SCPL: Enhancing Neural Network Training Throughput with Decoupled Local Losses and Model Parallelism

The necessity of aligning an enterprise's information system (IS) model to its business process (BP) model is incontestable to the consistent analysis of the business performance. However, the main difficulty of establishing/maintaining BP-IS model alignment stems from the dissimilarities in the knowledge of the information system developers and the business process experts. To overcome these limits, the authors propose a model-driven architecture compliant methodology that helps software analysts to build an IS analysis model aligned to a given BP model. The proposed methodology allows mastering transformation from computation independent model to platform independent model. The CIM level expresses the BP, which is modelled through the standard BPMN and, at the PIM level represents the aligned IS model, which is generated as use case diagram, system sequence diagrams, and class diagram. CIM to PIM transformation accounts for the BP structural and semantic perspectives to generate an aligned IS model that respects the best-practice granularity level and the quality of UML diagrams.

The standardized business processing rules and information processing flow, as well as the risk control according to the internal control norms, reflect the reliability, validity and robustness of the Enterprise Information System (EIS). The business objectives effectively met in EIS requires the integration of internal control concepts, rule & regulatory, standardized processes and measures in EIS. Furthermore, its implementation is a complex and large systematic engineering, and related to corporate governance structure, enterprise internal control environment, as well as many other factors. We introduce domain analysis and formal methods, which are a mathematic-based methodology in software engineering, to specify, model, verify the EIS if the desired internal control properties are contained in the software system during the design stage. So we can find defects and vulnerabilities quickly and effectively in the early-stage during EIS implementation, to reduce the risk of the inefficient or failure of internal control in EIS. In this paper, we first study the background of internal control in EIS, especially Chinese enterprise internal control environment, then introduce how to apply domain analysis and formal methods into EIS design to ensure internal control met business objective, at last, we take the sales activities of internal control under Chinese enterprise environment as an example to illustrate our method.

Yuri A. Zelenkov - Professor, Department of Information Systems and Digital Infrastructure Management, National Research University Higher School of EconomicsAddress: 20, Myasnitskaya Street, Moscow, 101000, Russian FederationE-mail: yzelenkov @hse.ru A modern enterprise has to react to permanent changes in the business environment by transformation of its own behavior, operational practices and business processes. Such transformations may range from changes of business processes to changes of information systems used to support the business processes, changes in the underlying IT infrastructures and even in the enterprise information system as a whole. The main characteristic of changes in a turbulent business environment and, consequently, in the enterprise information system is unpredictability. Therefore, an enterprise information system should support the operational efficiency of the current business model, as well as provide the necessary level of agility to implement future unpredictable changes of requirements. This article aims to propose a conceptual model of an agile enterprise information system, which is defined as a working system that should eliminate the largest possible number of gaps caused by external events through incremental changes of its own components. A conceptual model developed according to the socio-technical approach includes structural properties of an agile enterprise information system (actors, tasks, technology, and structure). Structural properties define its operational characteristics, i.e. measurable indicators of agility - time, costs, scope and robustness of process of change. Different ways to build such an agile system are discussed on the basis of axiomatic design theory. We propose an approach to measurement of time, cost, scope and robustness of changes which helps to make quantitative estimation of the achieved level of agility.

PurposeThe purpose of this paper is to identify the success factors of open‐source software in the enterprise level. It expands the application of the information systems (IS) success model in the literature to enterprise information systems (EIS). The paper presents a simplified open‐source EIS success model by removing several constructs in the existing open‐source software models.Design/methodology/approachTo test the research model, a survey questionnaire was developed based on previous studies dealing with IS success models and adapting them to the open‐source EIS context. The research instrument contained 30 items that represent seven constructs in the research model. Data were collected from 250 open‐source enterprise software end‐users. Due to its confirmatory nature, this study applied the structural equation model.FindingsThe results of the study indicate that only community service quality has a positive direct effect on open‐source EIS use, while information quality, EIS quality, and user satisfaction do not. Open‐source EIS quality has a direct positive effect on user satisfaction, which in turn has a positive effect on individual net benefits, which also positively affects organizational net benefits.Research limitations/implicationsThis study focused on the open‐source EIS users' perspective. Future studies could expand the scope by covering a broader open‐source EIS aspect such as motivation of its use, development processes, social dynamics in the development group, diffusion process, and the like. A longitudinal study could provide a more concrete trend of open‐source EIS use by organizations. The small sample size of this study is also a limitation.Practical implicationsThe present research provides a practical evidence of relationships in the open‐source EIS application model. The developers in on‐line open‐source communities need to take the success factors identified in this study into account when developing open‐source EIS.Originality/valueThere is a paucity of empirical studies in open‐source EIS applications. The paper expends the traditional IS success model to the open‐source EIS context by collecting and analyzing data from 150 real‐world open‐source EIS users.

Usage of large language models (LLMs) in enterprise information systems has been growing more common in recent years. The purpose of the study is to consider the evolution of technologies used in enterprise information systems from mathematical models to neural networks, including large language models, as well as suggest recommendations on the application of LLMs in enterprise systems and assume future tendencies and risks of using artificial intelligence (AI) in such systems. Methods and materials. A retrospective method was used to analyze the development of artificial intelligence algorithms. Several approaches to using LLMs in enterprise systems and decision support systems were considered and compared by several criteria identified based on the existing studies. Results. The conducted study includes several recommendations on the best practices of applying LLM-based approaches to enterprise systems. It also covers the advantages and disadvantages of a new LLM-based programming approach, where a person acts as a system architect, while a model performs the technical tasks. The study describes the most recent advanced technology, agentic AI, which allows large language models to interact with their external environments and perform diverse tasks using various tools. The study also includes assumptions about future tendencies of AI usage in enterprise information systems and the corresponding risks. Conclusion. The results of this study can be used as a base for managers’ decision making regarding the feasibility of using LLM-based methods considered in the study and their corresponding risks when building enterprise information systems.

Because of the special position of electric power industry in national economy and social life, electric power information security has been given particular attention. With the development of electric power informationization, the threat of information security has become more and more serious. Risks and losses, caused by potential safety hazard in electric power enterprise information system, have become more and more serious. The special nature of electric power enterprise information system security has particular requirements for information security defense system. In order to meet the special requirements of electric power enterprise information systems security, the concept of network business security has been proposed, constructing security defense system from the two levels of data security and network business security. According to the specific situation of power grid enterprise information system's businesses, classification and protection methods for power grid enterprise information system have been proposed. What's more, on the basis of the classification and protection methods, security defense model of power grid enterprise information system has been designed, providing a theoretical basis for the security construction of power grid enterprise information system.

Domain theory provides the basis for the ontology based concept model, which has been proved to be effective in building the enterprise information system. However, most of existing enterprise information system poses great difficult in solving the changing requirement from users. In this case, enterprise has to build a new information system. This paper presents presented a domain theory based adaptive concept model for enterprise information system. This model is capable of coping with the changing requirements from users, and can be used as an effective tool for the building of enterprise information system.

In this paper, we propose a novel approach for efficient training of deep neural networks in a bottom-up fashion using a layered structure. Our algorithm, which we refer to as deep cascade learning, is motivated by the cascade correlation approach of Fahlman and Lebiere, who introduced it in the context of perceptrons. We demonstrate our algorithm on networks of convolutional layers, though its applicability is more general. Such training of deep networks in a cascade directly circumvents the well-known vanishing gradient problem by ensuring that the output is always adjacent to the layer being trained. We present empirical evaluations comparing our deep cascade training with standard end-end training using back propagation of two convolutional neural network architectures on benchmark image classification tasks (CIFAR-10 and CIFAR-100). We then investigate the features learned by the approach and find that better, domain-specific, representations are learned in early layers when compared to what is learned in end-end training. This is partially attributable to the vanishing gradient problem that inhibits early layer filters to change significantly from their initial settings. While both networks perform similarly overall, recognition accuracy increases progressively with each added layer, with discriminative features learned in every stage of the network, whereas in end-end training, no such systematic feature representation was observed. We also show that such cascade training has significant computational and memory advantages over end-end training, and can be used as a pretraining algorithm to obtain a better performance.

Enterprise information systems with a great use of Web 2.0 technologies will be more open, free, and more efficient. With the contrast between classic Web technologies and Web 2.0 technologies, we represent a sample of enterprise information system based on Web 2.0, and show how the use of Web 2.0 technologies changes the system data exchange model of the enterprise information systems and how it improves the efficiency and effectiveness of information systems.

The paper aims to present a solution of managing an enterprise information system or enterprise resource planning (ERP) in small-mid size enterprises (SMEs). The research was focused on developing an integrated enterprise information system for resolving faced problems in managing SMEs’ business. The methods were investigating business requirements regarding information system, system analysis, system design which included processes, data, and interface model, implementation of system design using selected programming language, and system testing which included unit and integrated testing. The research produced a tested model of the enterprise information system and a working enterprise resource planning information system software named Qinova ERP. The system is implemented in a real environment, a hazardous waste transporter company in Indonesia, and the system has been tackling their business information system management problems. The system is ready to be employed to the different companies in the same industry with minor revision and to the other industries with some customizations to adapt their business requirements.

With the trend of economic globalization and localization services, employees are distributed in different regions, the old enterprise information system in closed environment has been difficult to support all the business in enterprises, also cannot meet the need of information sharing between upstream and downstream enterprises and partners in the supply chain. The new business model requires companies to have distributed information systems, remote access and other characteristics. VPN (virtual private network) is high cost and lack of flexibility, Web services-based information system can achieve low-cost real-time collection to process and share distributed information, which is the ideal model of enterprise information system. However, there is a big gap in current usability between the Web services and old desktop applications. This paper combines the usage patterns, business needs of enterprise information systems and technical characteristics of Web services, proposes the usability requirements of enterprise information systems based on Web services from different views of internal users, external customers and strategic partners.

The training process of deep neural networks (DNNs) is usually pipelined with stages for data preparation on CPUs followed by gradient computation on accelerators like GPUs. In an ideal pipeline, the end-to-end training throughput is eventually limited by the throughput of the accelerator, not by that of data preparation. In the past, the DNN training pipeline achieved a near-optimal throughput by utilizing datasets encoded with a lightweight, lossy image format like JPEG. However, as high-resolution, losslessly-encoded datasets become more popular for applications requiring high accuracy, a performance problem arises in the data preparation stage due to low-throughput image decoding on the CPU. Thus, we propose L3, a custom lightweight, lossless image format for high-resolution, high-throughput DNN training. The decoding process of L3 is effectively parallelized on the accelerator, thus minimizing CPU intervention for data preparation during DNN training. L3 achieves a 9.29\(\times \) higher data preparation throughput than PNG, the most popular lossless image format, for the Cityscapes dataset on NVIDIA A100 GPU, which leads to 1.71\(\times \) higher end-to-end training throughput. Compared to JPEG and WebP, two popular lossy image formats, L3 provides up to 1.77\(\times \) and 2.87\(\times \) higher end-to-end training throughput for ImageNet, respectively, at equivalent metric performance.KeywordsDNN trainingData preparationImage processing

Backpropagation (BP) is the cornerstone of today's deep learning algorithms, but it is inefficient partially because of backward locking, which means updating the weights of one layer locks the weight updates in the other layers. Consequently, it is challenging to apply parallel computing or a pipeline structure to update the weights in different layers simultaneously. In this letter, we introduce a novel learning structure, associated learning (AL), that modularizes the network into smaller components, each of which has a local objective. Because the objectives are mutually independent, AL can learn the parameters in different layers independently and simultaneously, so it is feasible to apply a pipeline structure to improve the training throughput. Specifically, this pipeline structure improves the complexity of the training time from , which is the time complexity when using BP and stochastic gradient descent (SGD) for training, to , where is the number of training instances and is the number of hidden layers. Surprisingly, even though most of the parameters in AL do not directly interact with the target variable, training deep models by this method yields accuracies comparable to those from models trained using typical BP methods, in which all parameters are used to predict the target variable. Consequently, because of the scalability and the predictive power demonstrated in the experiments, AL deserves further study to determine the better hyperparameter settings, such as activation function selection, learning rate scheduling, and weight initialization, to accumulate experience, as we have done over the years with the typical BP method. In addition, perhaps our design can also inspire new network designs for deep learning. Our implementation is available at https://github.com/SamYWK/Associated_Learning.

Enterprise Information System (IS) is often reduced to a position to solve problems, most often administrative ones, by means of informatics. This paper considers another point of view of IS utility where it creates new opportunities and greatly expands the means to deal with complex change situations. The case studied in this paper is the enterprise reorganization. This exploratory research reveals the role of IS in such a critical situation. It unearths key IS modeling and architecture principles and discovers knowledge and methods of reasoning required to support IS evolution as an underlying way to the enterprise reorganization. It concludes with the emergence of a new research area: the Computer Aided Information Steering.

Simulating simulation-agnostic SysML models for enterprise information systems via DEVS