CalcGraph: taming the high costs of deep learning using models

in Software and Systems Modeling (2022)

Models based on differential programming, like deep neural networks, are well established in research and able to outperform manually coded counterparts in many applications. Today, there is a rising ... [more ▼]

Models based on differential programming, like deep neural networks, are well established in research and able to outperform manually coded counterparts in many applications. Today, there is a rising interest to introduce this flexible modeling to solve real-world problems. A major challenge when moving from research to application is the strict constraints on computational resources (memory and time). It is difficult to determine and contain the resource requirements of differential models, especially during the early training and hyperparameter exploration stages. In this article, we address this challenge by introducing CalcGraph, a model abstraction of differentiable programming layers. CalcGraph allows to model the computational resources that should be used and then CalcGraph’s model interpreter can automatically schedule the execution respecting the specifications made. We propose a novel way to efficiently switch models from storage to preallocated memory zones and vice versa to maximize the number of model executions given the available resources. We demonstrate the efficiency of our approach by showing that it consumes less resources than state-of-the-art frameworks like TensorFlow and PyTorch for single-model and multi-model execution. [less ▲]

F-Alloy: a relational model transformation language based on Alloy

in Software and Systems Modeling (2019), 18(1), 213-247

Model transformations are one of the core artifacts of a model-driven engineering approach. The relational logic language Alloy has been used in the past to verify properties of model transformations. In ... [more ▼]

Model transformations are one of the core artifacts of a model-driven engineering approach. The relational logic language Alloy has been used in the past to verify properties of model transformations. In this paper we introduce the concept of functional Alloy modules. In essence a functional Alloy module can be viewed as an Alloy module representing a model transformation. We describe a sublanguage of Alloy called F-Alloy specifically designed to concisely specify functional Alloy modules. The restrictions on F-Alloy’s syntax are meant to allow efficient execution of the specified transformation, without the use of backtracking, by an adapted interpretation algorithm. F-Alloy’s semantics is given in this paper as a direct translation to Alloy; hence, F-Alloy specifications are also analyzable using the powerful automatic analysis features of Alloy. [less ▲]

Model-Based Simulation of Legal Policies: Framework, Tool Support, and Validation

in Software and Systems Modeling (2018), 17(3), 851-883

Simulation of legal policies is an important decision-support tool in domains such as taxation. The primary goal of legal policy simulation is predicting how changes in the law affect measures of interest ... [more ▼]

Simulation of legal policies is an important decision-support tool in domains such as taxation. The primary goal of legal policy simulation is predicting how changes in the law affect measures of interest, e.g., revenue. Legal policy simulation is currently implemented using a combination of spreadsheets and software code. Such a direct implementation poses a validation challenge. In particular, legal experts often lack the necessary software background to review complex spreadsheets and code. Consequently, these experts currently have no reliable means to check the correctness of simulations against the requirements envisaged by the law. A further challenge is that representative data for simulation may be unavailable, thus necessitating a data generator. A hard-coded generator is difficult to build and validate. We develop a framework for legal policy simulation that is aimed at addressing the challenges above. The framework uses models for specifying both legal policies and the probabilistic characteristics of the underlying population. We devise an automated algorithm for simulation data generation. We evaluate our framework through a case study on Luxembourg’s Tax Law. [less ▲]

A generic model decomposition technique and its application to the Eclipse modeling framework

in Software and Systems Modeling (2015), 14(2), 921-952

Model-driven software development aims at easing the process of software development by using models as primary artifacts. Although less complex than the real systems they are based on, models tend to be ... [more ▼]

Model-driven software development aims at easing the process of software development by using models as primary artifacts. Although less complex than the real systems they are based on, models tend to be complex nevertheless, thus making the task of handling them non-trivial in many cases. In this paper we propose a generic model decomposition technique to facilitate model management by decomposing complex models into smaller sub-models that conform to the same metamodel as the original model. The technique is based upon a formal foundation that consists of a formal capturing of the concepts of models, metamodels, and model conformance; a formal constraint language based on EssentialOCL; and a set of formally proved properties of the technique. We organize the decomposed sub-models in a mathematical structure as a lattice, and design a linear-time algorithm for constructing this decomposition. The generic model decomposition technique is applied to the Eclipse Modeling Framework (EMF) and the result is used to build a solution to a specific model comprehension problem of Ecore models based upon model pruning. We report two case studies of the model comprehension method: one in BPMN and the other in fUML. [less ▲]

Applying UML/MARTE on industrial projects: challenges, experiences, and guidelines

in Software and Systems Modeling (2014)

VPML: an approach to detect design patterns of MOF-based modeling languages

in Software and Systems Modeling (2013)

Does aspect-oriented modeling help improve the readability of UML state machines?

in Software and Systems Modeling (2012)

Aspect-oriented modeling (AOM) is a relatively recent and very active field of research, whose application has, however, been limited in practice. AOM is assumed to yield several potential benefits such ... [more ▼]

Aspect-oriented modeling (AOM) is a relatively recent and very active field of research, whose application has, however, been limited in practice. AOM is assumed to yield several potential benefits such as enhanced modularization, easier evolution, increased reusability, and improved readability of models, as well as reduced modeling effort. However, credible, solid empirical evidence of such benefits is lacking. We evaluate the “readability” of state machines when modeling crosscutting behavior using AOM and more specifically AspectSM, a recently published UML profile. This profile extends the UML state machine notation with mechanisms to define aspects using state machines. Readability is indirectly measured through defect identification and fixing rates in state machines, and the scores obtained when answering a comprehension questionnaire about the system behavior. With AspectSM, crosscutting behavior is modeled using so-called “aspect state machines”. Their readability is compared with that of system state machines directly modeling crosscutting and standard behavior together. An initial controlled experiment and a much larger replication were conducted with trained graduate students, in two different institutions and countries, to achieve the above objective. We use two baselines of comparisons—standard UML state machines without hierarchical features (flat state machines) and standard state machines with hierarchical/concurrent features (hierarchical state machines). The results showed that defect identification and fixing rates are significantly better with AspectSM than with both flat and hierarchical state machines. However, in terms of comprehension scores and inspection effort, no significant difference was observed between any of the approaches. Results of the experiments suggest that one should use, when possible, aspect state machines along with hierarchical and/or concurrent features of UML state machines to model crosscutting behaviors. [less ▲]