Dear Professor Maiti,

I recently read your VECTOR+ paper and found it quite interesting.

I’ve been working on a project where I use pre-1950 materials data to try and generate relevant compounds that were discovered much later. I use different generation methods and rank candidates based on novelty and feasibility.

I found some similarities with VECTOR+, particularly in exploring chemical space while balancing property optimization and novelty. I was curious whether the activity guided embeddings could limit the generation of genuinely novel structures, and how you approach this trade off.

Attatching a short summary of my understanding of the vector+ paper and what ive been working on.

Vector+ summary - https://drive.google.com/file/d/1fdkQpz2eD_zN_4cd2ie7FQOMPs94ZBnl/view?usp=sharing

My project - https://drive.google.com/file/d/18QVPd_zITamtg6ky60rujaLiRQ2KuEWw/view?usp=sharing

I’d love to learn more about your work. Would it be possible to visit your lab sometime?

Regards,  

Pranjal

26B2718