Roadmap from startup dev to formal MLE
This is a recopilation about my learning material to become a better engineer of data. I always work in computer vision, development, tracking and training of models, etc. But I want (just like you, if you've already gained more experience) formalize knowledge and learn about edge/frontier knowledge.
This is my list from actually I learn:
MLOps
Models lives in production, and are born from data. Be careful: explore how to work well with data and how models are tracked in production.
- Technical Debt in ML systems: https://papers.nips.cc/paper/2015/file/86df7dcfd896fcaf2674f757a2463eba-Paper.pdf
- From model serving to Data serving: https://www.youtube.com/watch?v=06-AZXmwHjo
- Challengues in deploy ML Systems: https://arxiv.org/pdf/2011.09926
Data Eng
Yes, you want to formalice the Data workflow and pipelines if you wanty to become a better ML engineer. Usually you work much more with data and not with models per se. Try to understood more about the data and you perform better in the 75% of your workflow, specially in
I recommend the course of DeepLearning.IA Introduction to Data engineering
For concepts:
Explore and extend a more detailed way of study data:
Usually your first problem is work with your data and your labels. I recommend to you to explore your data labels (speccialy in images) with a t-SNE plot. See this explaination of the application, I found them a very pretty application of t-SNE over images to show and explore:
https://csgaobb.github.io/Projects/DLDL.html
This may be a good way of evaluate the feasibility of your data/images.