arXiv subjects

Yaé U. Gaba

Publications and source records attributed to Yaé U. Gaba.

4 recordsLinked to original sources

arXiv2026

Interpretable hybrid credit scoring for thin-file and underbanked populations

We extend a residual-learning hybrid credit scoring framework (logistic regression scorecard plus a gradient-boosting correction on its residuals, decomposed at each prediction into an interpretability ratio $ρ(x)$ that measures the share attributable to the linear branch) along three axes: an East African empirical instantiation on the Zindi Financial Inclusion in Africa data (Kenya, Rwanda, Tanzania, Uganda); a fairness audit at the granularity of the framework's three interpretability regions; and a thin-file segmentation analysis. On the Taiwan Credit Default benchmark retained for continuity, the calibrated hybrid attains AUC $= 0.776$ ($Δ\mathrm{AUC} = +0.057$ vs.\ standalone logistic regression, $+0.001$ vs.\ standalone XGBoost), reduces Brier Score by 23\%, and concentrates the highest-default-rate borrowers (69.5\%) in the fully interpretable region. On Zindi, the calibrated hybrid attains AUC $= 0.869$ ($Δ\mathrm{AUC} = +0.015$ vs.\ LR, $p < 0.001$; $-0.004$ vs.\ XGBoost), cuts Brier from $0.158$ to $0.085$ (a 46\% reduction), and replicates the regional routing pattern. The fairness audit detects severe routing into the opaque ML-driven region along socioeconomic axes: rural respondents by 18 percentage points relative to urban, primary-or-less-educated by 32 points relative to secondary-and-above, and Ugandan respondents by 22 points relative to Kenyan, while gender shows essentially no routing disparity. The audit pipeline surfaces subgroup-routing violations that aggregate fairness metrics miss, in a form directly usable by African central-bank supervisors of digital credit.

q-fin.RM↗

arXiv2026

Policy-Invariant Reward Shaping from LLM Feedback: A Framework for Hybrid RL Agents

Combining large language models with reinforcement learning is increasingly explored, yet the theoretical status of LLM-derived reward signals is often left implicit. We formalize the hybrid LLM-planner and RL-controller architecture as a Goal-Augmented Markov Decision Process and show that when the LLM per-state progress score is used as a bounded potential function, the resulting shaping term preserves the optimal policy set even when the LLM scores are inaccurate. This guarantee is stronger than what general LLM-as-reward approaches provide. We verify the result numerically on a small MDP under four potential configurations, including an adversarial one scaled to twenty times the base reward magnitude.

cs.LG↗

arXiv2026

On $U$-startpoints and $U$-fixed points in quasi-uniform spaces: a revisit and extensions

We extend the $\U$-fixed-point and $\U$-startpoint framework for quasi-uniform spaces. We reformulate the original existence framework through the filter axioms and supply a two-sided contraction condition on a generating family of quasi-pseudometrics. For this condition we prove a Banach-style fixed-point theorem on $T_0$ bicomplete quasi-uniform spaces, with a Picard-type error bound; a Boyd--Wong companion and a stability estimate follow. The conjugate quasi-uniformity yields an endpoint version, a selector argument extends the result to multivalued maps and to common startpoints of commuting families, a Knaster--Tarski variant covers monotone selectors on order-complete spaces, and a direct quasi-Hausdorff route handles weakly contractive multivalued maps. Moreover, the $T_0$-quotient removes the $T_0$ assumption from the corresponding involution-based fixed-point theorem.

math.GN↗

arXiv2026

Expansive homeomorphisms on complexity quasi-metric spaces

The complexity quasi-metric of Schellekens is a topological framework in which the asymmetry of computational comparisons -- ``$A$ is at most as fast as $B$'' carrying different information than ``$B$ is at most as slow as $A$'' -- is built into the distance itself. This paper develops the theory of expansive homeomorphisms on the resulting space. The central result is that the scaling transformation $ψ_α(f)(n)=αf(n)$ is expansive on the complexity space $(\C,d_\C)$ if and only if $α\neq 1$. The $δ$-stable sets of this dynamics turn out to coincide with asymptotic complexity classes, giving a dynamical characterisation of objects familiar from complexity theory. We then show that the canonical coordinates of $ψ_α$ are hyperbolic with contraction rate $λ=1/α$, and we connect orbit separation in the dynamical system to the classical time hierarchy theorem of Hartmanis and Stearns. Unstable sets, conjugate dynamics, and topological entropy estimates for the scaling map are also worked out. Concrete algorithms and Python implementations accompany every proof, so each result can be checked computationally; SageMath snippets sit alongside the examples, and the full code is in the \href{https://github.com/gabayae/expansive-homeomorphisms-complexity-qmetric}{companion repository}.

cs.CC↗