and diverse. However, diversity measure remains a mystery although there were many attempts. We conjecture that a deficiency of previous diversity measures lies in the fact that they consider only behavioral diversity, i.e., how the classifiers behave when making predictions, neglecting the fact that classifiers may be potentially different even when they make the same predictions. Based on this recognition, in this paper, we advocate to consider structural diversity in addition to behavioral diversity, and propose the TMD （tree matching diversity） measure for decision trees. To investigate the usefulness of TMD, we empirically evaluate performances of selective ensemble approaches with decision forests by incorporating different diversity measures. Our results validate that by considering structural and behavioral diversities together, stronger ensembles can be constructed. This may raise a new direction to design better diversity measures and ensemble methods.

Intelligent Development Environment and Software Knowledge Graph

Software intelligent development has become one of the most important research trends in software engineering. In this paper, we put forward two key concepts -- intelligent development environment （IntelliDE） and software knowledge graph -- for the first time. IntelliDE is an ecosystem in which software big data are aggregated, mined and analyzed to provide intelligent assistance in the life cycle of software development. We present its architecture and discuss its key research issues and challenges. Software knowledge graph is a software knowledge representation and management framework, which plays an important role in IntelliDE. We study its concept and introduce some concrete details and examples to show how it could be constructed and leveraged.

Derivative-free reinforcement learning:a review

Reinforcement learning is about learning agent models that make the best sequential decisions in unknown environments.In an unknown environment,the agent needs to explore the environment while exploiting the collected information,which usually forms a sophisticated problem to solve.Derivative-free optimization,meanwhile,is capable of solving sophisticated problems.It commonly uses a sampling-andupdating framework to iteratively improve the solution,where exploration and exploitation are also needed to be well balanced.Therefore,derivative-free optimization deals with a similar core issue as reinforcement learning,and has been introduced in reinforcement learning approaches,under the names of learning classifier systems and neuroevolution/evolutionary reinforcement learning.Although such methods have been developed for decades,recently,derivative-free reinforcement learning exhibits attracting increasing attention.However,recent survey on this topic is still lacking.In this article,we summarize methods of derivative-free reinforcement learning to date,and organize the methods in aspects including parameter updating,model selection,exploration,and parallel/distributed methods.Moreover,we discuss some current limitations and possible future directions,hoping that this article could bring more attentions to this topic and serve as a catalyst for developing novel and efficient approaches.

Instance selection method for improving graph-based semi-supervised learning

Graph-based semi-supervised learning is an important semi-supervised learning paradigm. Although graphbased semi-supervised learning methods have been shown to be helpful in various situations, they may adversely affect performance when using unlabeled data. In this paper, we propose a new graph-based semi-supervised learning method based on instance selection in order to reduce the chances of performance degeneration. Our basic idea is that given a set of unlabeled instances, it is not the best approach to exploit all the unlabeled instances; instead, we should exploit the unlabeled instances that are highly likely to help improve the performance, while not taking into account the ones with high risk. We develop both transductive and inductive variants of our method. Experiments on a broad range of data sets show that the chances of performance degeneration of our proposed method are much smaller than those of many state-of-the-art graph-based semi-supervised learning methods.

The complexity of variable minimal formulas

Based on the common properties of logic formulas:equivalence and satisfiability,the concept of variable minimal formulas with property preservation is introduced.A formula is variable minimal if the resulting sub-formulas with any variable omission will change the given property.Some theoretical results of two classes:variable minimal equivalence(VME) and variable minimal satisfiability(VMS) are studied.We prove that VME is NP-complete,and VMS is in DP and coNP-hard.

Self-Supervised Task Augmentation for Few-Shot Intent Detection

Few-shot intent detection is a practical challenge task,because new intents are frequently emerging and collecting large-scale data for them could be costly.Meta-learning,a promising technique for leveraging data from previous tasks to enable efficient learning of new tasks,has been a popular way to tackle this problem.However,the existing meta-learning models have been evidenced to be overfitting when the meta-training tasks are insufficient.To overcome this challenge,we present a novel self-supervised task augmentation with meta-learning framework,namely STAM.Firstly,we introduce the task augmentation,which explores two different strategies and combines them to extend meta-training tasks.Secondly,we devise two auxiliary losses for integrating self-supervised learning into meta-learning to learn more generalizable and transferable features.Experimental results show that STAM can achieve consistent and considerable performance improvement to existing state-of-the-art methods on four datasets.

PTM： A Topic Model for the Inferring of the Penalty

Deciding the penalty of a law case has always been a complex process, which may involve with much coordination. Despite the judicial study based on the rules and conditions, artificial intelligence and machine learning has rarely been used to study the problem of penalty inferring, leaving the large amount of law cases as well as various factors among them untouched. This paper aims to incorporate the state-of-the-art artificial intelligence methods to exploit to what extent this problem can be alleviated. We first analyze 145 000 law cases and observe that there are two sorts of labels, temporal labels and spatial labels, which have unique characteristics. Temporal labels and spatial labels tend to converge towards the final penalty, on condition that the cases are of the same category. In light of this, we propose a latent-class probabilistic generative model, namely Penalty Topic Model （PTM）, to infer the topic of law cases, and the temporal and spatial patterns of topics embedded in the case judgment. Then, the learnt knowledge is utilized to automatically cluster all cases accordingly in a unified way. We conduct extensive experiments to evaluate the performance of the proposed PTM on a real large-scale dataset of law cases. The experimental results show the superiority of our proposed PTM.

ZenLDA: Large-Scale Topic Model Training on Distributed Data-Parallel Platform

Recently, topic models such as Latent Dirichlet Allocation(LDA) have been widely used in large-scale web mining. Many large-scale LDA training systems have been developed, which usually prefer a customized design from top to bottom with sophisticated synchronization support. We propose an LDA training system named ZenLDA, which follows a generalized design for the distributed data-parallel platform. The novelty of ZenLDA consists of three main aspects:(1) it converts the commonly used serial Collapsed Gibbs Sampling(CGS) inference algorithm to a Monte-Carlo Collapsed Bayesian(MCCB) estimation method, which is embarrassingly parallel;(2)it decomposes the LDA inference formula into parts that can be sampled more efficiently to reduce computation complexity;(3) it proposes a distributed LDA training framework, which represents the corpus as a directed graph with the parameters annotated as corresponding vertices and implements ZenLDA and other well-known inference methods based on Spark. Experimental results indicate that MCCB converges with accuracy similar to that of CGS, while running much faster. On top of MCCB, the ZenLDA formula decomposition achieved the fastest speed among other well-known inference methods. ZenLDA also showed good scalability when dealing with large-scale topic models on the data-parallel platform. Overall, ZenLDA could achieve comparable and even better computing performance with state-of-the-art dedicated systems.

A Semi-Supervised Attention Model for Identifying Authentic Sneakers

To protect consumers and those who manufacture and sell the products they enjoy,it is important to develop convenient tools to help consumers distinguish an authentic product from a counterfeit one.The advancement of deep learning techniques for fine-grained object recognition creates new possibilities for genuine product identification.In this paper,we develop a Semi-Supervised Attention(SSA)model to work in conjunction with a large-scale multiple-source dataset named YSneaker,which consists of sneakers from various brands and their authentication results,to identify authentic sneakers.Specifically,the SSA model has a self-attention structure for different images of a labeled sneaker and a novel prototypical loss is designed to exploit unlabeled data within the data structure.The model draws on the weighted average of the output feature representations,where the weights are determined by an additional shallow neural network.This allows the SSA model to focus on the most important images of a sneaker for use in identification.A unique feature of the SSA model is its ability to take advantage of unlabeled data,which can help to further minimize the intra-class variation for more discriminative feature embedding.To validate the model,we collect a large number of labeled and unlabeled sneaker images and perform extensive experimental studies.The results show that YSneaker together with the proposed SSA architecture can identify authentic sneakers with a high accuracy rate.