Modern AI practices all strive towards the same goal: better results. In the context of deep learning, the term "results"often refers to the achieved accuracy on a competitive problem set. In this paper, we adopt an idea from the emerging field of Green AI to consider energy consumption as a metric of equal importance to accuracy and to reduce any irrelevant tasks or energy usage. We examine the training stage of the deep learning pipeline from a sustainability perspective, through the study of hyperparameter tuning strategies and the model complexity, two factors vastly impacting the overall pipeline's energy consumption. First, we investigate the effectiveness of grid search, random search and Bayesian optimisation during hyperparameter tuning, and we find that Bayesian optimisation significantly dominates the other strategies. Furthermore, we analyse the architecture of convolutional neural networks with the energy consumption of three prominent layer types: convolutional, linear and ReLU layers. The results show that convolutional layers are the most computationally expensive by a strong margin. Additionally, we observe diminishing returns in accuracy for more energy-hungry models. The overall energy consumption of training can be halved by reducing the network complexity. In conclusion, we highlight innovative and promising energy-efficient practices for training deep learning models. To expand the application of Green AI, we advocate for a shift in the design of deep learning models, by considering the trade-off between energy efficiency and accuracy.

@en

The batch size is an essential parameter to tune during the development of new neural networks. Amongst other quality indicators, it has a large degree of influence on the model’s accuracy, generalisability, training times and parallelisability. This fact is generally known and commonly studied. However, during the application phase of a deep learning model, when the model is utilised by an end-user for inference, we find that there is a disregard for the potential benefits of introducing a batch size. In this study, we examine the effect of input batching on the energy consumption and response times of five fully-trained neural networks for computer vision that were considered state-of-the-art at the time of their publication. The results suggest that batching has a significant effect on both of these metrics. Furthermore, we present a timeline of the energy efficiency and accuracy of neural networks over the past decade. We find that in general, energy consumption rises at a much steeper pace than accuracy and question the necessity of this evolution. Additionally, we highlight one particular network, ShuffleNetV2 (2018), that achieved a competitive performance for its time while maintaining a much lower energy consumption. Nevertheless, we highlight that the results are model dependent.@en

Software reuse is a widely adopted practice among both researchers and practitioners. The relation between security and reuse can go both ways: a system can become more secure by relying on mature dependencies, or more insecure by exposing a larger attack surface via exploitable dependencies. To follow up on a previous study and shed more light on this subject, we further examine the association between software reuse and security threats. In particular, we empirically investigate 1244 open-source projects in a multiple-case study to explore and discuss the distribution of security vulnerabilities between the code created by a development team and the code reused through dependencies. For that, we consider both potential vulnerabilities, as assessed through static analysis, and disclosed vulnerabilities, reported in public databases. The results suggest that larger projects in size are associated with an increase on the amount of potential vulnerabilities in both native and reused code. Moreover, we found a strong correlation between a higher number of dependencies and vulnerabilities. Based on our empirical investigation, it appears that source code reuse is neither a silver bullet to combat vulnerabilities nor a frightening werewolf that entail an excessive number of them.

@en

Reuse is a common and often-advocated software development practice. Significant efforts have been invested into facilitating it, leading to advancements such as software forges, package managers, and the widespread integration of open source components into proprietary software systems. Reused software can make a system more secure through its maturity and extended vetting, or increase its vulnerabilities through a larger attack surface or insecure coding practices. To shed more light on this issue, we investigate the relationship between software reuse and potential security vulnerabilities, as assessed through static analysis. We empirically investigated 301 open source projects in a holistic multiple-case methods study. In particular, we examined the distribution of potential vulnerabilities between the native code created by a project’s development team and external code reused through dependencies, as well as the correlation between the ratio of reuse and the density of vulnerabilities. The results suggest that the amount of potential vulnerabilities in both native and reused code increases with larger project sizes. We also found a weak-to-moderate correlation between a higher reuse ratio and a lower density of vulnerabilities. Based on these findings it appears that code reuse is neither a frightening werewolf introducing an excessive number of vulnerabilities nor a silver bullet for avoiding them.

@en