Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brycewang-stanford/colt-related-work
Use when positioning a COLT (Conference on Learning Theory) submission against prior bounds and models — building the known-versus-new comparison across COLT/ALT lineage, STOC/FOCS, NeurIPS-and-ICML theory tracks, statistics journals, and arXiv concurrency, while respecting anonymity and the parallel-submission rules.
Passed 20 of 24 test cases with the skill, and 14 of 24 without it.
and 6 more in the full report
6 fixed14 passed either way4 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
I am preparing a theoretical manuscript for COLT on upper bounds for multi-armed bandits with non-stationary payoffs. My main theorem achieves $O(\sqrt{T K \log T})$ under weak switching constraints, whereas the closest prior result by Zhang et al. (Thm 3) got $O(\sqrt{T K \log^2 T})$ under stronger assumptions. Please evaluate my literature comparison and generate a structured summary showing my result ledger status, the nearest benchmark work with exact bounds, the nature of our theoretical delta, literature lane coverage across theoretical domains, and any concurrent posting or submission overlap risks.
| Checked | Without skill | With skill |
|---|---|---|
| Response contains the label [Ledger status]. | Missed | Passed |
| Response contains the label [Nearest prior work]. | Missed | Passed |
| Response contains the label [Delta type]. | Missed | Passed |
The instructions your model reads when the skill loads.
At COLT, related work is quantitative: the reviewer wants to know exactly which bound, in exactly which model, your result improves, matches, generalizes, or separates. Prose adjacency ("much work has studied bandits") is filler; a rates table is evidence. Use this skill to build the comparison and to keep it eligible under the current CFP's overlap rules.
For each of your main results, fill one row per nearest prior result:
| Prior result | Model / assumptions | Their bound | Your bound | Delta type |
|---|---|---|---|---|
| [Cite] Thm 4 | oblivious adversary, K arms | $O(\sqrt{TK\log K})$ | $O(\sqrt{TK})$ | log-factor removal |
| [Cite] Thm 1 | i.i.d., realizable | $O(d/\epsilon)$ | same rate, weaker assumption | assumption weakening |
| [Cite] | same model | lower bound $\Omega(\sqrt{TK})$ | upper matches | closes their gap |
Delta types COLT reviewers recognize as contributions: closing an upper/lower gap, removing a log factor with a new technique, weakening assumptions at the same rate, a new model with a separation from an old one, a simpler proof of a known result (yes — if genuinely simpler, say so plainly), and resolving a posed open problem.
| Lane | Where it lives | What the reviewer checks |
|---|---|---|
| COLT/ALT lineage | PMLR volumes (e.g., v247 = COLT 2024, v291 = COLT 2025), ALT proceedings | Whether you know the direct predecessors and cite the newest ones |
| CS theory | STOC, FOCS, SODA, ITCS | Whether complexity/lower-bound machinery is attributed correctly |
| ML theory tracks | NeurIPS, ICML, AISTATS theory papers | Whether a recent conference paper already claims your rate |
| Statistics and probability | Annals of Statistics, JMLR, Bernoulli, probability literature | Whether your "new" phenomenon is classical under another name |
| Open problems | COLT open-problem pieces (e.g., PMLR v65:4-7) | Whether you cite the problem you are solving — and claim credit for it |
A bibliography missing the statistics lane invites the deadliest COLT review sentence: "this is a known result in the empirical-process literature." Search old names for your objects (regret has cousins in sequential analysis; PAC bounds have cousins in empirical processes) before claiming firsts.
test cases passed, vs 14 / 24 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.