CORTEXA
← Browse
arxivstat.MLcs.AIcs.LG2026-06-26

Spectral Perturbation of the Empirical Fisher Information Matrix under Weight Quantization

Rahid Zahid Alekberli, Hikmat Karimov

We study the spectral perturbation of the empirical Fisher Information Matrix (FIM) of a parametric statistical model under two structured perturbations: departure of the input from a reference (in-distribution) ensemble, and finite-precision (quantized) perturbation of the model's parameters. For the first, under an explicit local curvature-monotonicity hypothesis on the dominant eigenvalue lambda_max of the FIM, we show departure from a reference manifold provably elevates lambda_max relative to a calibration baseline (Proposition 3.2), and discuss why this hypothesis is required, since curvature need not increase monotonically under every perturbation. Our principal result is a directional eigenvalue perturbation bound, via Weyl's inequality, showing lambda_max under a quantization noise perturbation is lower bounded by its unperturbed value up to a third-order remainder, and, under a mild genericity condition, strictly exceeds it at leading order (Theorem 4.3). We give two tractable approximations to lambda_max -- one heuristic, one with a rigorous two-sided bound -- and a completeness result for a threshold-based partition of an augmented state space. These results motivate using sigma_t = lambda_max(F_t)/lambda_base as a runtime monitoring statistic for deployed language models: the quantization result offers a mechanism for an empirical observation of our own, where a calibration threshold for this statistic was approximately 244 times larger than a preliminary full-precision estimate on a 4-bit quantized model, a single measurement rather than a value derived in closed form. We report supporting measurements (twelve models, n=1,080 trajectories) broadly consistent with our predictions, discuss the scope and limitations of every result, and state as an open problem the closed-form prediction of the quantization inflation magnitude our bound does not supply.

View free PDFSource page

Related papers

arxivcs.LGcs.AIcs.CVcs.ITstat.ML2026-07-03

CuBAS: Information Geometric Curvature-Based Adaptive Sampling for Supervised Classification

Alexandre L. M. Levada

The informativeness of a training set is as consequential as its size, yet most sampling strategies remain agnostic to the intrinsic geometry of the data distribution. We introduce CuBAS (Curvature-Based Adaptive Sampling), an information-geometric framework for adaptive data sel…

View free PDFSource page
arxivstat.MLcs.AIcs.LG2026-06-29

A Stochastic--Geometric Theory of Scaling Laws in Grokking

Róisín Luo, Christian Gagné, Jonas Ngnawé, Ihsan Ullah, Karyn Morrissey

Delayed generalization (\ie~grokking) refers to the phenomenon in which a neural network fits its training data early in training but only begins to generalize after a prolonged delay, often through an abrupt transition. Despite extensive empirical study, its underlying mechanism…

View free PDFSource page
arxivcs.LGcs.AIstat.ML2026-06-26

S-GAI: Spectral Geometry-Aware Initialization for Sigmoidal MLPs -- From Dataset Geometry to Network Weights

Yi-Shan Chu

Classical universal approximation theorems establish the expressive power of sigmoidal multilayer perceptrons, but they do not prescribe how initial weights should encode the geometry of a data distribution. We propose S-GAI, a spectral geometry-aware initialization framework for…

View free PDFSource page
arxivmath.OCcs.AIcs.LGstat.ML2026-07-24

Explicit Iteration Complexity of Exact Data-Driven Inverse Optimization for Integer Linear Programs

Akira Kitaoka

A data-driven inverse optimization problem (DDIOP) is the problem of estimating the objective-function parameters (weights) that explain observed optimal-solution data, and it arises in many applications, including integer linear programming (ILP). It is known that, by applying g…

View free PDFSource page
arxivstat.MLcs.AIcs.LG2026-07-15

Price of Fairness in Bandits: A Tight Minimax Characterization

Dhruv Sarkar, Soumyadeep Dutta, Sayak Ray Chowdhury

In bandit problems, standard regret-minimizing algorithms treat exploration as an amortized cost, which can expose early participants to unfair ex-ante losses in settings such as clinical trials. Recent work addresses this by evaluating the sequence of per-round expected rewards…

View free PDFSource page
arxivcs.LGcs.AIcs.CLstat.ML2026-06-26

What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs

Nhi Nguyen, Shauli Ravfogel, Rajesh Ranganath

Large language models (LLMs) are increasingly deployed in high-stakes domains, where free-text explanations such as chain-of-thought and post-hoc rationales are used to justify model outputs. Yet it remains unclear whether these explanations are sufficient, i.e., if they contain…

View free PDFSource page