Due to rapid improvements in computing performance and the amount of data that is accumulated, deep learning is gaining strength. Accordingly, IT Giants develop their deep learning framework to provide developers with a development environment, such as Google’s Tensorflow and Facebook’s PyTorch.

Figure 1. Deep learning frameworks

Each framework has its characteristics and strength, and they are properly used for appropriate purposes. For engineers with research purposes, it is necessary to note Figure 2. According to data from RISELab, the recent trend in the papers uploaded on arXiv.org shows that TensorFlow and PyTorch are mainly used in research purposes. It seems that choosing Tensorflow or PyTorch would be the best choice for people who are planning to dive in deep learning. It will allow understanding other’s research more easily and quickly.

Figure 2. The number of papers posted on arXiv.org that mention each framework. Source: Data from RISELab and graphic by Ben Lorica.

In the future, I will select various topics and upload a series of tutorials, and code will also be uploaded in two versions, Tensorflow and PyTorch. First, I will upload a post about basic topics, such as the MNIST tutorial. After that, I will discuss energy-related time series and reinforcement learning topics, which are my research interests. I hope this will help engineers who are new to using deep learning field or would like to be more adept at dealing with the frameworks.