CORTEXA
← Browse
zenodoPreprint2026-07-28

Importance for Retention Is Not Importance for Acquisition: A Falsifiable Test in Neural Networks

Joy Bose

Neural networks are empirically successful and mathematically only partially understood. This gap is often discussed as a single undifferentiated mystery, which makes the field feel either more solved or more mysterious than it actually is. We decompose the mystery into four distinct questions with different, unequal degrees of resolution: (1) why local, gradient-based optimization finds good solutions in a non-convex landscape; (2) whether a trained network's generalization can be certified rather than merely observed; (3) what a trained unit or circuit actually represents; and (4) whether these separate mathematical languages describe one underlying object. We survey the current state of each question, and identify a specific, underappreciated gap in the third: post-hoc, data-free measures of a unit's functional importance, such as the recently proposed HOPE framework, describe what a trained network now contains, but say nothing about when or how readily that content was acquired during training. We argue that importance for retention and importance for acquisition are conceptually distinct properties that existing post-hoc importance methods do not generally distinguish, and we propose a primary, falsifiable experiment that would test whether the two in fact coincide on networks whose target function is fully known, together with a second, cheaper, and independent experiment testing a related but distinct question about compression-based generalization certificates. Neither experiment requires new theory to run. We treat the first as the paper's central contribution and the second as a complementary check. A small, reduced-scale pilot of the primary experiment, run on a single-hidden-layer network rather than the full protocol proposed in the paper, is also reported; it shows a consistent, statistically significant association in every run, but at a scale too narrow to draw conclusions about the general claim, and we are explicit about why. We argue against premature claims of a single unifying mathematical object for learned knowledge until the full protocol, not this pilot, has actually been run.  

View free PDFSource page

Related papers

zenodoPreprint2026-07-29

MonteCarloJackknife.jl: Fast and Scalable Monte Carlo Approximation of Delete-d Jackknife Estimators in Julia

Soner AYDIN

This paper introduces MonteCarloJackknife.jl, an open-source Julia package that implements Monte Carlo approximation of delete-d jackknife estimators. Rather than exhaustively enumerating all deletion subsets, the package randomly samples a user-specified number of subsets, compu…

View free PDFSource page
zenodoPreprint2026-07-29

FOREX-SHIELD: A Multi-Modal Cyber-Defense Pipeline Combining Adversarially Hardened DeepLOB, Financial Transformers, and Zero-Knowledge Proofs for High-Frequency Foreign Exchange Settlement

Saiful Islam Tanvir

High-Frequency Foreign Exchange (FX) electronic execution networks process in excess of $7.5 trillion in daily spot volume across geographically distributed matching engines. Modern institutional trading infrastructure relies heavily on automated limit order book (LOB) forecastin…

View free PDFSource page
zenodoPreprint2026-07-28

Evaluation of complementary aspects of explainable AI techniques SHAP and LIME for deep neural networks for data sets in NLP domain

Ramesh Adeep Mohamed Arnest, Gursel Serpen

Abstract - Rapid advancements in large language models have enabled significant progress in solving complex real-world problems using deep neural networks (DNN). However, the black box nature of these DNN models poses significant challenges when it comes to explaining their decis…

View free PDFSource page