How Do Deep-Learning Framework Versions Affect the Reproducibility of Neural Network Models?

In the last decade, industry’s demand for deep learning (DL) has increased due to its high performance in complex scenarios. Due DL method’s complexity, experts and non-experts rely on blackbox software packages such as Tensorflow Pytorch. The frameworks are constantly improving, new versions released frequently. As a natural process development, contain improvements/changes methods their implementation. Moreover, may be bug-polluted, leading model decreasing or stopping from working. aforementioned changes implementation can lead variance obtained results. This work investigates effect of different major releases these performance. We perform our study using variety standard datasets. Our shows that users should consider changing framework version affect they possibility bug-polluted before starting debug source code had an excellent change. also importance virtual environments, Docker, when delivering product clients.