A small research lab inside a small company.

TsugiLabs is the research division of TsugiCinema Inc., a Delaware C-Corp. The parent company builds hardware and distribution for cinema-grade home video. The lab works one layer down from that: continual learning, model adaptation, signal-quantized optimization, and on-device inference under the kind of constraints that come from shipping silicon.

The kintsugi metaphor that names the parent company applies here too. A model is a fragile thing. It breaks under distribution shift. It forgets what it has already learned. The traditional repair is invisible: train more, average more, hide the seam. We are interested in the opposite. Make the seam visible. Make the fix legible. Treat the place where the model was bent back into shape as the most informative part of the artifact.

What we publish

Two streams. The blog is for short notes and write-ups that are useful but not paper-shaped. The papers page is for extended abstracts, workshop submissions, and preprints. Both have RSS feeds. Neither has a comment section.

What we do not publish

Pre-disclosure mechanism details for technology under provisional patent prosecution. The build pipeline runs a fail-closed disclosure lint on every commit so we cannot accidentally publish material that is still under embargo. If a piece of work belongs in a patent application before it belongs in a paper, that is the order it ships in.

Principal

Tong Liu. CEO of TsugiCinema Inc., does the hardware architecture, patents, and a substantial chunk of the engineering. Other contributors will appear here as their roles formalize.

Contact

research@tsugilabs.com for research correspondence. partnerships@tsugicinema.com for biz-dev and licensing questions.