Research dossier
Structured inductive biases for neural networks
A research dossier on mechanisms — biological, control-theoretic, optimization, systems — that earn their place by beating a fair, parameter-matched, unstructured baseline.
Maps
The citation map of every filed source — nodes are papers, links are shared references and direct citations. Sign in to build maps of your own.
Source index
All filed sources, grouped by theme, each with the one line on why it matters and a link to the original.
How it works
- 1
Set categoriesplanned
Group a map into the areas you actually care about — one per question you are chasing. A category is a label first; give it search topics only when you want it searched.
- 2
Deep search, per categoryplanned
Each category runs its own literature sweep on the cadence you choose — daily, weekly, monthly. Everything it finds lands in the review queue; nothing enters a map unreviewed.
- 3
Map it, ask it
Accepted sources join the map — nodes are papers, links are shared references. The chat beside it moves the map as you talk, opens the node under discussion, and cites what it used.
- 4
Pattern search
Standing questions asked of the map itself: the same paper filed twice under a preprint and a DOI, a source stranded with nothing to connect it to, the paper bridging two of your categories, a queue nobody cleared.
Steps marked planned are not built yet. Today sources are added by hand and pass through the same review gate the scheduled search will use.
What people use it for
Track a field without reading it daily
Set the cadence once. The queue collects what is new; you spend your attention on accept or reject, not on searching.
Check an idea against prior art
Before building a mechanism, ask what already does it. Typed relations — grounds, baseline, competes, prior-art, refutes — say how each source bears on the idea.
Keep a claim honest
Every source carries its evidence label and its caveats, so a map records how strongly a thing is known, not only that it was found.