This work examines directional fidelity in a computable hard-marginal formulation that preserves direct black-box marginals in a computable hard-marginal formulation that preserves direct black-box marginals.
Abstract
Local conditional predictions from a black box need not admit a common joint distribution. We study two aspects of their coherent projection: finite-access precision and directional fidelity. For arbitrary deterministic binary conditional tables, we establish matching expected minimax excess conditional-Kullback--Leibler risk $\Theta(\min\{R\sqrt{d/B},d^2/B\})$ under independent uniform exact-table queries, with repeated draws charged, for a local bounded pairwise joint class. The result holds for sufficiently large $d$, $R\sqrt d\le M_0$, and $Cd^3\le B\le(d2^{d-1})^{1/3}$, where $M_0$ is fixed. A moment estimator attains the upper bound. The main technical contribution is a matching lower-bound construction that respects shared pairwise interactions and frozen exact-table responses, with comparison to the best projection of the same realized table. We then examine directional fidelity in a computable hard-marginal formulation that preserves direct black-box marginals. A preregistered TabPFN-v2 case study combines a Car anchor with a frozen two-dataset, three-split extension. All $1{,}152$ extension optimization instances pass independent numerical checks at registered tolerances. Directional contrasts are dataset-dependent and asymmetric: optimizing one direction need not improve the other. Together, the analyses distinguish accurate approximation of a selected projection from improvement in conditional directions outside its fitted objective.
This work derives complementary task-compression and individual-sample refinements, identifies a sharp obstruction showing why selective privacy alone need not give a vanishing certificate, and invert the bound to obtain the exact leakage allowance for a prescribed generalization tolerance.
A. Zamani, Parastoo Sadeghi, Mikael Skoglund· 0 citations
We study the offline release of a public family of $k$ statistical queries on a database of $N$ records from a public universe of size $d$, under pure $\varepsilon$-differential privacy. We give an efficient algorithm that, with probability at least $1-\beta$, answers all queries simultaneously to error $$O\left(\min\l...
Linear sketches estimate properties of high-dimensional inputs from a fixed low-dimensional projection. At constant accuracy, randomized Euclidean-norm sketches, notably the Johnson-Lindenstrauss transform, support exponentially many nonadaptive queries in sketch dimension $k$. For adaptive black-box queries, known pri...
This work implements its observation chain using the high-order noisy conditional sampler of [OpenAI 2026], obtaining improved dimension dependence under functional inequalities and obtaining a corresponding guarantee from a warm start with a known bound on its R\'enyi divergence.
The main converse recovers the phase transition of the optimal Type II error under the exponentially decaying Type I error constraint $\varepsilon_n=e^{-nr}$.
Roberto Bruno, Adrien Vandenbroucque, A. Esposito· arXiv.org· 0 citations
It is determined that sharp minimax rates for distribution estimation under pure local differential privacy at critical query smoothness are determined, and matching lower bounds allow finite public blackboard interaction and repeated participation under each user's total privacy budget.
Yi Liu, Xiao-Tian Chang, Qi-Rui Hu· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.