Exploring AI and ML

Learn, Build, Deploy

Presented by Evelyn J. Boettcher

2023-12-31

Disclaimer



The views expressed in this talk are those of the author
and do not necessarily reflect the official policy or position of the

  • United States Government,
  • Department of Defense,
  • United States Air Force,
  • Air Force Research Lab, or
  • Air University.

If you want to learn more about Machine Learning and AI in Dayton, please attend a GemCity Machine Learning meetup.

GemCity ML/AI

We meet every third Thursday at 6pm as part of GemCity Tech meetup group.


GemCity.TECH

GemCity TECH’s mission is to grow the local industry and the community by providing a centralized destination for technical training, workshops and providing a forum for collaborating.


Currently, support several special interest groups from a variety of technical disciplines.

Discord

MeetUp

Outline

This presentation will cover the following topics:

  • Introduction to AI and machine learning (ML)
  • Supervised and unsupervised learning
  • Transfer Learning
  • Classification and classes
  • Dataset creation: We will be building a ML model using Teachable Machines

Build a Machine Learning model


The learning outcomes are as follows:

  • Understand the difference between AI and machine learning.
  • Explain supervised and unsupervised learning.
  • Classifier and classes
  • Importance of clean datasets and how bad data affects ML performance
  • Your own ML algorithm that you trained without any coding.

Artificial Intelligence (AI)

AI is more than Large Language Models (LLM)

LLM: Large Language Models: aka Chat GPT, Claude AI, Bard, Gemini

  • Type of AI program that can recognize and generate text.
  • LLMs are trained on huge sets of data — hence LARGE, Costly to build
  • LLMs are built on machine learning: specifically, a type of neural network called a transformer model.

Transformers:

  • Uses Attention models: (greatly improves speed)
  • Transformers learn the context and track relationships between sequence components

GPT: Generative Pre-trained Transformers

  • GPT is neural network-based language prediction models built on the Transformer architecture.
  • GPT models analyzes natural language queries (prompts), and predict the best response based on their understanding of language.
  • GPT models rely on the knowledge gained after training with hundreds of billions of parameters on massive language datasets.

What is AI and ML?

Artificial Intelligence (AI)

  • A field in computer science
  • AI is something that has the ability “learn” to do something without instructions.

Machine Learning (ML)

  • Machine learning is a field of study in artificial intelligence
  • ML uses statistical algorithms that can learn from data and generalize to unseen data (Testing/User data)
  • Perform tasks without explicit instructions.

Velev and Zlateva (2023)

How do ML algorithms learn

ML Algorithms learn via supervised or unsupervised learning.



Supervised Learning

Supervised learning is when you have labels for each data point.

Unsupervised Learning

Unsupervised learning is when it “learns” something about the data, but do not have the data truth.

Example of Data for Supervised Learning

Data Truth (Label) Data Truth (Label)
cat dog


x 100’s more Labeled images

Example of data for unsupervised learning

Data Data


x 500’s more images

Goal of unsupervised learning

Find features that separate the data into groups / clusters. Then hope there is a small number of labeled features so that we can classify those groups.


Why Do Unsupervised

Labeled data is really expensive. Sometimes “truth” can not be obtained.


Example: atmospheric distortion correction

Gedunken Experiment

Say there is a town with two employers (A and B).
Predict if person works for company A or B, based on where they live.


First stab: See if people are clustered around their work place.

So if we had a clustering algorithm we can predict where a person might work.

K-Means

K-means is the most common clustering algorithm.
K-means clusters, n data points (e.g. All your data) into k clusters by placing each data point to the nearest k.


Example: point p is 5 ft from \(k_i\) and 10 feet from \(k_j\). K-means would then place point p into the \(k_i\) group. Once, all the points are placed into a cluster or group, the new means for each cluster (\(k_i\) and \(k_j\)) are updated. The location for each mean cluster k is calculated by finding the mean from all the points that belong to that cluster.


Transfer Learning

Transfer Learning can allow you to use what is learned from a large dataset to your smaller (refined) data similar to the larger model.

Sarkar (2018)

  • Example: ImageNet Dataset has 1.4 million images and 1000 classes.

To learn more go to TensorFlow Transfer Learning Tutorial

Classes, Classification and Models

We are going to build a model that classifies between two (or more) classes.

Classes

The label or category:

  • Cat, Dog


Classification

A category into which something is put.

Model

Tries to predict the correct label (class) of a given input data.
The model is trained using the training data, and then it is evaluated on test data.

Classification

Say you are tasked to group monkeys into two classes:

  • Class_1: Biting Monkey
  • Class_2: Non-Biting Monkeys


Developing a Model

  • Create a rule to define: Biting and NonBiting
  • Separate your monkeys by that rule
  • Repeat until all of your monkeys can follow your rule.

Image from Huffington Post.

Class_1: Biting Class_2: Non Biting
card c card d
card a card b
… …

Classification Model: Evaluation

  • What Rule did you use to determine which class the Monkey belongs to?

  • Does your model hold up?

Visual
  • Class_1= Biting Monkey
  • Class_2= Non-Biting Monkeys

Rule

Classification: Man vs Machine

In this monkey example, we (human) could infer the model (set of rules). Software Engineering is when a human makes a rule that the machine follows.


Most Machine learning classifiers, the “rules” of the model are unknown to us. We only get the output of the machine assigned classes.


Machine learning is when the machine creates the rules for the machine to follow.

Building a ML Model

There are four basic steps to building a model

  • Train the model
  • Export the model
  • Use the model
  • Forth Step: Repeat


Why a forth step

Your model, app etc will not work the first time.

computer_fire

Training a model to classify sounds

We are going to train a ML model to classify sounds.
WITHOUT any coding.

  • Google “Teachable Machine”
    • Click Get Started
    • Click Audio Project (standard model)
  • Create three classes:
    • Background
    • Clapping
    • Snapping

Teachable Machine

Building ML with Teachable Machines

Thank you


If you want to take this to the classroom.
I have a No Code ML Tutorial for Elementary School Children



  • Train a model to classify hand gestures
    • Learn what a classifier is
  • Use conditional statements to turn those classes into an emoji
  • Build a Web application

github

References

Gem City Tech ML/AI

Sarkar, Dipanjan (DJ). 2018. “A Comprehensive Hands-on Guide to Transfer Learning with Real-World Applications in Deep Learning.” Medium. November 2018. https://towardsdatascience.com/a-comprehensive-hands-on-guide-to-transfer-learning-with-real-world-applications-in-deep-learning-212bf3b2f27a.
Velev, Dimiter, and Plamena Zlateva. 2023. “Issues of Artificial Intelligence Application in Digital Marketing.” In. https://doi.org/10.3233/FAIA230716.