|
Stability of interpretable circuits across models of varying size
showcase (WIP)
| repo
| brief report
This repo contains my ongoing experiments to better understand the recent advances in mechanistic interpretability research.
|
|
Finetuning open source models under low-resource constraint
NeurIPS LLM Efficiency Challenge, 2023
challenge
| repo
| models
| data
I participated in the LLM efficiency challenge and finetuned performant, open source models using custom open source datasets.
|
|
Applying protein language models to predicting disease causing genetic mutations in medically actionable genes
Onuralp Soylemez, Pablo Cordero
NeurIPS Workshop on Learning Meaningful Representations of Life, 2022
workshop
| paper
| code
We developed a protein language model evaluation framework and revealed unappreciated structural features of proteins that are missed by other structure predictors like AlphaFold.
|
|
Fine-tuning large foundational models in cancer biology
Pablo Cordero, Onuralp Soylemez, Darren Zhu
Bio x ML hackathon, 2022
hackathon
| code
| 5-min summary
We ranked 2nd place ($3,000 prize award) with our project on finetuning large language models for single cell biology on smaller datasets like DepMap cancer dependency.
|
|
Prioritization of drug targets using Bayesian tensor factorization
ICML Workshop on Computational Biology, 2022
workshop
| paper
| code
Drug targets with human genetics evidence are shown to have better odds to succeed. We used Bayesian tensor factorization to integrate different types of human genetics evidence from rare genetic diseases to complex disorders.
|
|
Accelerating scientific discovery process using large language models
interview
| code
Proof-of-concept "chatGPT for genetics".
|