MoSeS (Modelling and Simulation for e-Social Science) is a research node of the National Centre for e-Social Science. MoSeS uses e-Science techniques to execute an events-driven model that simulates discrete demographic processes; this allows us to project the UK population 25 years into the future. This paper describes the architecture, simulation methodology and latest results obtained by MoSeS.

No abstract

Few-shot text classification has recently been promoted by the meta-learning paradigm which aims to identify target classes with knowledge transferred from source classes with sets of small tasks named episodes. Despite their success, existing works building their meta-learner based on Prototypical Networks are unsatisfactory in learning discriminative text representations between similar classes, which may lead to contradictions during label prediction. In addition, the task-level and instance-level overfitting problems in few-shot text classification caused by a few training examples are not sufficiently tackled. In this work, we propose a contrastive learning framework named ContrastNet to tackle both discriminative representation and overfitting problems in few-shot text classification. ContrastNet learns to pull closer text representations belonging to the same class and push away text representations belonging to different classes, while simultaneously introducing unsupervised contrastive regularization at both task-level and instance-level to prevent overfitting. Experiments on 8 few-shot text classification datasets show that ContrastNet outperforms the current state-of-the-art models.

With the popularization of the Internet lifestyle and the innovation of learning methods, more and more online learning systems have emerged, allowing users to study in the system anytime and anywhere. While providing convenience to users, online learning systems also bring troubles to users, who cannot quickly find the resources they are interested in from the huge amount of learning resources. In this paper, we apply deep learning to an English online learning platform and analyze learners and learning contents by clustering algorithm and association rules. Based on this, a content organization system is developed using genetic algorithms, which is applied to the case of this paper to provide learners with personalized learning content. With the hope that the system can be extended to other online learning platforms in the future, three data mining techniques were selected to solve the problems found in the English online learning platform, and we designed how these techniques should be applied to the online learning platform. The first technique is the cluster mining technique, which is used to analyze learners’ profiles, classify learners in different categories, provide them with personalized learning content, and organize group learning. The second technique is association rules, which is used to analyze the correlation between learning contents. For the adaptive student-teacher knowledge migration strategy, the teacher model can guide the student model to track online and migrate the task-specific knowledge to the online tracking student model through the network parameters. Finally, a case study is selected and the above design is applied to this case study, and the results are analyzed in detail. The data mining technology is applied to the English online learning platform, and an innovative English learning content organization system is developed. It is hoped that the results of this study will have some practical value for promotion and provide an idea for the construction of the online learning platform, and it is also expected that the idea can improve the quality of online learning to a certain extent. Specifically, the online student model is adaptively updated by the teacher model parameters and the online student model parameters together.

There is a growing interest in training deep neural networks (DNNs) in a GPU cloud environment. This is typically achieved by running parallel training workers on multiple GPUs across computing nodes. Under such a setup, the communication overhead is often responsible for long training time and poor scalability. This paper presents AIACC-Training, a unified communication framework designed for the distributed training of DNNs in a GPU cloud environment. AIACC-Training permits a training worker to participate in multiple gradient communication operations simultaneously to improve network bandwidth utilization and reduce communication latency. It employs auto-tuning techniques to dynamically determine the right communication parameters based on the input DNN workloads and the underlying network infrastructure. AIACC-Training has been deployed to production at Alibaba GPU Cloud with 3000+ GPUs executing AIACC-Training optimized code at any time. Experiments performed on representative DNN workloads show that AIACC-Training outperforms existing solutions, improving the training throughput and scalability by a large margin.