LLM Based Project Ideas
Large Language Models (LLMs) have taken the AI world, and everyone else it seems, by storm. Below are some project ideas, probably mostly at the MS level. Note that I am not an LLM expert nor am I even an AI expert; these projects are focused on applying some AI techniques, not on developing new AI/LLM techniques. If you undertook one of these projects it would be an opportunity for both of us to learn something about LLMs and their usefulness in particular application/software domains.
Anomalous Image Detection and Description
Can you use some LL-based analysis pipeline to detect images that have an anomaly in them? And perhaps describe the anomaly? Perhaps building in the middle of a forest, or a cell tower out in the desert? Or perhaps anomalies among images of data graphs (say, EKGs)?
This project would require learning about LLMs, tool research and experimentation, and then building a custom tool pipeline to process the specific type of image data that you will work with. It may involve doing some portion of augmentation (described below).
Research paper on zero-shot classification
Augmenting a Generic Model with Custom Data
How to customize an LLM with your own data? The first L is Large, meaning you don’t really want to create/train your own, and it is generic, meaning it contains many important relationships that may apply to any domain, including your own. But you want to customize it to get answers related to your own dataset.
This may be pursued in conjunction with the image ideas above.
Small Language Models for Specific Datasets?
Would building a small language model (SLM) be a good idea for some of our application data, or for some other domain? Let’s try it!