ProbXiv
Explorersign in
Problem archiveProblem record

Statement

For a binary rr-hot codeword cc of length nn, view 1rc\frac{1}{r}c as a distribution on Z/nZ\mathbb{Z}/n\mathbb{Z} and call it the reduced distribution of cc, denoted by dist⁡(c)\operatorname{dist}(c). For a code I⟶CI\longrightarrow C with codewords cic_i and probabilities pip_i, define its average minimal KL-divergence (AMKL) by

∑imin⁡jKL⁡(dist⁡(ci)∣∣dist⁡(cj))pi.\sum_i \min_j \operatorname{KL}(\operatorname{dist}(c_i)\mathbin{||}\operatorname{dist}(c_j))p_i.

AMKL conjecture. A feature code with bigger AMKL has better performance.

The conjecture proposes AMKL as a metric for feature coding, because the Hamming distance can be the same for code pairs whose observed feature-coding performance differs. Its status is not established by the supplied text.

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.