arXiv subjects

Pengfei Li

Publications and source records attributed to Pengfei Li.

At least 73 records · Page 4Linked to original sources

arXiv2025

TabPFN: One Model to Rule Them All?

Hollmann et al. (Nature 637 (2025) 319-326) recently introduced TabPFN, a transformer-based deep learning model for regression and classification on tabular data, which they claim "outperforms all previous methods on datasets with up to 10,000 samples by a wide margin, using substantially less training time." Furthermore, they have called TabPFN a "foundation model" for tabular data, as it can support "data generation, density estimation, learning reusable embeddings and fine-tuning". In this paper, we provide a tailored explanation of how TabPFN works for a statistics audience, by emphasizing its interpretation as approximate Bayesian inference. We then explore the significance of TabPFN to the field of statistics: We show that an out-of-the-box application of TabPFN can sometimes outperform specialized state-of-the-art methods for semi-supervised parameter estimation, prediction under covariate shift, and heterogeneous treatment effect estimation. As a partial explanation for the predictive effectiveness of TabPFN, we show that it can simultaneously adapt to both nonparametric structure and parametric structure, for instance, sometimes outperforming LASSO even when assumptions are correctly specified. All experiments can be reproduced using the code provided at https://github.com/qinglong-tian/tabpfn_study (https://github.com/qinglong-tian/tabpfn_study).

cs.LG↗

arXiv2025

Predicting Public Health Impacts of Electricity Usage

The electric power sector is a leading source of air pollutant emissions, impacting the public health of nearly every community. Although regulatory measures have reduced air pollutants, fossil fuels remain a significant component of the energy supply, highlighting the need for more advanced demand-side approaches to reduce the public health impacts. To enable health-informed demand-side management, we introduce HealthPredictor, a domain-specific AI model that provides an end-to-end pipeline linking electricity use to public health outcomes. The model comprises three components: a fuel mix predictor that estimates the contribution of different generation sources, an air quality converter that models pollutant emissions and atmospheric dispersion, and a health impact assessor that translates resulting pollutant changes into monetized health damages. Across multiple regions in the United States, our health-driven optimization framework yields substantially lower prediction errors in terms of public health impacts than fuel mix-driven baselines. A case study on electric vehicle charging schedules illustrates the public health gains enabled by our method and the actionable guidance it can offer for health-informed energy management. Overall, this work shows how AI models can be explicitly designed to enable health-informed energy management for advancing public health and broader societal well-being. Our datasets and code are released at: https://github.com/Ren-Research/Health-Impact-Predictor.

cs.LG↗

arXiv2025

The odd primordial halo of the Milky Way implied by Gaia. A shallow core, but a steep decline

Primordial dark matter halos are well understood from cold dark matter-only simulations. Since they can contract significantly as baryons settle into their centers, direct comparisons with observed galaxies are complicated. We present an approach to reversing the halo contraction by numerically calculating the halo response to baryonic infall and iterating the initial condition. This allowed us to derive spherically averaged primordial dark matter halos for observed galaxies. We applied this approach to the Milky Way and found that the latest Gaia measurements for the rotation velocities imply an odd primordial Galactic halo: Its concentration and total mass differ by more than 3$σ$ from the predictions, and the density profile presents an inner core that is too shallow and an outer decline that is too steep to be compatible with the cold dark matter paradigm.

astro-ph.GA↗

arXiv2025

Bohdi: Heterogeneous LLM Fusion with Automatic Data Exploration

Heterogeneous Large Language Model (LLM) fusion integrates the strengths of multiple source LLMs with different architectures into a target LLM with low computational overhead. While promising, existing methods suffer from two major limitations: 1) reliance on real data from limited domain for knowledge fusion, preventing the target LLM from fully acquiring knowledge across diverse domains, and 2) fixed data allocation proportions across domains, failing to dynamically adjust according to the target LLM's varying capabilities across domains, leading to a capability imbalance. To overcome these limitations, we propose Bohdi, a synthetic-data-only heterogeneous LLM fusion framework. Through the organization of knowledge domains into a hierarchical tree structure, Bohdi enables automatic domain exploration and multi-domain data generation through multi-model collaboration, thereby comprehensively extracting knowledge from source LLMs. By formalizing domain expansion and data sampling proportion allocation on the knowledge tree as a Hierarchical Multi-Armed Bandit problem, Bohdi leverages the designed DynaBranches mechanism to adaptively adjust sampling proportions based on the target LLM's performance feedback across domains. Integrated with our proposed Introspection-Rebirth (IR) mechanism, DynaBranches dynamically tracks capability shifts during target LLM's updates via Sliding Window Binomial Likelihood Ratio Testing (SWBLRT), further enhancing its online adaptation capability. Comparative experimental results on a comprehensive suite of benchmarks demonstrate that Bohdi significantly outperforms existing baselines on multiple target LLMs, exhibits higher data efficiency, and virtually eliminates the imbalance in the target LLM's capabilities. Our code is available at https://github.com/gjq100/Bohdi.git.

cs.LG↗

arXiv2025

A Survey of Reinforcement Learning for Large Reasoning Models

In this paper, we survey recent advances in Reinforcement Learning (RL) for reasoning with Large Language Models (LLMs). RL has achieved remarkable success in advancing the frontier of LLM capabilities, particularly in addressing complex logical tasks such as mathematics and coding. As a result, RL has emerged as a foundational methodology for transforming LLMs into LRMs. With the rapid progress of the field, further scaling of RL for LRMs now faces foundational challenges not only in computational resources but also in algorithm design, training data, and infrastructure. To this end, it is timely to revisit the development of this domain, reassess its trajectory, and explore strategies to enhance the scalability of RL toward Artificial SuperIntelligence (ASI). In particular, we examine research applying RL to LLMs and LRMs for reasoning abilities, especially since the release of DeepSeek-R1, including foundational components, core problems, training resources, and downstream applications, to identify future opportunities and directions for this rapidly evolving area. We hope this review will promote future research on RL for broader reasoning models. Github: https://github.com/TsinghuaC3I/Awesome-RL-for-LRMs

cs.CL↗

arXiv2025

The radial acceleration relation at the EDGE of galaxy formation: testing its universality in low-mass dwarf galaxies

A tight correlation between the baryonic and observed acceleration of galaxies has been reported over a wide range of mass ($10^8 < M_{\rm bar}/{\rm M}_\odot < 10^{11}$) - the Radial Acceleration Relation (RAR). This has been interpreted as evidence that dark matter is actually a manifestation of some modified weak-field gravity theory. In this paper, we study the radially resolved RAR of 12 nearby dwarf galaxies, with baryonic masses in the range $10^4 < M_{\rm bar}/{\rm M}_\odot < 10^{7.5}$, using a combination of literature data and data from the MUSE-Faint survey. We use stellar line-of-sight velocities and the Jeans modelling code GravSphere to infer the mass distributions of these galaxies, allowing us to compute the RAR. We compare the results with the EDGE simulations of isolated dwarf galaxies with similar stellar masses in a $Λ$CDM cosmology. We find that most of the observed dwarf galaxies lie systematically above the low-mass extrapolation of the RAR. Each galaxy traces a locus in the RAR space that can have a multi-valued observed acceleration for a given baryonic acceleration, while there is significant scatter from galaxy to galaxy. Our results indicate that the RAR does not apply to low-mass dwarf galaxies and that the inferred baryonic acceleration of these dwarfs does not contain enough information, on its own, to derive the observed acceleration. The simulated EDGE dwarfs behave similarly to the real data, lying systematically above the extrapolated RAR. We show that, in the context of modified weak-field gravity theories, these results cannot be explained by differential tidal forces from the Milky Way, nor by the galaxies being far from dynamical equilibrium, since none of the galaxies in our sample seems to experience strong tides. As such, our results provide further evidence for the need for invisible dark matter in the smallest dwarf galaxies.

astro-ph.GA↗

arXiv2025

Transfer Learning under Group-Label Shift: A Semiparametric Exponential Tilting Approach

We propose a new framework for binary classification in transfer learning settings where both covariate and label distributions may shift between source and target domains. Unlike traditional covariate shift or label shift assumptions, we introduce a group-label shift assumption that accommodates subpopulation imbalance and mitigates spurious correlations, thereby improving robustness to real-world distributional changes. To model the joint distribution difference, we adopt a flexible exponential tilting formulation and establish mild, verifiable identification conditions via an instrumental variable strategy. We develop a computationally efficient two-step likelihood-based estimation procedure that combines logistic regression for the source outcome model with conditional likelihood estimation using both source and target covariates. We derive consistency and asymptotic normality for the resulting estimators, and extend the theory to receiver operating characteristic curves, the area under the curve, and other target functionals, addressing the nonstandard challenges posed by plug-in classifiers. Simulation studies demonstrate that our method outperforms existing alternatives under subpopulation shift scenarios. A semi-synthetic application using the waterbirds dataset further confirms the proposed method's ability to transfer information effectively and improve target-domain classification accuracy.

stat.ME↗

arXiv2025

Semiparametric Learning from Open-Set Label Shift Data

We study the open-set label shift problem, where the test data may include a novel class absent from training. This setting is challenging because both the class proportions and the distribution of the novel class are not identifiable without extra assumptions. Existing approaches often rely on restrictive separability conditions, prior knowledge, or computationally infeasible procedures, and some may lack theoretical guarantees. We propose a semiparametric density ratio model framework that ensures identifiability while allowing overlap between novel and known classes. Within this framework, we develop maximum empirical likelihood estimators and confidence intervals for class proportions, establish their asymptotic validity, and design a stable Expectation-Maximization algorithm for computation. We further construct an approximately optimal classifier based on posterior probabilities with theoretical guarantees. Simulations and a real data application confirm that our methods improve both estimation accuracy and classification performance compared with existing approaches.

stat.ME↗

arXiv2025

Transformation of topological optical states via spiral modulation in fractional-diffraction systems

We propose a scheme for manipulations of a variety of topological states in fractional optical systems through spiral modulation of the local refraction index. An analytical approximation, based on a truncated finite-mode system, and direct simulations reveal that the spiral modulation supports direct mutual conversion between eigenmodes with topological-charge difference Δm=1, driven by the resonant coupling between the modes and the underlying spiral modulation. The conversion between eigenmodes with Δm=2 requires involvement of an intermediary mode and precise tuning of the resonance condition. We further explore the conversion of degenerate modes under the action of azimuthal modulation. Modulated degenerate modes exhibit incomplete conversion, evolving into intermediate states with odd parity symmetry. Finally, we examine nonlinear effects on the spiral-modulation-induced mode conversion, identifying an essential nonlinearity-induced resonance-frequency shift that critically affects the conversion efficiency.

physics.optics↗

arXiv2025

Emulating extended Lyman-alpha haloes around star-forming galaxies

Extended Ly$α$ emission is commonly observed around star-forming galaxies, opening a window for probing the neutral hydrogen gas in the circumgalactic medium (CGM). In this paper, we develop a prescription of spherically symmetric CGM gas properties and build emulators to model circularly-averaged surface brightness (SB) profiles of the extended Ly$α$ emission. With CGM gas properties parametrized by the density, velocity, and temperature profiles, a self-shielding calculation is carried out to obtain the neutral gas distribution with ionizing photons from the ultraviolet (UV) background and star formation in the galaxy. Our calculation reveals three types of systems with distinct neutral gas distribution: non-shielded systems with the CGM being highly ionized across all radii, shielded systems with a neutral gas shell shielding the UV background, and transitional systems in between. Ly$α$ SB profiles are obtained through Ly$α$ radiative transfer (RT) simulations, performed for the CGM models with three kinds of Ly$α$ sources: the star formation from central and satellite galaxies, and the recombination in the CGM. We build emulators to efficiently predict Ly$α$ SB profiles for given model parameters and Ly$α$ sources, based on Gaussian process regression. After being trained with only 180 RT simulations for each Ly$α$ source, the emulators reach an overall accuracy at the level of $\sim 20$ per cent. By applying the emulators to fit mock Ly$α$ SB profiles constructed from our model, we find a reasonable recovery of model parameters, indicating the potential of extracting physical information of the CGM and galaxies from the observed extended Ly$α$ emission.

astro-ph.GA↗

arXiv2025

Pseudo Empirical Likelihood Inference for Non-Probability Survey Samples

In this paper, the authors first provide an overview of two major developments on complex survey data analysis: the empirical likelihood methods and statistical inference with non-probability survey samples, and highlight the important research contributions to the field of survey sampling in general and the two topics in particular by Canadian survey statisticians. The authors then propose new inferential procedures on analyzing non-probability survey samples through the pseudo empirical likelihood approach. The proposed methods lead to asymptotically equivalent point estimators that have been discussed in the recent literature but possess more desirable features on confidence intervals such as range-respecting and data-driven orientation. Results from a simulation study demonstrate the superiority of the proposed methods in dealing with binary response variables.

stat.ME↗

arXiv2025

Observation of edge solitons and transitions between them in a trimer circuit lattice

In nonlinear topological systems, edge solitons either originate from linear topological edge modes or emerge as nonlinearity-induced localized states without topological protection. While electric circuits (ECs) provide a platform for realizing various types of topological insulators, observation of edge solitons and transitions between them in EC lattices remains a challenging problem. Here, we realize quench dynamics in nonlinear ECs to experimentally demonstrate both topologically nontrivial and trivial edge solitons in a trimer EC lattice and transitions between them. In the weakly nonlinear regime, we observe two types of topologically nontrivial edge solitons that originate from the corresponding linear topological edge states, characterized by the presence of mutually antisymmetric or symmetric peaks at two edge sites. Under strong nonlinearity, topologically trivial edge solitons with antisymmetric, symmetric, and asymmetric internal structures are discovered. The work suggests possibilities for exploring sophisticated nonlinear states and transitions between them in nonlinear topological systems.

nlin.PS↗

arXiv2025

Retrospective score tests versus prospective score tests for genetic association with case-control data

Since the seminal work by Prentice and Pyke (1979), the prospective logistic likelihood has become the standard method of analysis for retrospectively collected case-control data, in particular for testing the association between a single genetic marker and a disease outcome in genetic case-control studies. When studying multiple genetic markers with relatively small effects, especially those with rare variants, various aggregated approaches based on the same prospective likelihood have been developed to integrate subtle association evidence among all considered markers. In this paper we show that using the score statistic derived from a prospective likelihood is not optimal in the analysis of retrospectively sampled genetic data. We develop the locally most powerful genetic aggregation test derived through the retrospective likelihood under a random effect model assumption. In contrast to the fact that the disease prevalence information cannot be used to improve the efficiency for the estimation of odds ratio parameters in logistic regression models, we show that it can be utilized to enhance the testing power in genetic association studies. Extensive simulations demonstrate the advantages of the proposed method over the existing ones. One real genome-wide association study is analyzed for illustration.

stat.ME↗

arXiv2025

Instability of inverse probability weighting methods and a remedy for non-ignorable missing data

Inverse probability weighting (IPW) methods are commonly used to analyze non-ignorable missing data under the assumption of a logistic model for the missingness probability. However, solving IPW equations numerically may involve non-convergence problems when the sample size is moderate and the missingness probability is high. Moreover, those equations often have multiple roots, and identifying the best root is challenging. Therefore, IPW methods may have low efficiency or even produce biased results. We identify the pitfall in these methods pathologically: they involve the estimation of a moment-generating function, and such functions are notoriously unstable in general. As a remedy, we model the outcome distribution given the covariates of the completely observed individuals semiparametrically. After forming an induced logistic regression model for the missingness status of the outcome and covariate, we develop a maximum conditional likelihood method to estimate the underlying parameters. The proposed method circumvents the estimation of a moment-generating function and hence overcomes the instability of IPW methods. Our theoretical and simulation results show that the proposed method outperforms existing competitors greatly. Two real data examples are analyzed to illustrate the advantages of our method. We conclude that if only a parametric logistic regression is assumed but the outcome regression model is left arbitrary, then one has to be cautious in using any of the existing statistical methods in problems involving non-ignorable missing data.

stat.ME↗

arXiv2025

Causal Mediation Analysis for Zero-inflated Mixture Mediators

Causal mediation analysis is an important statistical tool to quantify effects transmitted by intermediate variables from a cause to an outcome. There is a gap in mediation analysis methods to handle mixture mediator data that are zero-inflated with multi-modality and atypical behaviors. We propose an innovative way to model zero-inflated mixture mediators from the perspective of finite mixture distributions to flexibly capture such mediator data. Multiple data types are considered for modeling such mediators including the zero-inflated log-normal mixture, zero-inflated Poisson mixture and zero-inflated negative binomial mixture. A two-part mediation effect is derived to better understand effects on outcomes attributable to the numerical change as well as binary change from 0 to 1 in mediators. The maximum likelihood estimates are obtained by an expectation maximization algorithm to account for unobserved mixture membership and whether an observed zero is a true or false zero. The optimal number of mixture components are chosen by a model selection criterion. The performance of the proposed method is demonstrated in a simulation study and an application to a neuroscience study in comparison with standard mediation analysis methods.

stat.ME↗

arXiv2025

Testing Homogeneity in a heteroscedastic contaminated normal mixture

Large-scale simultaneous hypothesis testing appears in many areas such as microarray studies, genome-wide association studies, brain imaging, disease mapping and astronomical surveys. A well-known inference method is to control the false discovery rate. One popular approach is to model the $z$-scores derived from the individual $t$-tests and then use this model to control the false discovery rate. We propose a new class of contaminated normal mixtures for modelling $z$-scores. We further design an EM-test for testing homogeneity in this class of mixture models. We show that the EM-test statistic has a shifted mixture of chi-squared limiting distribution. Simulation results show that the proposed testing procedure has accurate type I error and significantly larger power than its competitors under a variety of model specifications. A real-data example is analyzed to exemplify the application of the proposed method.

stat.ME↗

arXiv2025

Maximum multinomial likelihood estimation in compound mixture model with application to malaria study

Malaria can be diagnosed by the presence of parasites and symptoms (usually fever) due to the parasites. In endemic areas, however, an individual may have fever attributable either to malaria or to other causes. Thus, the parasite level of an individual with fever follows a two-component mixture, with the two components corresponding to malaria and nonmalaria individuals. Furthermore, the parasite levels of nonmalaria individuals can be characterized as a mixture of a zero component and a positive distribution. In this article, we propose a nonparametric maximum multinomial likelihood approach for estimating the proportion of malaria using parasite-level data from two groups of individuals collected in two different seasons. We develop an EM-algorithm to numerically calculate the proposed estimates and further establish their convergence rates. Simulation results show that the proposed estimators are more efficient than existing nonparametric estimators. The proposed method is used to analyze a malaria survey data.

stat.ME↗

arXiv2025

A powerful procedure that controls the false discovery rate with directional information

In many multiple testing applications in genetics, the signs of test statistics provide useful directional information, such as whether genes are potentially up- or down-regulated between two experimental conditions. However, most existing procedures that control the false discovery rate (FDR) are $p$-value based and ignore such directional information. We introduce a novel procedure, the signed-knockoff procedure, to utilize the directional information and control the FDR in finite samples. We demonstrate the power advantage of our procedure through simulation studies and two real applications.

stat.ME↗