Notes on ML research, internships, and things I've learned along the way.
Why unimodal fits stretch across empty valleys, the exact log 2 threshold where mode-seeking kicks in, and how Forward vs. Reverse KL shapes diffusion, distillation, and RLHF.
The programs, deadlines, and faculty I shortlisted for efficient ML and vision PhDs, with a copyable Google Sheet template.
The SOP I submitted to UT Austin, with section-by-section lessons for research applicants across universities.
How recent visual generative models route tokens through mixture-of-experts — from batch-level context to semantic prototypes.
Common questions about the Google DeepMind India Pre-Doctoral Researcher program, from someone who went through it.
Curated papers and resources on assessing aesthetic quality in graphic design.
When PyTorch tensor ops run faster on CPU vs GPU — and why the answer isn't always obvious.
Reflections from a research internship at Adobe Research.
Notes on multimodal image-to-image translation with BicycleGAN.