Taming the Deep Learning Beast, Hands on with PyTorch

The overall aim of the session was for attendees to leave with a decent understanding of what ‘training a model’ actually means in the context of deep learning.

Last week, we attended the Cambridge Python User Group Meetup at the Raspberry Pi offices on Hills Road in Cambridge. The Python meetup is a great place for people of all levels of Python knowledge to come together to learn more about the programming language and meet other like-minded people. The sessions are usually arranged on the first Tuesday of each month at 7.00pm. This month, I led a workshop for the 70 attendees titled ‘Taming the Deep Learning Beast: Hands on with PyTorch’. In the session, attendees built and trained an image classifier using the deep learning framework PyTorch.

We used PyTorch as it a great tool which combines the power of the most advanced deep learning frameworks with an intuitive, pythonic API that makes writing machine learning code a pleasure and not a pain. It was really rewarding to give this workshop to a room full of eager participants who, I hope, enjoyed the session. By using Google Colab, every participant was able to train their models on a GPU without any hassle of installing the appropriate libraries. All I had to do was share a URL! We had lots of great questions both during and after the presentation.

What did I learn? The workshop was also a great learning experience for me. I learnt the importance of pitching level; with content needing to be easy enough for everyone to follow, but also challenging enough so that people are stimulated. This is a difficult balance to achieve, particularly with something as technical as a crash course in deep learning! I’ve found a key to doing so is to make things as interactive as possible. I find even something like asking the audience to answer really simple questions that they all know the answer to is a great way to get them involved in the discussion. People tend to feel very nervous about answering a question in front of a large audience but by making it so easy that no one can get it wrong, they are nearly always bought out of their shell. We’re hiring! As well as the opportunity to lead a workshop, I also had the chance to let attendees know about our current vacancies at Speechmatics. We’re growing considerably and as such, have lots of open roles which are very relevant to keen Python developers. If you’re interested in learning more about our opportunities, check out our careers page. Sam Ringer, Machine Learning Engineer, Speechmatics