pipette
ESEspañol

Orbit Reduction and Learned Run Distributions for Finite-Blocklength Binary Deletion Channels

Hassan Khodaiemehr, Chen Feng

Preprint

In the authors' words

For a binary deletion channel operating on fixed-length inputs, the relevant figure of merit is the finite-blocklength capacity , not only the infinite-blocklength limit . We show that an optimal input may be chosen constant on complement and permutation-equivalence orbits, reducing the optimization to one weight per orbit, and introduce the optimized run distribution (ORD), an -parameter run-count model that coincides with for and is optimal among all run-count-constant inputs. An exact embedding-count dynamic program evaluates for these structured laws. A hybrid neural--exact procedure recovers certified ORD weights for ; variational critics (InfoNCE, NWJ, DV/MINE, SMILE) are used only as inner search objectives. Direct score-function learning of ORD weights collapses toward a flat run-length distribution (RLD) for . We therefore introduce ORD continuum transfer: a normalized run-count profile learned from exact small- ORD solutions is resampled at target lengths up to . Transferred ORD consistently outperforms RLD at moderate deletion probabilities; at the gain vanishes by roughly --. Exact ORD rates converted by Fertonani--Duman's length-entropy inequality are valid lower bounds on ; nested-Monte-Carlo evaluations of large- inputs are reported as diagnostics and are not claimed as capacity lower bounds. A fixed- sample-budget study quantifies nested-MC bias. All primary reported rates are values of .

Main resultLimitation the authors admit

Appeared: Tuesday, September 22. arXiv. Preprint, not yet peer-reviewed.

Authors' comment: 15 pages, 16 figures, a preliminary conference version of parts of this work was presented at CWIT 2024 conference