The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Machine Learning as a Software Engineering Discipline with Dillon Erb - #404
8 snips
Aug 27, 2020 
Dillon Erb, Co-founder and CEO of Paperspace, discusses how to tackle the challenges of building scalable machine learning workflows. He emphasizes the importance of integrating software engineering principles into machine learning, focusing on reproducibility and effective workflow management. Dillon also highlights how MLOps platforms can assist developers, and explores the relationship between Gradient and Kubeflow. Additionally, he touches on innovations in machine learning visualization that enhance practical applications across various industries.
AI Snips
Chapters
Books
Transcript
Episode notes
Early TensorFlow Users on Paperspace
- Early Paperspace users inquired about using GPUs with TensorFlow, even before major cloud providers offered GPUs.
- This clued Dillon Erb into the growing importance of GPUs for deep learning and data processing beyond visual compute.
ML as a Software Engineering Discipline
- Dillon Erb believes machine learning is part of traditional software engineering, leveraging practices like CICD.
- While ML has unique aspects, core software principles like version control and staging environments still apply.
Notebooks and Reproducibility
- Use tools that emphasize reproducibility and version control, like Gradient's notebook lab.
- Treat notebooks as containers within a larger workflow, enabling reproducibility and history tracking through Docker commits.
