Despite the success of self-supervised pre-training in texts and images, applying it to multivariate time series (MTS) falls behind tailored methods for tasks like forecasting, imputation and anomaly detection. We propose a general-purpose framework, named UP2ME (Univariate Pre-training to Multivariate Fine-tuning). It conducts task-agnostic pre-training when downstream tasks are unspecified. Once the task and setting (e.g. forecasting length) are determined, it gives sensible solutions with frozen pre-trained parameters, which has not been achieved before. UP2ME is further refined by fine-tuning. A univariate-to-multivariate paradigm is devised to address the heterogeneity of temporal and cross-channel dependencies. In univariate pre-training, univariate instances with diverse lengths are generated for Masked AutoEncoder (MAE) pre-training, discarding cross-channel dependency. The pre-trained model handles downstream tasks by formulating them into specific mask-reconstruction problems. In multivariate fine-tuning, it constructs a dependency graph among channels using the pre-trained encoder to enhance cross-channel dependency capture. Experiments on eight real-world datasets show its SOTA performance in forecasting and imputation, approaching task-specific performance in anomaly detection. Our code is available at https://github.com/Thinklab-SJTU/UP2ME.

Software architecture security design is a key stage in developing business-oriented system, such as business-critical system, ICT system and AI system. Many typical accidents also remind us that the security of software architecture even plays a more important role than the code security in most software systems. However, there are very few researches which focus on the security of software architecture. Especially, we don’t find a systematic and feasible quantitative method to evaluate the security of software architecture. To fill this gap, we launched a research project focusing on software architecture security since 2019 and try to provide a series of quantitative methods for evaluating the security of software architecture. In this paper, a business-oriented quantitative method was proposed to evaluate the security of software architecture from the view of business-critical security insurance. In our method, both business dependency graph (BDG) and multi-hierarchical dependency graph (MHDG) are defined and constructed first for describing businesses and their relationships, and system components and their relationships; then, attack points and business key-points are identified and labeled on BDG and MHDG; and further, all potential attack paths and effective attack paths in the system based on MHDG are detected; and finally, a set of security indicators is defined and used to evaluate the security of software architecture quantitatively. For more effective experiments, we use software architectures with different styles and different evolution versions. Experimental results show that our method can reflect effectively the security characteristics of software architecture with different styles, can find the security changes of different architecture evolution versions for the same system, and can perform quantitative evaluation of software architecture security efficiently.