Selected work

Representative Papers

Online Policy Evaluation for MDPs with Dynamic UBSR Measures

Jul 25, 2026

This work addresses the limitations of existing risk-aware reinforcement learning methods for policy evaluation, which are often confined to specific risk measures or rely on simulators, thereby hindering their applicability in fully online settings. Focusing on Markov decision processes under the dynamic utility-based shortfall risk (UBSR) measure, the study introduces UBSR-TD—an efficient online policy evaluation algorithm based on linear function approximation—by extending the risk-neutral temporal difference (TD) algorithm through a tailored loss function, along with an accelerated variant. Theoretical analysis establishes its almost sure convergence, while numerical experiments confirm its empirical effectiveness. Furthermore, the approach demonstrates practical utility in managing perishable inventory with uncertain shelf life, thereby overcoming key applicability barriers in online risk-aware policy evaluation.

0 citationsRead paper

Controllable and Content-Based Recommendations

Jul 23, 2026

This work addresses the limited interpretability and controllability of traditional recommender systems, which rely on dense implicit representations that hinder user intervention. The authors propose CCBR, a novel framework that, for the first time, directly generates textual summaries from multimodal item content—such as images, audio, and video—to construct user profiles. These summaries serve as a textual bottleneck in collaborative filtering, enabling controllable, text-based recommendations. By integrating collaborative filtering with multimodal understanding and text generation, CCBR allows users to flexibly steer recommendation outcomes through either textual or multimodal inputs. Experiments demonstrate that CCBR achieves performance on par with conventional implicit models across multiple multimodal datasets, significantly outperforms the state-of-the-art controllable method TEARS, and effectively validates the practicality and interpretability of its user-guided mechanism.

0 citationsRead paper

Choosing the threshold in extreme value analysis

Jun 26, 2026

Threshold selection critically influences inference in extreme value analysis, yet its associated uncertainty is often overlooked. This study presents the first systematic evaluation of over forty threshold selection methods, integrating Hill estimators, visual diagnostics, goodness-of-fit tests, and extended generalized Pareto models. Through comprehensive simulation experiments and empirical analysis of the Padua daily rainfall series, the work assesses the statistical performance and applicability conditions of each method. The authors propose an automated strategy for optimal threshold selection and identify several approaches that exhibit superior robustness and efficiency. These findings offer practical guidance for improving reliability in extreme value modeling.

0 citationsRead paper

Incomplete Matrix Regression

Jun 24, 2026

This work proposes a distribution-free penalized regression framework to jointly model covariate effects and low-rank latent structure in sparse, noisy observations that exhibit row/column covariates and structural dependencies. The method integrates Lasso, ridge-type kernel regularization, and low-rank decomposition, and is efficiently solved via a scalable alternating least squares algorithm that flexibly accommodates prior similarity information. Theoretically, non-asymptotic error bounds are established, while computationally the approach achieves substantial reductions in complexity. Empirical evaluations on both synthetic and real-world data demonstrate that the proposed method attains prediction accuracy comparable to existing sophisticated approaches at significantly lower computational cost. The accompanying algorithm is publicly available as the R package IMR.

0 citationsRead paper

Financing Artificial Intelligence Infrastructure: Mapping AI Infrastructure Investment and Compute Governance Across Africa

Jun 24, 2026

Current AI governance frameworks disproportionately emphasize the geographic accessibility of computational power while neglecting the underlying dynamics of capital flows, ownership structures, and control mechanisms, thereby falling short of achieving genuine compute equity. This study systematically examines 46 AI infrastructure projects across Africa between 2019 and 2025, representing a total investment of $12.7 billion, integrating systematic literature review, publicly available data, and a value chain analysis framework. It reveals that 73% of funding is concentrated in capital and physical infrastructure, with compute control heavily centralized among a few global technology giants, and investments markedly clustered in South Africa, Kenya, Nigeria, and Egypt. Introducing the concept of “asymmetric interdependence,” this work advocates for incorporating capital, ownership, and control dimensions into compute governance to transcend the prevailing paradigm centered solely on technological access.

0 citationsRead paper

Recent publications

Latest Papers

Online Policy Evaluation for MDPs with Dynamic UBSR Measures

Jul 25, 2026

This work addresses the limitations of existing risk-aware reinforcement learning methods for policy evaluation, which are often confined to specific risk measures or rely on simulators, thereby hindering their applicability in fully online settings. Focusing on Markov decision processes under the dynamic utility-based shortfall risk (UBSR) measure, the study introduces UBSR-TD—an efficient online policy evaluation algorithm based on linear function approximation—by extending the risk-neutral temporal difference (TD) algorithm through a tailored loss function, along with an accelerated variant. Theoretical analysis establishes its almost sure convergence, while numerical experiments confirm its empirical effectiveness. Furthermore, the approach demonstrates practical utility in managing perishable inventory with uncertain shelf life, thereby overcoming key applicability barriers in online risk-aware policy evaluation.

0 citationsRead paper

Controllable and Content-Based Recommendations

Jul 23, 2026

This work addresses the limited interpretability and controllability of traditional recommender systems, which rely on dense implicit representations that hinder user intervention. The authors propose CCBR, a novel framework that, for the first time, directly generates textual summaries from multimodal item content—such as images, audio, and video—to construct user profiles. These summaries serve as a textual bottleneck in collaborative filtering, enabling controllable, text-based recommendations. By integrating collaborative filtering with multimodal understanding and text generation, CCBR allows users to flexibly steer recommendation outcomes through either textual or multimodal inputs. Experiments demonstrate that CCBR achieves performance on par with conventional implicit models across multiple multimodal datasets, significantly outperforms the state-of-the-art controllable method TEARS, and effectively validates the practicality and interpretability of its user-guided mechanism.

0 citationsRead paper

Choosing the threshold in extreme value analysis

Jun 26, 2026

Threshold selection critically influences inference in extreme value analysis, yet its associated uncertainty is often overlooked. This study presents the first systematic evaluation of over forty threshold selection methods, integrating Hill estimators, visual diagnostics, goodness-of-fit tests, and extended generalized Pareto models. Through comprehensive simulation experiments and empirical analysis of the Padua daily rainfall series, the work assesses the statistical performance and applicability conditions of each method. The authors propose an automated strategy for optimal threshold selection and identify several approaches that exhibit superior robustness and efficiency. These findings offer practical guidance for improving reliability in extreme value modeling.

0 citationsRead paper

Incomplete Matrix Regression

Jun 24, 2026

This work proposes a distribution-free penalized regression framework to jointly model covariate effects and low-rank latent structure in sparse, noisy observations that exhibit row/column covariates and structural dependencies. The method integrates Lasso, ridge-type kernel regularization, and low-rank decomposition, and is efficiently solved via a scalable alternating least squares algorithm that flexibly accommodates prior similarity information. Theoretically, non-asymptotic error bounds are established, while computationally the approach achieves substantial reductions in complexity. Empirical evaluations on both synthetic and real-world data demonstrate that the proposed method attains prediction accuracy comparable to existing sophisticated approaches at significantly lower computational cost. The accompanying algorithm is publicly available as the R package IMR.

0 citationsRead paper

Financing Artificial Intelligence Infrastructure: Mapping AI Infrastructure Investment and Compute Governance Across Africa

Jun 24, 2026

Current AI governance frameworks disproportionately emphasize the geographic accessibility of computational power while neglecting the underlying dynamics of capital flows, ownership structures, and control mechanisms, thereby falling short of achieving genuine compute equity. This study systematically examines 46 AI infrastructure projects across Africa between 2019 and 2025, representing a total investment of $12.7 billion, integrating systematic literature review, publicly available data, and a value chain analysis framework. It reveals that 73% of funding is concentrated in capital and physical infrastructure, with compute control heavily centralized among a few global technology giants, and investments markedly clustered in South Africa, Kenya, Nigeria, and Egypt. Introducing the concept of “asymmetric interdependence,” this work advocates for incorporating capital, ownership, and control dimensions into compute governance to transcend the prevailing paradigm centered solely on technological access.

0 citationsRead paper