WRA ML Curriculum – Overview & Progress

Introduction

I asked to 10 high school students: what comes up in your mind when you hear the term “artificial intelligence”?

‘2001 soace odyssey’

There are some unknown ones like ‘asdf’ or ‘asodfhj’ (which I believe is another way of saying “I don’t know,” a valid answer), but the general consensus is pretty clear: ‘cutting-edge’ and ‘intelligent’ ‘robots’ that often ‘destroy or replace humanity’ in Hollywood movies.

We are obscured by this Hollywood notion. The terminator movies warned us how an AI system would destroy humanity, and the Marvel movies expressed an idea that only a multimillionaire MIT graduate playboy with a tiny reactor attached to his chest can create such cutting-edge technology and even fails to create a safe one.

That was 2015 (Age of Ultron). We are in 2019. Four years in technology sector is quite a long time. It is not ‘cutting-edge’ anymore. We don’t have to graduate MIT to learn AI. There are programs that make AI as intuitive as stacking Lego blocks together.

Tensorflow is (primarily) a Python library that facilitates creating deep neural networks.

High school students can do this. If they can do it, why don’t we give them opportunities to learn?

Objectives

• Become aware of today’s prevalence of automated systems that utilize artificial intelligence and discuss the ethics and consequences of them
  • YouTube Recommended
  • Uber path optimization/demand prediction
  • Surveillance
  • Autonomous car
  • moral dilemma
• Learn how to handle data
  • Know the definition of data
  • learn types of data, depending on preference
    • Unstructured data
      • Images, videos
      • raw audio/spectrogram
      • text files
    • Structured/tabular data
  • learn to process data
    • Standardization/Normalization
    • Visualization
    • Augmentation
    • basic Numpy array handling
• Intuitively understand and learn to make machine learning models
  • Linear & logistic regression
  • Decision Trees
  • Neural Networks
    • MLP
    • CNN
    • RNN/LSTM
• recognize the importance of interdisciplinary collaboration
  • projects with diverse datasets
    • Science
      • biological/medical data
    • Literature/Language
      • Translation
      • Language model
    • Art
      • image
      • video
  • and more

Assessments & Evaluations

Finding a good way of assessing students in tech classes is always a challenge. Yes, I can simply give them daily quizzes over boring lectures, but that goes against the philosophy of this course–this course should be geared toward the practice, not the theory.

Luckily, in ML, there is an objective metric for comparison: the loss value. We can grade students based on how well their models work. So, the following are the ideas for grading.

• small-sized assessment equivalent to quizzes: N/A. It may not have one.
• medium-sized assessment equivalent to tests: Data MasterChef
  • Inspired by Kaggle competitions
  • Students are given a dataset relevant to what they have learned and tasked to create machine learning models.
  • Then, it will be graded holistically; accuracy is not everything.
    • ex) there is a chance that almost every team fails to create an effective model. In this case, they can explain the circumstances.
  • But, whoever reaches the “state-of-the-art” gets an extra credit as an incentive.
• Large-sized assessment equivalent to finals: ML Project
  • After the introduction, students will decide what project they will pursue over the course
  • students can do it either by themselves or as a team
  • In the end, they will present to the rest of the class about why, how, and what.

Resources & References

I don’t trust myself. Some may say I am just having impostor syndrome, but, trust me, I wouldn’t completely trust a high school student who is trying to create a official course in a place where he was just taught.

So, here’s my solution to that: getting inspiration from other reliable courses I have found.

MIT AI Ethics Education Curriculum

This course was created by Blakeley H. Payne with support from the MIT Media Lab Personal Robots Group, directed by Cynthia Breazeal.

Primarily targeting middle school students, the course focuses on introducing machine learning algorithms, giving a chance to think about how ML will shape the world, and showing them right ways to handle the fire. Activities include, but not limited to, AI Bingo, Ethical Matrix, Intro to Supervised ML & Algorithmic Bias, and YouTube Recommendation algorithm analysis.

This will serve really well as an introductory course, but the problem is that it is primarily designed for middle school students. Therefore, this will be only used to create the first 4-5 class periods as a hook.

Teachable Machine by Google

Create a ML model with zero lines of code. That pretty much sums up what this website does. The AI ethics course also uses this website for introducing supervised learning.

But, they are high school students; they can do more than this. I want them to be able to handle basics of Tensorflow at least. Just like the AI ethics course above, this will be only used for introduction.

Fast.ai : Making neural nets uncool again

Finally a website that is spittin’ straight facts.

Well, this website is more than just spitting straight facts. It is like neural network Khan Academy for high school students and/or undergrads (probably more toward undergrads, though, considering the difficulty).

As opposed to the middle school AI course, this one will be too difficult if I want to cover all in a year-long course especially since it deals with the fundamentals like gradient descent or backpropagation. I am not planning on teaching multivariable calculus to sophomores.

However, there are still “easy” sections that I can potentially refer to for my course. For example, some parts of the newest “A code first introduction to NLP” provide interesting tutorials without losing any interest over a flood of math equations.

Mr. Gerber and I are planning on visiting Carnegie Mellon University as well to see how they are introducing AI to undergraduates.