pipette
ESEspañol

Finite-Horizon Fisher Memory in Two-Sided Power-Bounded Recurrent Systems

Jeonghoon Lee

PreprintCode available

In the authors' words

We analyse allocation, admission and post-write retention in finite-horizon linear-Gaussian noisy recurrent memories. At every horizon, the directional Fisher memory satisfies : non-normality redistributes information but cannot raise its spherical average, while normal carriers satisfy . For bi-power-bounded carriers, we derive uniform lag bounds, identify the limit of with the inverse of the classical Ces\`aro asymptotic limit of , and give finite-horizon error bounds. A time-varying coupling defines an end-to-end store operator. The writer-optimal direction need not be store-optimal. After writing ends, an invertible hold preserves the full stored Fisher matrix. Additive contamination bounded by times the closure covariance retains at least of that matrix; a covariance-aware decoder attains the corresponding accuracy. With recurrent carriers held fixed, training input masks and linear readouts approached the task-specific optimum in 160 runs, with median normalized Rayleigh efficiency above . Binary accuracy matched the Gaussian prediction to mean absolute error below over more than four orders of magnitude in . In a separate pre-specified study of 320 runs, trained masks followed the designated input-time objective in both carrier types, in 16 of 16 draws. These studies used development-seen carriers and are pre-specified validations, not blind holdouts. The same fixed design reproduced the objective-specific result in 16 of 16 draws on carriers unused before run commitment. Exact isolation preserved information, while a decoder fixed at its training horizon fell to chance; inverse-adjoint transport restored its sampled decisions to numerical precision.

Main resultLimitation the authors admit

Appeared: Thursday, October 1. arXiv. Preprint, not yet peer-reviewed.

Authors' comment: 34 pages, 7 figures. Reproducibility materials: https://github.com/jeonghoon-ad/finite-horizon-fisher-memory (release v1.0)