Data Science and Neural Networks - Big Love or just a Love Affair?
summary
Guidance on when neural networks win versus simpler models: neural nets approximate complex functions but need large, well‑labelled and representative datasets; labels can be costly (Mechanical Turk) and distributional shifts break models (credit scoring example). Practical wins listed are image classification and speech recognition where labels are easy or abundant; time series forecasting and customer‑service text need care. Prefer transparent models like decision tree transparency when interpretability or sparse data matters. Training often uses frameworks from TensorFlow/UCI experiments.
AI-written summary · about
chapters (10)
- 0:00Data science & neural networks
- 1:50AI network roots
- 4:35Simple networks & uses
- 7:12Training with labels
- 11:08Data quality impact
- 14:11When to avoid NNs
- 16:25Image classification
- 19:07Choosing models based on data quality
- 21:13Neural networks future
- 24:50Conclusion & Q&A
related talks
35mIntelligent Ruby: Getting Started with Machine Learning
26mThe scaling laws of AI Why neural networks continue to grow
01h 02mTutorial on Deep Learning I
41mModels Schm-odels: Why you Should Care about Data-Centric AI
41mSpotlight on AI use cases and innovators around the world
48mKeynote – Sameer Singh
42mWomen’s Networks: How Neural Networks Help the Beauty Industry
06mMachine learning is about your data and deployment