Practical Deep Learning: Learning Game Plan

I dug this video up from the older version of Practical Deep Learning for Coders. It’s a “learn how to learn” type video.

My take aways:

• Practical work is the goal. Don’t fool yourself into thinking you know what is taught by just watching the videos or reading the book. Watch the video, watch it again but with the tools at hand, pausing to try stuff out, do a project using what you learned. The title of this blog is me trying to remind myself of this concept.
• I have my own ML workstation setup with a GPU. I used poetry to set it up. This was a pain due to lack of attention to repeatability in Jupyter Notebooks. It would be very useful if they had a “lock” feature that records all python packages in the environment they are run it and their exact version. Below is what it looks like after finishing Lesson 1 (the next video).
• Be tenacious. Finish a project.
• One message that’s very close to my heart is not keep on getting ready to do a project, like stopping to learn linear algebra (and then remembering how I never learned all the math terms, and therefore want to go back even deeper) in order to do well on this course. Try to do a complete project, then on the next project go deeper, dig in deeper when the code needs it.
• Show your work to the world. Blog not to be a breaking news source, but blog for the audience of yourself 6 months ago.

pyproject.toml

[tool.poetry]
name = "dl"
version = "0.1.0"
description = ""
authors = []
readme = "README.md"

[tool.poetry.dependencies]
python = "^3.10"
fastai = "2.7.11"
duckduckgo-search = "^2.8.5"
ipykernel = "^6.22.0"
ipywidgets = "^8.0.5"
jupyterlab = "^3.6.3"
jupyterlab-git = "^0.41.0"

[build-system]
requires = ["poetry-core"]
build-backend = "poetry.core.masonry.api"

Do Your Own Deep Neural Network in Rust idea

Project idea: Take DNNs from Scratch in Zig and build it in Rust.

LLMs and the Truth™

The Economist covers It doesn’t take much to make machine-learning algorithms go awry. Will we see a core of knowledge built that considered “The Truth”, and then all other input data is evaluated on how likely that’s true based on the givens? LLMs judging what is fed to their younger siblings?

Back to Machine Learning

In 2012 I spent a lot of my rare free time working thru both Sebastian Thrun and Peter Norvig’s Intro to Artificial Intelligence (which I can’t find anymore) and Andrew Ng’s Machine Learning course. Andrew’s was the better of the two.

One thing that blew me away was how he used k-means clustering on our homework submissions to discover where there were a large number of students that had a common misconception, and make a clarification video.

At the time I was working in R&D at Rosetta Stone, and we wanted to bring data science and machine learning to bear on how to improve our language learning offerings. It was a linear course, and we dreamed of building a model of what our learners knew, and then challenging them on what they didn’t know. Duolingo had a much better vision and execution for this.

With the breakthroughs in the last few years: Stable Diffusion, LLaMA, LLaVA (yes, I’m ignoring things I can’t run on my own computer), I wanted to dive back in and learn more. To that end I’m taking Jeremy Howard’s Practical Deep Learning for Coders.

I have a lot of data on comic books. I’m hoping to build some practical applications for this. One idea is to feed each panel into something like LLaVA and have it describe what’s going on in it, then have it summarize the story.

Hands On Rust

I’ve been working thru Hands On Rust. You learn rust and make a game. Play my game. See it on github.