Deep learning at scale: A field manual
summary
Argues that data is code and must get the same practices as software: versioning via immutable stores or copy‑on‑write snapshots, metadata captured for model provenance and lineage, and abstraction layers over unavoidable silos so models see unified datasets. Recommends building unit data tests alongside test sets to validate specific failure modes, tracking every dataset like source control, and treating deployment scaling as a systems problem of data flow, not just bigger models.
AI-written summary · about
chapters (3)
related talks
41mCan Data & ML really learn from software engineering?
26mCompleting the machine learning loop
59mDevOps for Machine Learning Projects
31mReliable Data for Large ML Models: Principles and Practices
26mUnlocking the Power of Data-Centric AI.
20mBetter Care and Feeding of Machine Learning Models
38mBuilding robust machine learning systems
27mDataEngBytes 2023 - BRI-T1-07 - Joe Reis