Site-hosted paper project pages first: the core finding, the figure to inspect, and where the paper, code, and artifact trail live. Older Medium posts are listed below so this site stays the single map to public writing.
paper project pages
-
A transferred collapse-warning signal fires on every tested LoRA fine-tuning run for masked diffusion language models, even when nothing is collapsing.
-
Two cheap attention-based diagnostics predict grokking phase regimes across 1,120 transformer runs, with architecture-specific critical weight decay.
-
Replacing LayerNorm with Dynamic Tanh helps or hurts depending on the token-to-parameter regime, tested from 64M to 3.78B parameters.
older technical writing
These remain external Medium posts, but they are now linked from the matching project areas instead of floating outside the site.
-
Older PPG blood-pressure estimation walkthrough, now tied to the biomedical signal ML project line.
-
Older NLP learning note; kept as part of the writing archive rather than promoted as a current research claim.
-
Layout-aware document classification walkthrough connected to the document understanding project line.
-
Object-detection walkthrough for driver-license field localization, connected to the document understanding project line.