[C3] Structuring Machine Learning Projects

First week: machine learning strategies (1) (ML Strategy (1))

Why is the ML strategy? (Why ML Strategy)

Orthogonalization (orthogonalization)

Single number evaluation index (Single number evaluation metric)

Meet and optimization metrics (Satisficing and Optimizing metric)

Divided into a training set, collection development, test set (Train / dev / test distributions)

Development and test sets the size (Size of the dev and test sets)

When changing the set development / test set and evaluation indicators (When to change dev / test sets and metrics)

Why is the human performance (Why human-level performance?)

Can avoid bias (Avoidable bias)

Understanding of human performance (Understanding human-level performance)

More than human performance (Surpassing human-level performance)

Improve the performance of your model (Improving your model performance)

Second week: machine learning strategies (2) (ML Strategy (2))

Error Analysis (Carrying out error analysis)

Clear Label erroneous data (Cleaning up incorrectly labeled data)

Quickly build your first system, and iterating (Build your first system quickly, then iterate)

On a different distribution of the training set and test set (Training and testing on different distributions)

Analysis of variance and bias data distribution does not match (Bias and Variance with mismatched data distributions)

Processing data mismatch (Addressing data mismatch)

Transfer learning (Transfer learning)

Multi-task learning (Multi-task learning)

What is the end of deep learning? (What is end-to-end deep learning?)

Whether end-depth learning methods (Whether to use end-to-end deep learning)