Comparing neural network training performance between Elixir and Python

The Conclusion from the paper:

    6 Conclusion
    In this article, we concluded that Elixir took more time in both datasets ana-      lyzed compared 
    to Python while demanding more Graphic RAM on peaks. We observed    that RAM usage is 
    an exception to the other metrics as Elixir con- sumed on average  42% and 33% less RAM 
    than Python, showing that Elixir has better memory management  for these test cases.
    As for usability, we observed that Python has much more documentation, material, 
    and available libraries for dataset preparation, model creation, and training.  This 
    was expected as Python is already much more consolidated in the machine learning 
    market. However, the Nx library was a significant improve- ment to Elixir, allowing 
    the language to be used for neural network creations much more efficiently than before Nx.

from the paper: "This work aims to compare the results of Python and Elixir on training convolutional neural networks (CNN) using MNIST and CIFAR10 datasets, concluding that Python achieved overall better results, and that Elixir is already a viable alternative."

I also want to note they are using older versions of Axion and Nx