bauxitiego/ml_course

"Learning Machine Learning" Course, Bogotá, Colombia 2019 #LML2019

Learning Machine Learning Course

Learning Machine Learning Course, Bogotá, Colombia (© Josh Bloom; June 2019)

Following along

To execute the notebooks with all the dependencies installed, we'll use MyBinder. Here's the stable Binder link:

Bleeding edge (might take a long time to build an image):

If those don't work you can follow along using the static notebooks (ie., no kernel environment) at nbviewer site.

Format

4 days of lectures, approximately one per morning and one per afternoon. Intermittent breakout work by participants.

Lecture Outline

Introductions & Course Goals

• (Lecture 1) Computational and Inferential Thinking
  • Logistics
  • Data Science, Machine learning, and statistics introduction
  • Exploration: visualization & data preparation
  • Featurization, Dirty Data, and Natural Language Processing

Statistical Learning and Prediction

• (Lecture 2) Bayesian Inference

  • Introduction
  • MCMC, hierarchical Bayes
  • Gaussian Processes

• (Lecture 3) Supervised Learning

  • Regression: logistic regression, kNN, error estimation
  • Classification: Random Forest & LightGBM

Neutral Networks

• (Lecture 4) Neural Networks

  • Introductory algorithms and frameworks
  • Fully connected networks for regression and classification

• (Lecture 5) Deep Convolutional Neural Networks

  • imaging classification
  • sequence inferencing/classification

• (Lecture 6) Generative and Compressive Modeling

  • auto-encoders
  • GANs
  • surrogate emulation

Self-supervised Learning & Novelty Detection (Lecture 7)

• transfer learning & hyperparameter optimization
• Clustering approaches
• Dimensionality reduction
• Anomaly/Novelty detections with random forests
• Outlier detection for Timeseries

ML In the Real World

(Lecture 8)

• Business considerations
• Deployability, Scaling, and Maintainability
• Bias, Reproducibility, GDPR, and Ethics in ML
• What we didn't cover: Reinforcement Learning, etc.

Major Frameworks:

• Python 3.6 (or 3.7)
• numpy, scipy, seaborn
• sklearn
• keras/tensorflow
• spacy
• pymc3