In this study, we explored an approach to automate the review process of software design documents by using LLM. We first analyzed the review methods of design documents and organized 11 review perspectives. Additionally, we analyzed the issues of utilizing LLMs for these 11 review perspectives and determined which perspectives can be reviewed by current general-purpose LLMs instead of humans. For the reviewable perspectives, we specifically developed new techniques to enable LLMs to comprehend complex design documents that include table data. For evaluation, we conducted experiments using GPT to assess the consistency of design items and descriptions across different design documents in the design process used in actual business operations. Our results confirmed that LLMs can be utilized to identify inconsistencies in software design documents during the review process.

Systems based on Deep Neural Networks (DNNs) are increasingly being used in industry. In the process of system operation, DNNs need to be updated in order to improve their performance. When updating DNNs, systems used in companies that require high reliability must have as few regressions as possible. Since the update of DNNs has a data-driven nature, it is difficult to suppress regressions as expected by developers. This paper identifies the requirements for DNN updating in industry and presents a case study using techniques to meet those requirements. In the case study, we worked on satisfying the requirement to update models trained on car images collected in Fujitsu assuming security applications without regression for a specific class. We were able to suppress regression by customizing the objective function based on NeuRecover, a DNN repair technique. Moreover, we discuss some of the challenges identified in the case study.

Systematic techniques to improve quality of deep neural networks (DNNs) are critical given the increasing demand for practical applications including safety-critical ones. The key challenge comes from the little controllability in updating DNNs. Retraining to fix some behavior often has a destructive impact on other behavior, causing regressions, i.e., the updated DNN fails with inputs correctly handled by the original one. This problem is crucial when engineers are required to investigate failures in intensive assurance activities for safety or trust. Search-based repair techniques for DNNs have potentials to tackle this challenge by enabling localized updates only on “responsible parameters” inside the DNN. However, the potentials have not been explored to realize sufficient controllability to suppress regressions in DNN repair tasks. In this paper, we propose a novel DNN repair method that makes use of the training history for judging which DNN parameters should be changed or not to suppress regressions. We implemented the method into a tool called Neurecover and evaluated it with three datasets. Our method outperformed the existing method by achieving often less than a quarter, even a tenth in some cases, number of regressions. Our method is especially effective when the repair requirements are tight to fix specific failure types. In such cases, our method showed stably low rates (<2 %) of regressions, which were in many cases a tenth of regressions caused by retraining.

Application Programming Interfaces (APIs) are common in software development to reuse other products. Although the documentation allows API consumers to learn about API usages, it can be unreliable. Here, we investigate the characteristics of inappropriate usage examples in web API documentation by extracting and comparing OpenAPI Specifications from usage example-response pairs. About 65.5% of the endpoints have some form of inappropriate usage examples. Furthermore, mismatches are classified into four categories: undocumented keys pattern, dynamic keys pattern, unreturned keys pattern, and type mismatched pattern. Our results suggest that the number of keys in the response is correlated with the number of mismatches. These findings should assist both API providers and consumers who deal with unreliable documentation in web APIs.

This paper proposes a method to support Personal Software Process (PSP) in a development organization by classify the operations on a computer into a purpose of the user. PSP requires the developers to record and analyze their activity during the development process. There are several methods and systems to support the PSP, they records the operations automatically and also records a purpose of the operations (task) which records manually by the developer. Such manual recording by the developers is a barrier to introduction of the PSP system, and the cause of inaccurate record histories. Our proposal method classifies the operations into the task automatically with the chronological operation history. The method hypothesize that the each task consists of successive operation. The method classify the each operation into the task with a machine learning algorithm, Random Forests. An Experiment result shows the proposal method with chronological operation history classify the operation into the tasks more accurately than the method without the chronological operation history.