machine_learning.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## What is Machine Learning\n",
    "\n",
    "<iframe src=\"https://giphy.com/embed/5bgLjE2iQypE1uYIeA\" width=\"480\" height=\"271\" style=\"\" frameBorder=\"0\" class=\"giphy-embed\" allowFullScreen></iframe><p><a href=\"https://giphy.com/gifs/mit-robots-jenga-machine-learning-5bgLjE2iQypE1uYIeA\">via GIPHY</a></p>\n",
    "\n",
    "https://news.mit.edu/2019/robot-jenga-0130\n",
    "\n",
    "Here are a couple of definition:\n",
    "\n",
    "Google:\n",
    "\n",
    "the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw inferences from patterns in data.\n",
    "\n",
    "Python Data Science Handbook:\n",
    "\n",
    "You can think of machine learning as a class of algorithms that allow a program to detect particular patterns in a dataset, and thus \"learn\" from the data to draw inferences from it\n",
    "\n",
    "My definition: \n",
    "\n",
    "Machine Learning is the iterative process of improving a function estimator. Ex: `y = f(x)`. We usually have the `y` - or want to find one that is interesting to us , we have the `x`, we need to find a model of `f()` that gives us the best results, sometimes with new `x`s. One way we do that is through algorithms that can improve themselves - which boils down to statistics and calculus.\n",
    "\n",
    "\n",
    "## How is it different than Artificial Intelligence (AI)?\n",
    "\n",
    "At the time of writing this, AI is taking th news, economy, and tech by storm. Every company seems to want to have \"AI\" attached to their processes. Quite honestly, the distinction between AI and ML is quite blurry and you may get many different answers when you ask a practitioner to compare.\n",
    "\n",
    "In general Machine Learning are known and well understood algorithms that help us predict an output based on a input, notably the creation of a function estimation. Artificial Intelligence is the highest level process, putting data in a process that can include data cleaning, data augmentation, passing the data through one the function estimator created, (possibly using that data to create a better function estimator), then transforming the result into something useful for the user.\n",
    "\n",
    "Aka, ML is a way to create the engine. AI uses the engine (with all the fun gizmos of a car).\n",
    "\n",
    "Here is a definition form Columbia Engineering\n",
    "\n",
    "    [A]rtificial intelligence refers to the general ability of computers to emulate human thought and perform tasks in real-world environments, while machine learning refers to the technologies and algorithms that enable systems to identify patterns, make decisions, and improve themselves through experience and data. \n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## A Brief History\n",
    "\n",
    "- Bayes Theorem ~ 1763\n",
    "- Perceptron  - 1958\n",
    "- Neural Network Research Takes off - 1960\n",
    "- AI Winter in 1970s\n",
    "- Convolution -  In theory - 1959, spatial 1989\n",
    "- AI Picks up due to internet age -1990s (more data is better)\n",
    "- Boosting - 1990\n",
    "- Speech Recognition - 1997\n",
    "- Facial Recognition - 2006\n",
    "- Alexanet Deep Learning - 2012\n",
    "- Resnet - 2016\n",
    "- Transforms \"Attention is all you need\"- 2017\n",
    "\n",
    "\n",
    "Citations: \n",
    "<details>\n",
    "\n",
    "https://www.dataversity.net/a-brief-history-of-machine-learning/\n",
    "\n",
    "https://en.wikipedia.org/wiki/Bayes%27_theorem\n",
    "\n",
    "https://arxiv.org/abs/1511.08458\n",
    "\n",
    "https://yann.lecun.com/exdb/publis/pdf/lecun-89.pdf\n",
    "\n",
    "https://papers.nips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html\n",
    "\n",
    "https://arxiv.org/abs/1512.03385\n",
    "</details>\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Modern History\n",
    "\n",
    "Currently large language models (LLMs) are very popular. With the release of Chat GPT3 from OPenAI, AI became very accessible to the everyday person. Its extreme popularity is creating an current AI boom which we are in today.\n",
    "\n",
    "Some of the Main Players are:\n",
    "- OpenAI - Chat GPT\n",
    "- Meta - Llama\n",
    "- Google - Gemini\n",
    "- Microsoft - Co Pilot\n",
    "- Honorable Mention -> [DeepMind with Alpha Go](https://www.youtube.com/watch?v=WXuK6gekU1Y), merged with Google (the modern )\n",
    "- Tesla/Google/More - Self Driving\n",
    "\n",
    "Currently multimodel models are beginning to be more popular, this includes image, video, voice, or any other medium combined together. \n",
    "\n",
    "Many people believe AI will catapult the human race forward in technology - and offer significant profits for companies. \n",
    "\n",
    "AI is trained using special processors - usually GPUs (though there are more). The main company that has a significant market share is NVIDIA, a GPU manufacturing company - which for a brief time was the most valuable company in the world ~ 3 trillion dollars.\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# The Main Types\n",
    "\n",
    "## Supervised Learning\n",
    "\n",
    "Supervised learning involves somehow modeling the relationship between measured features of data and some label associated with the data; once this model is determined, it can be used to apply labels to new, unknown data. This is further subdivided into classification tasks and regression tasks: in classification, the labels are discrete categories, while in regression, the labels are continuous quantities. We will see examples of both types of supervised learning in the following section. (jakevdp)\n",
    "\n",
    "\n",
    "## Unsupervised Learning\n",
    "\n",
    "Unsupervised learning involves modeling the features of a dataset without reference to any label, and is often described as \"letting the dataset speak for itself.\" These models include tasks such as clustering and dimensionality reduction. Clustering algorithms identify distinct groups of data, while dimensionality reduction algorithms search for more succinct representations of the data. We will see examples of both types of unsupervised learning in the following section. (jakevdp)\n",
    "\n",
    "## Reinforcement Learning\n",
    "In the context of reinforcement learning, an autonomous agent must learn to perform a task by trial and error, without any guidance from the human operator. DeepMind demonstrated that a reinforcement learning system based on deep learning is capable of learning to play Atari video games, reaching human-level performance on many tasks (Mnih et al., 2015). Deep learning has\n",
    "also significantly improved the performance of reinforcement learning for robotics (Finn et al., 2015). (Goodfellow, Bengio, Courville)\n",
    "\n",
    "Citations:\n",
    "<details>\n",
    "jakevdp | https://jakevdp.github.io/PythonDataScienceHandbook/05.01-what-is-machine-learning.html\n",
    "Goodfellow, Bengio, Courville | https://drive.google.com/file/d/1t2gPY6O0MV8yduok6EjAl4kIs3jSJp40/view\n",
    "</details>\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cool Examples\n",
    "\n",
    "- LLMs - ChatGPT4, Gemini, Llama, and more\n",
    "- Alpha Go - https://www.youtube.com/watch?v=WXuK6gekU1Y\n",
    "- Intel's OpenVino Project - Image Recognition\n",
    "- F1Tenth - https://f1tenth.org/\n",
    "- Tesla Self Driving / Waymo - https://www.youtube.com/watch?v=JHgr9SgeicM\n",
    "- Image Generation - https://deepai.org/machine-learning-model/text2img\n",
    "- Evolution Simulation - https://www.youtube.com/watch?v=TZfh8hpJIxo&t=19s\n",
    "- Learning how to walk - https://www.youtube.com/watch?v=L_4BPjLBF4E\n",
    "- Playing Tag - https://www.youtube.com/watch?v=hCmrMOzx5VA\n",
    "- Image Super resolution - https://paperswithcode.com/task/image-super-resolution\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Models and concepts we will discuss in class\n",
    "\n",
    "Classical Machine Learning\n",
    "- Linear Regression - Prediction\n",
    "- Logistical Regression - Classification\n",
    "- Support Vector Machines - Classification\n",
    "  - Note there is also Support Vector Regression - we may not go into this (Prediction)\n",
    "- Decision Tree and Random Forest - Classification\n",
    "\n",
    "Non Classical (Neural Networks)\n",
    "- CNNs, RNNs - Prediction or Classification \n",
    "- LLMs - This is a concept lecture about LLMS\n",
    "- GANs - Generation (Deepfakes, Image Generation, and more)\n",
    "- Transfer Learning - Concept of using other people's models\n",
    "- Digital Image Processing - what can we do without neural networks?\n",
    "- Maybe more?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# The Main Concepts\n",
    "\n",
    "There are some main concepts that almost all (not not all) machine learning models share. They are described below:\n",
    "\n",
    "1. Create your function estimation algorithm | Machine Learning Model\n",
    "\n",
    "2. Define the problem in a mathematical way so a computer can have a goal to work towards. | Error and Loss Function\n",
    "\n",
    "![](./images/Loss_graph.PNG)\n",
    "\n",
    "3. The loss can be changed | Parameter Adjustments\n",
    "\n",
    "4. Moving the loss in the correct direction | Stochastic Gradient Descent\n",
    "\n",
    "5. Finding faster ways to do things | Leveraging Math\n",
    "\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": ".venv",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "name": "python",
   "version": "3.11.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}