I earned my ML Engineer certificate!
That’s my main major update but I also wanted to share my interest in a few project ideas. Most I’ve already started or have the groundwork for.
If anyone is interested in joining, I’d be more than happy to chat and give a proper overview of the projects! Just let me know.
Note: I enjoy experimental projects the most so expect iterative processes, lots of experimentation and hopefully, blurring lines between possible and improbable.
With that being said, here are my project ideas:
A Neural Network
trained to look at a piece of art (specifically: the colors, the sentiment assigned to the colors, the angles of lines, identify the amount of subjects, the ratio of curved to angled lines, the ratio of cool to warm hues, and some more features) and tell you within reasonable accuracy (to be defined) what the artist was trying to make you feel.
I created a color dataset generator to make this project a little bit easier for me, but I can create more programs as need be. This project would be purely experimental but it’d be really cool to see the end result!
2: Unique Metric
This one is my personal favorite passion project. I already have the main function groundwork laid to return a vector of features that averages the model’s responses. Here’s a VERY general breakdown:
Two models (they don’t have to be complex, but they have to be distinct) are both provided a theoretical problem (X(t) for my purposes). They both generate potential solutions. The Unique Metric creates a weighted average from their responses and influences the loss of the output, giving the models a theoretical Y to strive for. The goal, and reasoning, is to influence the models to work together to generate a more balanced approach and an “optimal” response that’s closer to the ACTUAL Y.
It’s hard to explain this project without diving into all the documentation and diagrams I’ve crafted for it, but for my purposes, I am going to use an Emotional (sentiment) trained model and a Logical model. One generates the most efficient, cold-hearted, to-the-point solution while the other generates one that is more based in sentiment, ethical considerations, probably too complicated of a possible answer. The Unique Metric will average the outputs (the values in the feature vectors) and provide a “goal” for the models to strive for. The hope is that the models, upon being penalized for falling outside of the accepted range the Unique Metric sets, will either work together to produce ONE well-balanced output or they will both produce outputs that are individually identical to one another, as a means of minimizing cost.
I hope this brief explanation makes sense. If someone is interested in working on the project with me, I promise I have more, better-outlined details than this.
Anyway, those are the two projects I’m working on! I’m excited to be certified and I hope to continue growing with this community. Thanks for having me.