Week 1: Synaptic Circuitry, Deeeeep Learning, and Gradiently Dissenting into Machine Learning’s Impact in Healthcare: (2/14/2024)
Andrew Y -
Hello everyone!
First, thank you so much for following this blog! Let’s dive deeper (metaphorically and literally!) into what I’ve been up to the past week.
Last week, we looked into the trio that drives modern healthcare: Medical Imaging, Computer Science, and Computational Biology & Bioinformatics. While Medical Imaging and Computational Biology lead the charge in patient care, it’s Computer Science and Artificial Intelligence that turbocharge disease detection, making modern Medical Diagnosis swift and effective.
To understand Computer Science, we first need to grasp the concept of a neural network. Imagine it as a bunch of marshmallows mushed together by toothpicks. Now imagine if the marshmallows were connected by toothpicks that had cables within them like a network of interconnected nodes (or neurons) to work together and process information, recognize patterns, and make decisions.
One key concept within neural networks is gradient descent. Think about it like navigating a terrain to find the lowest point in a valley, where the valley represents the error or loss in the model’s predictions. By adjusting the parameters of the neural network in the direction that reduces this error, gradient descent helps the network learn and improve its performance.
Now, let’s zoom into deep learning, a subset of machine learning where neural networks with multiple layers (hence “deep”) are employed. These layers enable the network to learn increasingly abstract and complex features from the input data, allowing for more sophisticated tasks such as image recognition, natural language processing, and medical diagnosis.
In medical imaging, this technology changes how we diagnose patients compared to traditional methods. Take, for instance, the detection of tumors in medical scans. Neural networks can be trained on vast datasets of labeled images to recognize subtle patterns indicative of various conditions. Once trained, these networks can analyze new images rapidly and accurately, assisting radiologists in making diagnoses more efficiently and reliably.
For stable diffusion, neural networks and deep learning play a more complicated role with three different architectures, or designs. There is the variational autoencoder, which takes an image, compresses it down for the model to train it, and then decompresses the generated image; the text encoder, which takes a user input based on a prompt and turns it into something that the autoencoder can understand. Finally, there is the U-Net, which can be thought of as the “brain” or “boss” of the image-generating process (aka. diffusion). Stanford-University Human-Centered Artificial Intelligence gives an in-depth explanation of why this works, so be sure to check it out: https://hai.stanford.edu/news/could-stable-diffusion-solve-gap-medical-imaging-data.
For this week, I’ve uploaded some links by famous mathematician 3Blue1Brown and other famous novels, which give a more in-depth analysis of the fundamentals of Machine Learning, Deep Learning, and AI and how they are being used in healthcare.
Also, I’ve started weekly articles and activities! Hopefully, you learn something from those.
Happy Valentine’s Day! Stay tuned for more updates!
Warmest Regards,
Andrew Yang
RESOURCES:
Interactives:
This gives a simple simulation of how neural networks work:
A Neural Network Playground
Youtube Links:
3Blue1Brown:
Youtube Channel: https://www.youtube.com/@3blue1brown
Neural network Playlist: 3Blue1Brown goes into depth about the logistics and science behind Neural Networks, and why Machine Learning works the way it does. https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi
Weekly Articles –
The Little Book of Deep Learning by Fleuret (2023): https://fleuret.org/public/lbdl.pdf
ScientificAmerican.com: AI Special Report: The Rise of AI (June 2016): https://drive.google.com/file/d/1S_7sVM6GdF57T8c5n-0w7Nys64Xbclq0/view?usp=sharing
TIME: The A to Z of Artificial Intelligence by Billy Perrigo (April 2023):
https://drive.google.com/file/d/1iVLD6mi9lZ8IC_-1HO9aIJRqof4Lz-EO/view?usp=sharing
Weekly Updates Posted Here:
Progress Site: Google Folder: https://drive.google.com/drive/folders/1PJROf3cJFZIWSS3Im-ZTZ6LLiTnyMXeW?usp=sharing
This folder gives access to all the images taken during model training, report analysis, progress checks per week, results, and other information that will be useful in the next ten weeks.
Comments:
All viewpoints are welcome but profane, threatening, disrespectful, or harassing comments will not be tolerated and are subject to moderation up to, and including, full deletion.