Data-Centric AI Cookbook-let's prep that data

on YouTube ↗30m

" summary

Shows that tiny concrete fixes beat vague promises: a penguin dataset of ~5000 images contains unit errors (cm vs mm) that break models, imputing missing flipper lengths by species-and-sex mean improves performance more than hyperparameter tuning, and the workflow focuses on finding mislabeled or OOD samples via Cleanlab out-of-distribution detection and embedding visualization. Emphasize a representative validation set, version and relabel data easily, and prioritize correcting labels over collecting more noisy examples.

AI-written summary · about

" chapters (11)

1. 0:00Data-centric AI introduction
2. 1:55Data quality matters most
3. 5:42Big data to good data
4. 8:32Unstructured data challenges
5. 11:21Foundation models dependency
6. 15:38Quantity vs quality in data
7. 17:23Good data characteristics
8. 19:32Tools for data issues
9. 24:10Representative train-test splits
10. 26:15Easy data iteration
11. 27:20Summary

" related talks

• 33mKeys to Data-Centric AI: Asking AI to Improve its own Dataset.Jonas Mueller · Scale By the Bay 2023
• 41mModels Schm-odels: Why you Should Care about Data-Centric AIMarysia Winkels · PyData London 2022
• 11mData-centric AI: A new paradigmMeenu Gaba · Women in ML Symposium 2023
• 28mData Curation: Transforming Bytes into AI GoldRichard Decal · AI in Production 2024
• 39mDoing Data: The Critical Process of Data Preparation (...)Ellen Friedman · Berlin Buzzwords 2019
• 37mBest Practices for Building AI DatasetsBradley Arsenault · Big Data Conference Europe 2019
• 26mUnlocking the Power of Data-Centric AI.Fabiana Clemente, Gonçalo Martins Ribeiro · Scale By the Bay 2023
• 21mDataEngBytes 2023 - SYD-T2-04 - Suneeta MallSuneeta Mall · DataEngBytes Sydney 2023