Evaluating our one-week hands-on workshop on applying machine learning to research.
5 Stimulate diverse teams and be aware of (gender-)biases.****We wanted to attract diverse teams and in particular avoid gender-biases since IT and machine-learning are (still) heavily male dominated. It is very easy to simply add a sentence saying that you welcome female participants, but that’s probably not going to do the job if the rest of the text is full of gender-coded works (“competitive”, “hacking”, …).
There are apps to check your text for masculine-coded words (e.g. http://gender-decoder.katmatfield.com/), but they only cover a short list of words and don’t get the context. So they can primarily serve to give you an idea of how subtle things can be. More particularly, we avoided wording that could make the workshop sound competitive, which started with not officially calling it a hackathon**!
Although we can’t prove this did the trick, I definitely think that the many discussions we had with a wide range of people about this subject before setting up the workshop really helped us a lot. And the outcome was great, too. Out of all the team members, **40% were female **(of 101 applicants in total). Among the selected teams we worked with during the workshop this number even went up to 59% female participants (of the final 27 participants).
(See also this blog post on hackathons with high female participant numbers)
6** Join forces with SURF.**** Not only did Maxwell Cai from SURF make a very valuable addition to our list of mentors (also giving an interesting introductory lecture on ML), we also benefited from the compute infrastructure provided by SURF. During the workshop, we had access to GPUs via simple Jupyter notebooks that we used intensively to train our deep learning models (for 4 teams using deep learning we had 8 GTX 1080Ti GPUs which was fully sufficient for the type of tasks we had).
Things we’d like to improve:
The positives notwithstanding, there is of course also a long list of things we can improve on. Luckily, most items on that list were clearly very minor issues. The more important lessons we learned were:
- Better representation of some research field****There are disciplines within the humanities and social sciences where I believe machine learning can add a lot of value. Out of 25 applications, only 2 came from those fields. So next time we will certainly make an effort to better target humanities and social science research institutions.
- Extend the pre-workshop training.**We had thought about it and did offer a free one-day Python course in the week before the workshop, which indeed turned out to be very helpful. But to give participants with less coding experience and little to no prior Python experience a bit more time to prepare, it would be better to organize such a training a bit more in advance. Next time we plan to add a separate onboarding day! **This will contain presentations of each team, first discussion rounds with the mentors and both basic and more intermediate Python courses to help people that have worked with it a little or which transition from very decent skills in similar languages (we had a lot of people working with ‘R’ or MATLAB).
- Stimulate more interaction between the teams. **This might not be so simple since the teams came from widely different research fields and had different levels of computational skills, from programming beginners to software developers. During the workshop we had lunch together, we went out for drinks, and we had standup updates to report on what each team was busy with. Still, next time we hope to be able to include more moments for teams to interact and learn from what other teams have done.
- **Some technical stuff:We quickly realized that it was essential to have at least one big screen per team together with flip-charts or whiteboards. It was easy to fix, but next time I would try to arrange more screens and sketching options.
