Classification

AI 101

Classification

Why classification?

Definitely AI topic!
- Recall Imagenet
- “Is this image a cat or dog?”
It is the basis for some more advanced ML/AI stuff.
We’ll use it within our labs.

Carolus Linnæus

High School Biology

The first time I heard of classification it was high school biology.
- Y’all already know I love a dichotomous key.
I was told Carolus Linnæus invented the idea of categorizing things.
- This seems unlikely to be true but we can go with it.
Specifically, Carl (we’re on a first name basis) classified living things into species.

Two Species

Felis catus

Homo sapiens

Life is complicated

Classification into species is a complex and multi-faceted process.
In that respect, it is the negation of American action horror film production.
So we study the simpler system: specifically, the Alien vs. Predator films.
- Specifically, from 1976 to 2024 (since I haven’t been keeping up the past few years).

Example

We Begin…

…in 1979 with the release of Alien, widely regard as the greatest film of the post-silent era by instructors of this class.

Classification

In order to differentiate Alien from latter installments, we make a few notes.
- Alien starred Sigourney Weaver*
- Alien was directed by Ridley Scott
- Alien contains the term “Alien” in its name.

Timeline

We can place Alien on a timeline as follows:

Code

import matplotlib.pyplot as plt

def plot(films):

    years = list(range(1976, 2027))  # Create a list of years from 1976 to 2024

    plt.figure(figsize=(9,1)) # Adjust figure size for better visualization
    plt.plot(years, [0] * len(years), color='black', linewidth=2) # Plot a horizontal line

    def add_film(name, year, color="green"):
      plt.annotate(
          name,
          xy=(year, 0.0),
          xytext=(year, 0.2),
          arrowprops=dict(color=color, shrink=0.05, width=1, headwidth=8),
          horizontalalignment='center',
          verticalalignment='bottom',
          fontsize=10,
          color=color
      )
    
    [add_film(*film) for film in films]
    
    plt.yticks([]) # Remove y-axis ticks as they are not relevant for a horizontal line
    plt.grid(True, axis='x', linestyle='--', alpha=0.7)
    plt.xlim(1976, 2027) # Set x-axis limits slightly beyond the data range
    plt.show()


films = [
    ["Alien", 1979],
]


plot(films)

An Aside

Alien featured not one but two high profile depiction of artifical intelligence:
- MU/TH/UR, the ship computer
- Ash, a seemingly human crew member
In fact, it raises important questions about AI ethics, especially related to labor issues.

Special Order 937

At one point in the film, the crew of spacecraft bring onboard a dangerous alien organism.
The corporate ownership of the vessel which to retrieve the organism, despite the danger it poses to the crew.
This leads to a famous sequence in cinema and AI history:

Special Order 937

Priority one — Ensure return of organism for analysis. All other considerations secondary. Crew expendable

Commands

This command, issued by MU/TH/UR to Ash, a “synthetic” passing as human, led to Ash bamboozling the human crew into various unsafe circumstances, and ultimately killing many of them.
This plot point raises interesting questions that remain unanswered!

Up Next

After a series of studio mishaps, the next official franchise film is Aliens in 1986.

Aliens is about how director James Cameron (Titanic, Avatar, Terminator) does not understand art.

Classification

In order to differentiate Alien from latter installments, we make a few notes.
- Aliens starred Sigourney Weaver
- Aliens was not directed by Ridley Scott
- Aliens contains the term “Alien” in its name.

Timeline

We can place Aliens on a timeline as follows:

Code

films = [
    ["Alien", 1979],
    ["Aliens", 1986],
]

plot(films)

Alien 3 and Alien Resurrection were the next films in the franchise
Both even worse and less interesting than Aliens, so we’ll just add them and move on.
- Both starred Sigourney Weaver
- Both were not directed by Ridley Scott
- Both contained the term “Alien” in their name.

Timeline

We can the Sigourney Weaver Alien films on a timeline as follows:

Code

films = [
    ["Alien", 1979],
    ["Aliens", 1986],
    ["Alien 3", 1992],
    ["Alien Res.", 1997],
]

plot(films)

The Big Mistake

The 2000s were a time of silliness and goofery.
And from the glorious excess can a blissful union of two great franchises.
- Alien, a four film franchise with one good film.
- Predator a two film franchise starring two future governors.
  - Arnold Schwarzenegger (R-CA)
  - Jesse Ventura (Reform-MN)

Alien vs. Predator and AVPR: Alien vs. Predator - Requiem were released in 2004 and 2007, respectively.
While playful romps, they broke from tradition and switched from women to men as leads, and were set in present rather than future (both borrowing from Predator).
- Both did not star Sigourney Weaver
- Both were not directed by Ridley Scott
- Both contained the term “Alien” in their name.

Timeline

We can add the AVP films on a timeline as follows:

Code

films = [
    ["Alien", 1979],
    ["Aliens", 1986],
    ["Alien 3", 1992],
    ["Alien Res.", 1997],
    ["AVP", 2004, "darkgoldenrod"],
    ["AVPR", 2007, "darkgoldenrod"],
]

plot(films)

Reverse Reverse

While both AVP films would soon come into canon conflict with the Ridley Scott films, for a brief shining moment they retroactively brought two addition films into the now combined franchise.
- Predator (1987)
- Predator 2 (1990)
The less said about them the better.

Classification

Both did not star Sigourney Weaver
Both were not directed by Ridley Scott
Both did not contain the term “Alien” in their name.
Both do contain the term “Predator” in their name.
- Like AVP films but unlike the other Alien films.

Timeline

We can add the pre-AVP Predator films.

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["AVP", 2004, "darkgoldenrod"],
    ["AVPR", 2007, "darkgoldenrod"],
    ["P", 1987, "red"],
    ["P2", 1990, "red"],
]

plot(films)

Modern Era

The AVP films renewed interest in the Predator sub-franchise, with several new releases.
- There’s post 2024 releases I haven’t considered as I haven’t watched them.
They are:
- Predators (2010) - the first “off-Earth” franchise film.
- The Predator (2018) - a soft reboot
- Prey - a prequel film and the first franchise film to star a woman in the leading role.

Timeline

We can add the “modern” Predator films.

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["AVP", 2004, "darkgoldenrod"],
    ["AVPR", 2007, "darkgoldenrod"],
    ["P", 1987, "red"],
    ["P2", 1990, "red"],
    ["P's", 2010, "darkred"],
    ["The", 2018, "darkred"],
    ["Prey", 2022, "darkred"],
]

plot(films)

Return of the King 👑

2012 marked the triumphant return of Ridley Scott the Alien franchise with Prometheus, a retelling of Frankenstein in the Alien universe raising essential questions about life, control, and - again - AI.

David

David is an android serving as a butler, maintenance man, and surrogate son to his creator… David is obsessed with the concept of creating life of his own. After Weyland is killed, David is freed from servitude…

Inflection Point

A major plot point in Prometheus is the radicalization of David against human creators, portrayed through a repartee with human scientist.

David : Why do you think your people made me?
Human : We made you because we could.
David : Can you imagine how disappointing it would be for you to hear the same thing from your creator?
Human : I guess it’s good you can’t be disappointed.

Frankenstein

Frankenstein; or, The Modern Prometheus written by Mary Shelley in 1818 is widely regarded as the greatest original English text by instructors of this class.

Classification

Prometheus did not star Sigourney Weaver…
- …but did feature a woman in a leading role (Noomi Rapace)
Was directed by Ridley Scott
Did not contain the term “Alien” in their name.
- Or “Predator”

A Wrinkle in Time

While both the AVP films and Prometheus may exist in the same universe as Alien, they most exist together.
- In AVP films, “aliens” are present on earth in the modern day.
- In Prometheus, the “aliens” are genetic engineered off-planet by David.

Timeline

We can add Prometheus

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["AVP", 2004, "darkgoldenrod"],
    ["AVPR", 2007, "darkgoldenrod"],
    ["P", 1987, "red"],
    ["P2", 1990, "red"],
    ["P's", 2010, "darkred"],
    ["The", 2018, "darkred"],
    ["Prey", 2022, "darkred"],
    ["Prom.", 2012, "darkgreen"],
]

plot(films)

Canon Timeline

Through properly Prometheus is mutually exclusive all Predator films.

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["Prom.", 2012, "darkgreen"],
]

plot(films)

Unfortunately, Prometheus was subject to fairly extreme studio interference (as was Alien but not quite as grievously).
The fallout led to the interesting but imperfect Alien: Covenant also from Scott and Alien: Romulus (from director Fede Álvarez).
Each featured leading roles for women, “Alien” in title, and a canon conflict with Predator films.

Canon Timeline

We use the “canon” timeline just for clarity.

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["Prom.", 2012, "darkgreen"],
    ["Cov.", 2017, "darkgreen"],
    ["Rom.", 2024, "darkgreen"],
]

plot(films)

Full Timeline

And the full timeline.

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["Prom.", 2012, "darkgreen"],
    ["Cov.", 2017, "darkgreen"],
    ["Rom.", 2024, "darkgreen"],
    ["AVP", 2004, "darkgoldenrod"],
    ["AVPR", 2007, "darkgoldenrod"],
    ["P", 1987, "red"],
    ["P2", 1990, "red"],
    ["P's", 2010, "darkred"],
    ["The", 2018, "darkred"],
    ["Prey", 2022, "darkred"],
]

plot(films)

Classification

Background

Classification is a supervised machine learning technique used to predict labels or categories based on input data. The goal is to assign each data point to a predefined class, such as spam vs. non-spam emails or diseased vs. healthy patients.

For example, a classification model might be trained on dataset of images labeled as either dogs or cats and it can be used to predict the class of new and unseen images as dogs or cats based on their features such as colour, texture or shape.

Supervision

We recall supervision:
- Uses labeled data.
- Maps inputs to outputs.
- Goal: Predict outcomes.
- Examples: Classification, regression.

Types of Classification

Binary Classification (Good vs. Bad)
Multi-class (Alien Quadrilogy vs. Prometheus Trilogy vs. AVP)
Multi-label (Alien vs. Sigourney vs. Ridley)

Process

Stages

Data Collection
Feature Extraction
Model Training
Iterate
1. Model Evaluation
2. Prediction

Data Collection

We start with some training data.
- For example, the films of the Alien franchise.

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["Prom.", 2012, "darkgreen"],
    ["Cov.", 2017, "darkgreen"],
    ["Rom.", 2024, "darkgreen"],
    ["AVP", 2004, "darkgoldenrod"],
    ["AVPR", 2007, "darkgoldenrod"],
    ["P", 1987, "red"],
    ["P2", 1990, "red"],
    ["P's", 2010, "darkred"],
    ["The", 2018, "darkred"],
    ["Prey", 2022, "darkred"],
]

plot(films)

Feature Extraction

We learn some information about the training data.
- For example, whether each film features Sigourney Weaver

Code

films = [
    ["Alien", 1979],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["Prom.", 2012, "black"],
    ["Cov.", 2017, "black"],
    ["Rom.", 2024, "black"],
    ["AVP", 2004, "black"],
    ["AVPR", 2007, "black"],
    ["P", 1987, "black"],
    ["P2", 1990, "black"],
    ["P's", 2010, "black"],
    ["The", 2018, "black"],
    ["Prey", 2022, "black"],
]

plot(films)

Aside: Features

In machine learning and pattern recognition, a feature is an individual measurable property or characteristic of a data set.

Can be: Released before 2000.
Can’t be: Is art.

Model Training

Pretend you are a model.
Here are the films I consider “good” in red.
How can we “predict” if a film is good.

Code

films = [
    ["Alien", 1979,"red"],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["Prom.", 2012,"red"],
    ["Cov.", 2017],
    ["Rom.", 2024,"red"],
    ["AVP", 2004],
    ["AVPR", 2007],
    ["P", 1987],
    ["P2", 1990],
    ["P's", 2010],
    ["The", 2018],
    ["Prey", 2022,"red"],
]

plot(films)

Model Evaluation

Imagine we use “Directed by Ridley Scott” as our model for a good film.
Hits in red, mispredicts in blue, missed predicts in purple, correct omits in green.

Code

films = [
    ["Alien", 1979,"red"],
    ["A's", 1986],
    ["A3", 1992],
    ["Res.", 1997],
    ["Prom.", 2012,"red"],
    ["Cov.", 2017,"blue"],
    ["Rom.", 2024,"purple"],
    ["AVP", 2004],
    ["AVPR", 2007],
    ["P", 1987],
    ["P2", 1990],
    ["P's", 2010],
    ["The", 2018],
    ["Prey", 2022,"purple"],
]

plot(films)

Prediction

This was accurate until 2017!
- It wouldn’t have been a very good predicting model!

Iteration

In 2017
- Studios pushed Ridley Scott to release a middling Alien: Covenant
- Out of nowhere, Prey was an incredible film in the Predator franchise.
  - And the sole Predator film not in cannon conflict with Prometheus
- Alien: Romulus was a superb Alien film, and the first in the franchise not directed by Ridley Scott

Iteration

However!
- Alien: Covenant was Ridley Scott’s first direct sequel.
- Prey is the sole Predator film to feature a woman in a leading role.
- Alien: Romulus was the second standalone preequel - after Prometheus.
It comes down to feature selection!

Film	Woman-led	Canon	Not a direct sequel	“Good”
Alien	✓	✓	✓	✓
Aliens	✓	✓
Predator			✓
AVP			✓
Prometheus	✓	✓	✓	✓
Covenant	✓	✓
Prey	✓	✓	✓	✓
Romulus	✓	✓	✓	✓

Simple Example

It’s not always complicated.
- How to capture the Alien original quadrilogy?
- Stars Sigourney Weaver.
Easy!

Film	Weaver	Alien in name	“Original”
Alien	✓	✓	✓
Alien 2¹		✓
Aliens	✓	✓	✓
Predator
AVP		✓
Romulus		✓
Avatar²	✓

Classification

Classification

Why classification?

Carolus Linnæus

High School Biology

Two Species

Life is complicated

Example

We Begin…

Classification

Timeline

An Aside

Special Order 937

Special Order 937

Commands

Think, Pair, Share

Up Next

Classification

Timeline

More

Timeline

The Big Mistake

More

Timeline

Reverse Reverse

Classification

Timeline

Modern Era

Timeline

Return of the King 👑

David

Inflection Point

Frankenstein

Think, Pair, Share

Classification

A Wrinkle in Time

Timeline

Canon Timeline

More

Canon Timeline

Full Timeline

Classification

Background

Supervision

Types of Classification

Process

Stages

Data Collection

Feature Extraction

Aside: Features

Model Training

Model Evaluation

Prediction

Iteration

Iteration

Simple Example