Ghost---Shadow
So many ideas, so little time.
Built from Ghost---Shadow's public GitHub profile, Looking to relocate. Not on Take My Repo yet, so there are no posts or updates here.
Reproduction of SLiC-HF: Sequence Likelihood Calibration with Human Feedback
This repository contains key takeaways and notes on all the papers I have read and online conferences I have watched, categorized and archived for quick future reference.
A Traffic Optimization system in C++ using a rudimentary ant colony optimization technique.
An academic project for implementing a distributed computing system.
The squiggly line theorem: a squigglier curve is closer to the data, in a direction the starting loss cannot see. Proved in Lean 4 for sines; a conjecture for neural networks.
Causal-consistency post-pre-training objective: a why/therefore cycle loss (untested)
InSQuaD is a research framework for efficient in-context learning that leverages submodular mutual information to optimize the quality-diversity tradeoff in example selection for large language models
A WATCH_DOGS-like smartphone controlled traffic light system, which is designed to give service vehicles a priority.
An implementation of Joel's Evidence Based Scheduling.
An academic project for implementing a server for a distributed computing system.
A fully vectorized PyTorch implementation of BLEU scores optimized for training neural networks.
A comprehensive study of the "Balmer Peak" phenomenon in binary search algorithms, where introducing controlled randomness (epsilon) can improve performance on certain array structures.
Ask ChatGPT how we can use LLMs to augment the business processes of websites by providing document.body.innerText and web url.
A program in c++ used to simplify expressions using Karnaugh Map.