Instituttet for Anvendt Datateknik (Denmark)

In the field of social media mental health monitoring research, machine learning algorithms serve as the core support, establishing a crucial link between multi-source data and mental health status assessment. In view of the inherent flaws of traditional models, this paper constructs a mental health risk prediction model based on the Bidirectional gated Recurrent Unit (BiGRU) network. The research first conducts data distribution analysis and correlation analysis on the dataset, and then introduces multiple machine learning algorithms for classification and comparison. The results show that the proposed model demonstrates a comprehensive leading advantage in all evaluation indicators. Its accuracy rate, recall rate, precision rate and F1 value all reach 95.4%. Not only is it significantly higher than the 82.9%, 82.9%, 83.5%, 82.8% of Naive Bayes, 89.9%, 89.9%, 90.4%, 89.9% of ExtraTrees, and 87.7%, 87.7%, 87.9%, 87.7% of gradient boosting decision trees, It far exceeds support vector machines whose core classification indicators are only in the range of 54.3% to 57.9%, and also has a slight improvement compared to the suboptimal CatBoost. In terms of the AUC metric reflecting the ability to distinguish categories, this model achieved a high score of 99.6%, approaching the perfect classification level. It is slightly better than CatBoost's 99.2% and 99.0%, and significantly outperforms other comparison models, demonstrating superior classification accuracy and category discrimination ability. This model provides an efficient solution for the precise identification of mental health risks on social media, which is of great significance for promoting the practical development of mental health monitoring technology.

Text-to-music (TTM) generation, which converts textual descriptions into audio, opens up innovative avenues for multimedia creation. Achieving high quality and diversity in this process demands extensive, high-quality data, which are often scarce in available datasets. Most open-source datasets frequently suffer from issues like low-quality waveforms and low text-audio consistency, hindering the advancement of music generation models. To address these challenges, we propose a novel quality-aware training paradigm for generating high-quality, high-musicality music from large-scale, quality-imbalanced datasets. Additionally, by leveraging unique properties in the latent space of musical signals, we adapt and implement a masked diffusion transformer (MDT) model for the TTM task, showcasing its capacity for quality control and enhanced musicality. Furthermore, we introduce a three-stage caption refinement approach to address low-quality captions' issue. Experiments show state-of-the-art (SOTA) performance on benchmark datasets including MusicCaps and the Song-Describer Dataset with both objective and subjective metrics. Demo audio samples are available at https://qa-mdt.github.io/, code and pretrained checkpoints are open-sourced at https://github.com/ivcylc/OpenMusic.

Along with the growth of satellite communication industry, the demands and benefits to perform satellite terminal antenna evaluation are increasing. UAV based in-situ measurement can increase the efficiency of the measurement procedure. Main beam localization is a necessary procedure to execute the antenna evaluation test. To accelerate the process of finding the antenna beam centre, this paper develop a meta-reinforcement learning based algorithm. The developed algorithm is compared with other methods and it showed the best performance in terms of accuracy, robustness and travelling efficiency not only in the simulated radiation pattern environment but also in the empirically obtained radiation pattern.

The main current trend in applied formal methods can be characterized by the term “lightweight”. Historically, formal methods have been viewed as pure alternatives to traditional development methodologies, demanding a revolutionary change in industry to adopt them. With a pragmatic, lightweight approach, the use of formal methods is complementing and improving existing development practices in a company in an evolutionary way, demonstrating more clearly the cost-effectiveness of formal methods. This paper presents our view on lightweight formal methods as a strategy for successful formal methods technology transfer to industry.KeywordsFormal MethodGraphical NotationMetro SystemTechnology Transfer ProcessSoftware Life CycleThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

This paper presents the findings from the later phases of a study of the effects of introducing formal specification to the commercialscale development of a small security-critical system component. The objectives and form of the study are briefly reviewed. Observations have been made of the effort profile across the project, compliance of the developed system with customer requirements and software characteristics. The results of these observations are presented. Conclusions and areas of further work are discussed.

Experience shows that it is not economically feasible to formally specify all parts of a system in an industrial application. Either one already has a number of existing components which are trusted and therefore desirable for reuse, or components are so simple that there is no gain in formally specifying their behavior. In both cases it may be felt that it is not worth spending time on developing a detailed formal specification of the entire system. This raises the question what tools should be provided for the analysis of the entire system in which actual code is combined with specifications. In this paper we propose an approach which enables integration of code into a formal specification for prototyping facilities. The integration of code is supported by an extension to the IFAD VDM-SL Toolbox such that heterogeneous models can be interpreted.

In the IPTES project a dual language approach is proposed for overcoming both the problems derived from the use of a user-friendly, high-level, but not-formally-defined language and from a lower-level, formal, but difficult-to-use language. The approach uses a user-friendly, high-level language as user interface and a lower-level, formal language asmachine language. In this way the users can both access the IPTES environment through a nice interface and can profit from non-ambiguity-checks and proofs algorithms based on the formal kernel machine language. The correspondence between the two languages is built-in in the IPTES environment that provides a transparent mapping mechanism that relates the users specifications expressed by means of the high-level interface-language with the formal definitions expressed in the formal machine language.

The use of formal notations in the development of real-time systems is discussed. The basic idea is to have a number of formal notations in which the designer stepwise can refine the given specification. For each of the formal notations more and more design decisions are made and noted for a limited aspect of the problem. The other design decisions will then be postponed to a later stage of development where that aspect is taken into account. An example of how such formal notations could be defined for the design of real-time systems is given. The idea behind these notations is based on an event-action model, and it is therefore most appropriate for the development of reactive systems. The stepwise development idea is illustrated by the CEDER method.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>