01-foundations.Rmd

# 1 Foundations

**Learning objectives:**

- Understand the purpose for the book
- Know the names of any python packages relevant to the book
- Have an overview of
  - what machine learning is
  - the different types of machine learning problem
  - different types of uncertainty

## Origin of the book {-}

2012
- Deep learning revolution
- ImageNet image classification challenge

Hardware advances
- GPUs

Crowd sourcing data collection
- Amazon Mechanical Turk

Unifying lens for the book is “Probabilistic modeling and Bayesian decision theory”

## Python packages {-}

These packages are relevant to the book:

- NumPy
  - multidimensional arrays & computational maths
- Scikit-learn
  - machine learning toolkit
- JAX
  - numerics on tensors and automatic differentiation
- PyTorch
  - tensor library for deep learning
- TensorFlow
  - framework for building ML pipelines (?)
- PyMC
  - probabilistic programming MCMC etc

## Notebooks for the book {-}

[github](https://github.com/probml/pyprobml/blob/auto_notebooks_md/notebooks.md)

The notebooks auto-open in Colab

They show how to make the figures for the book

## What is Machine Learning {-}

To discuss:

- What is machine learning?
- What is machine learning from a probabilistic perspective?
- Why take a probabilistic approach to ML?

## Types of Machine Learning Problem {-}

- supervised learning (classification, regression, )
- unsupervised learning (clustering, latent variables)
- reinforcement learning (learn how to interact with env)

## Types of 'uncertainty' {-}

- Input/Output mapping isn’t known or knowable (model uncertainty)
- Randomness is intrinsic in the mapping (data uncertainty)

## Meeting Videos {-}

### Cohort 1 {-}

`r knitr::include_url("https://www.youtube.com/embed/MxdYkiNTGKU?si=O5b8HWZVlm5p23Y-")`

<details>
<summary> Meeting chat log </summary>

```
00:04:09    Derek Sollberger (he/him):    Hello!
00:04:41    Sohan Aryal:    Hello everyone,
first time actually involving in a book club,
00:05:08    jRad:    Hi, second one for me, been quite a while!
00:05:20    Sohan Aryal:    Reacted to "Hi, second one for m..." with 😯
00:54:33    Derek Sollberger (he/him):    Should the same person handle each two-week pair?
00:59:33    Derek Sollberger (he/him):    If no one minds, I would like to volunteer for the second
  half of the LDA chapter (on Bayesian classification)
01:05:26    Rahul:    Thank you very much, Russ!
01:05:29    Schafer, Toryn:    Thanks!
01:05:34    Derek Sollberger (he/him):    Thank you all.  Thanks Russ!
01:05:36    David Díaz:    Thanks!
01:05:36    Russ Hyde:    Thanks everyone
```
</details>