Why and how make we ML reproducible - Jesper Dramsch

There are different areas where researchers can increase the quality of their research artefacts that use ML. These increases in quality are achieved by using existing solutions to minimize the impact these methods take on researcher productivity.

This talk loosely covers the topics I discussed in my Euroscipy tutorial, which will be used for the interactive session here:

github.com/JesperDramsch/ml-for-science-reproducib…

Topics covered:

Why make it reproducible?
• Model Evaluation
• Benchmarking
• Model Sharing
• Testing ML Code
• Interpretability
• Ablation Studies

These topics are used as examples of “easy wins” researchers can implement to disproportionately improve the quality of their research output with minimal additional work using existing libraries and reusable code snippets.

This talk was part of the workshop "Real-world Perspectives to Avoid the Worst Mistakes using Machine Learning in Science" at Pydata Global 2022, organised by Jesper Dramsch, Gemma Turon, and ‪@ValerioMaggio‬.

The workshop has received funding from the Software Sustainability Institute through the 2022 fellowship programme received by Jesper Dramsch.
dramsch.net/ssi

Find the programme of the workshop, transcripts and more resources here:
realworld-ml.xyz/

#MachineLearning #PydataGlobal