Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/colt-topic-selection
Use when deciding whether a result is COLT-shaped (Conference on Learning Theory) — a theorem-first learning-theory contribution — versus better routed to ALT, STOC/FOCS, NeurIPS/ICML/AISTATS, JMLR, or a statistics journal, and whether the right vehicle is a full paper or a COLT open-problem piece.
Passed 16 of 23 test cases with the skill, and 13 of 23 without it.
and 4 more in the full report
7 fixed9 passed either way3 still failing4 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
We proved a new O(T^{1/3}) regret bound for non-stationary multi-armed bandits under dynamic regret constraints, improving on the prior O(T^{1/2}) rate. We have zero experiments and pure theory. Please evaluate whether this project belongs at COLT and output your evaluation using the bracketed output format including fit classification, headline claim, delta type, alternative vehicle, and pre-writing blocker.
| Checked | Without skill | With skill |
|---|---|---|
| Output classifies fit as [Fit] core COLT. | Missed | Passed |
| Output identifies delta type as rate improvement over prior bound. | Passed | Passed |
The instructions your model reads when the skill loads.
Use this before writing begins. COLT solicits papers on theoretical aspects of machine learning, described in the 2026 CFP (checked 2026-07-08) as a subject at the intersection of computer science, statistics, and applied mathematics, with an explicitly inclusive view that includes theory shedding light on empirical phenomena. The practical bar: the contribution must be a theorem — a rate, a separation, a characterization, a hardness result, or an algorithm whose guarantee is the point.
| Signal in the project | Best venue reading |
|---|---|
| Regret/sample-complexity/oracle-complexity bound, new or improved rate | Core COLT |
| Matching lower bound via a new instance construction | Core COLT |
| Theory explaining a deep-learning phenomenon, theorem-first | COLT (in-scope by the CFP's inclusive view) or ML-conference theory track |
| Learning theory with a long, self-contained development (60+ pages of ideas, not just proofs) | JMLR or Annals of Statistics — journal-length exposition |
| Algorithmic result where combinatorial/complexity machinery dominates the learning content | STOC / FOCS / SODA |
| Learning theory, but the community fit is the smaller algorithmic-learning-theory circuit | ALT (sister venue, autumn deadline cycle — verify) |
| Theorems plus substantial experiments as co-equal evidence | AISTATS or NeurIPS/ICML |
| Statistical methodology with inference guarantees and applied audience | AISTATS or a statistics journal |
| Probabilistic/Bayesian modeling contribution, uncertainty-first | UAI |
| A precise, motivated question you cannot answer | COLT Open Problem piece (see below) |
COLT has a tradition of publishing short open-problem pieces in its proceedings — citable, reviewed, and historically influential (verified instance: Agarwal, Krishnamurthy, Langford, Luo & Schapire, "Open Problem: First-Order Regret Bounds for
test cases passed, vs 13 / 23 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.