Where we are now

class: center, middle, inverse, title-slide

# Where we are now
### Ken Arnold
### 2021-03-17

---

# Increasing Structure

* Initial goal: practice more self-directed learning (e.g., for industry)
* but... this is leaving some people behind.
* so:
  * main part of the course is structured
  * Homework and Project is unstructured / under-specified
  
---

# Main elements

* Active reading: primary way of "covering" material
  * Watch lecture videos if it helps you
  * Ask *way* more questions! (e.g., "this section seems really complicated... how important is it really?")
* Fundamentals notebooks: practice with basic skills (working out at the gym)
* Labs: more narrative exercises (team drills)
* Homework: application and extension (playing the game)

---

# New Fundamentals notebooks pushed in the last week

* practice with JSON (optional)
* manipulating batches of images
* computing the gradient of a function
* practice with softmax and sigmoid

---

# Poll

* git? Moodle?
* pace?

---

# Homework 2

Goal:

* Interview: "I trained a classifier on MNIST"
* Connect ch1/2 high-level classifier usage with ch4/5 low/mid-level gradient descent
* Practice with tensor shapes, backprop, cross-entropy loss, learning rate tuning, etc.

---

# Review Lab 2

---

# Lab 3

Goals:

* "Hello world" of learning from a data stream (Stochastic Gradient Descent)
* Recognize problematic situations in numerical computing
* Practice debugging techniques

Strategy:

* Don't avoid the errors. If you've prematurely fixed something, go back and break it again and see what it does to the output