Difference between revisions of "Machine Learning"
|Line 63:||Line 63:|
#''Hidden Technical Debt in Machine Learning Systems'', [[File:NIPS-5656-hidden-technical-debt-in-machine-learning-systems.pdf]].
Revision as of 21:52, October 7, 2018
Machine learning, everybody’s doing it. Scientists are applying machine learning to their scientific algorithms and using the results to justify various conclusions. Is machine learning the silver bullet that will help us answer our scientific questions?
The purpose of this cluster is to educate ourselves and the ESIP community about machine learning through asking and answering questions and sharing experiences and resources. The scope of the cluster includes topics like the following:
- What is machine learning? What can machine learning do? How is machine learning different from data science? From data analytics?
- What types of machine learning algorithms are there and how do they compare to each other? Under what circumstances is each best applied and/or not applicable?
- What are symbolic and subsymbolic approaches to machine learning and their subtypes?
- What are machine learning SWOTs: strengths, weaknesses, opportunities, threats? Some considerations include:
- models, model choices and biases
- p-hacking, data dredging
- What machine learning tools are available?
- In particular, regarding deep learning/neural networks
- How do supervised, semi-supervised, and unsupervised networks differ?
- How to integrate truth labeled data?
- Under what circumstances is it okay to not understand what the network learned? When is it not okay?
- How to deal with a lack of training data?
- How to (try to) understand what was learned?
Possible cluster outputs could include:
- A machine learning tool survey and SWOT analysis
- Training material, on line [and maybe a guide to available platforms and resources? e.g., using AWS ML platform]
- Training and/or other sessions at ESIP
- Recommendations regarding the application of ML in earth and space sciences