arXiv subjects

Xiaodong Yan

Publications and source records attributed to Xiaodong Yan.

At least 19 recordsLinked to original sources

arXiv2026

EFQ-Softmax: Exp-Free Quantization for Softmax

Low-bit attention accelerates Transformer inference by moving the $QK^\top$ and $PV$ matrix multiplications to FP8 or FP4 matrix engines. However, the softmax path often evaluates shifted-score exponentials in higher precision, forms a temporary probability block, and quantizes it before low-bit $PV$ multiplication. This exp-then-quantize path creates a mismatch between a high-precision probability producer and a low-bit matrix consumer. We propose EFQ-Softmax (Exp-Free Quantization for Softmax), a low-bit probability-generation method that directly maps shifted attention scores to block-scaled E2M1 operands. For each microscaling block, EFQ-Softmax selects an exponent-only scale from the local maximum, maps the shifted scores to a normalized residual domain, and generates nonnegative E2M1 probability codes using a single affine rule. The resulting operand is used consistently in both the $\widetilde{P}V$ numerator update and the $\widetilde{P}\mathbf{1}$ denominator update. The FlashAttention-style row-maximum update, historical rescaling, high-precision accumulation, and final normalization remain unchanged. We evaluate end-to-end quality on Qwen3-8B, Qwen3-VL-8B-Instruct, and WAN2.2-TI2V-5B, and separately measure kernel-level performance on the A5 vector unit. EFQ-Softmax improves the Qwen3-8B seven-task mean from 0.6749 with MXFP4 to 0.6773 and the Qwen3-VL nine-task mean from 0.7826 to 0.8000. On WAN2.2, it maintains temporal consistency and visual quality comparable to the FP16 and MXFP4 baselines under VBench. On the A5 vector unit, EFQ-Softmax reduces the vector-stage latency of the fused probability-generation kernel by 40.33% on average across sequence lengths from 16K to 128K. These results show that direct low-bit probability generation can replace the conventional exp-then-quantize path while preserving end-to-end model quality.

cs.LG↗

arXiv2026

Robust Code RL via Faulty-Code-Driven Test case Synthesis and Dense Reward Shaping

Reinforcement Learning from Verifiable Rewards (RLVR) is pivotal for enhancing LLM code generation, yet its efficacy is often hindered by insufficient test case coverage, leading to reward hacking and policy degradation. To address this, we propose RobustTests, a framework featuring a faulty-code-driven test case synthesis strategy. By leveraging "near-correct" faulty codes, RobustTests captures latent logical discrepancies and employs validator agents with behavioral feature clustering to filter invalid or redundant test cases. Additionally, a stepwise dense reward function based on pass rates is introduced to mitigate false negatives and enhance training robustness. Using this pipeline, we construct an augmented version of the CodeContests+ dataset with superior diagnostic utility. Experimental results show that RL fine-tuning of Qwen3-32B via RobustTests achieves a 3% absolute gain on LiveCodeBench, demonstrating its effectiveness in advancing LLM code generation proficiency. Codes and data are available at https://huggingface.co/datasets/sid6/RobustTests.

cs.AI↗

arXiv2026

Polar director structure of SmAP$_\text{F}$ phase of bent-core liquid crystals in thin planar cells with bias electric field

We study the polar director structure in thin planar cells filled with bent-core liquid crystals in the ferroelectric smectic-A phase (SmAP$_\text{F}$). We analyze a continuum phenomenological model proposed in the physics literature and present rigorous proofs of the existence and uniqueness of the equilibrium solutions. We further investigate the qualitative properties of nontrivial solutions and examine the effects of a bias electric field, surface anchoring, and cell thickness on the polar director configuration. Our results are consistent with previous experimental and numerical simulations reported in the physics literature. In addition, our analysis reveals new parameter-dependent behaviors supported by our numerical simulations and extends results reported from previous literature.

math.AP↗

arXiv2026

Asymptotic Theory and Sequential Testing for Adaptive Bandits

Multi-armed bandit (MAB) processes constitute a foundational subclass of reinforcement learning problems and represent a central topic in statistical decision theory. Yet, conducting valid sequential testing under adaptive allocation remains challenging due to the lack of asymptotic theory under non-i.i.d. reward sequences and sublinear sample sizes for some arms. To address this open challenge, we propose an Urn Bandit (UNB) process to integrate the reinforcement mechanism of urn probabilistic models with MAB principles, ensuring almost sure concentration of allocation proportions on optimal arms. We establish a joint functional central limit theorem (FCLT) for consistent estimators of expected rewards under non-i.i.d. reward sequences with non-sub-Gaussian tails and pairwise cross-arm dependence. To overcome the limitations of existing methods that focus mainly on cumulative regret and therefore provide only algorithmic performance guarantees without supporting valid sequential testing, we develop an asymptotic theory for sequential test statistics under the proposed UNB process. The resulting framework enables a broad class of sequential inference procedures, such as A/B testing and policy evaluation. Simulation studies and real data analysis demonstrate that UNB maintains testing performance comparable to that of the equal randomization (ER) design while achieving improved reward accumulation relative to ER.

stat.ME↗

arXiv2026

Mixed Effects Mixture of Experts: Modeling Double Heterogeneous Trajectories

Linear mixed-effects model (LMM) is a cornerstone of longitudinal data analysis, but is limited to adeptly make heterogeneous analyses predictable under both group-specific fixed effects and subject-specific random effects. To address this challenge, we propose a novel statistical framework by using a large model prototype: a mixed effects mixture of experts model (MEMoE). This framework integrates the divide-and-conquer paradigm of Mixture of Experts Models with classical mixed-effect modeling. In the proposed MEMoE, each expert is a full LMM dedicated to capturing the longitudinal trajectory of a specific latent subpopulation, while another model gating function learns to route subjects to the most appropriate expert in a data-driven manner based on baseline covariates. We develop a robust inferential procedure for parameter estimation based on the Laplace Expectation-Maximization algorithm, with standard errors calibrated using robust sandwich estimators to account for potential model misspecification. Extensive simulation studies and an empirical application demonstrate that MEMoE outperforms both traditional single-population LMM and conventional Mixture of Experts models in terms of parameter recovery, classification accuracy, and overall model fit.

stat.ME↗

arXiv2025

Intrinsic Benefits of Categorical Distributional Loss: Uncertainty-aware Regularized Exploration in Reinforcement Learning

The remarkable empirical performance of distributional reinforcement learning (RL) has garnered increasing attention to understanding its theoretical advantages over classical RL. By decomposing the categorical distributional loss commonly employed in distributional RL, we find that the potential superiority of distributional RL can be attributed to a derived distribution-matching entropy regularization. This less-studied entropy regularization aims to capture additional knowledge of return distribution beyond only its expectation, contributing to an augmented reward signal in policy optimization. In contrast to the vanilla entropy regularization in MaxEnt RL, which explicitly encourages exploration by promoting diverse actions, the novel entropy regularization derived from categorical distributional loss implicitly updates policies to align the learned policy with (estimated) environmental uncertainty. Finally, extensive experiments verify the significance of this uncertainty-aware regularization from distributional RL on the empirical benefits over classical RL. Our study offers an innovative exploration perspective to explain the intrinsic benefits of distributional learning in RL.

cs.LG↗

arXiv2025

Optimization via Strategic Law of Large Numbers

This paper proposes a unified framework for the global optimization of a continuous function in a bounded rectangular domain. Specifically, we show that: (1) under the optimal strategy for a two-armed decision model, the sample mean converges to a global optimizer under the Strategic Law of Large Numbers, and (2) a sign-based strategy built upon the solution of a parabolic PDE is asymptotically optimal. Motivated by this result, we propose a class of {\bf S}trategic {\bf M}onte {\bf C}arlo {\bf O}ptimization (SMCO) algorithms, which uses a simple strategy that makes coordinate-wise two-armed decisions based on the signs of the partial gradient of the original function being optimized over (without the need of solving PDEs). While this simple strategy is not generally optimal, we show that it is sufficient for our SMCO algorithm to converge to local optimizer(s) from a single starting point, and to global optimizers under a growing set of starting points. Numerical studies demonstrate the suitability of our SMCO algorithms for global optimization, and illustrate the promise of our theoretical framework and practical approach. For a wide range of test functions with challenging optimization landscapes (including ReLU neural networks with square and hinge loss), our SMCO algorithms converge to the global maximum accurately and robustly, using only a small set of starting points (at most 100 for dimensions up to 1000) and a small maximum number of iterations (200). In fact, our algorithms outperform many state-of-the-art global optimizers, as well as local algorithms augmented with the same set of starting points as ours.

math.OC↗

arXiv2025

Every Step Evolves: Scaling Reinforcement Learning for Trillion-Scale Thinking Model

We present Ring-1T, the first open-source, state-of-the-art thinking model with a trillion-scale parameter. It features 1 trillion total parameters and activates approximately 50 billion per token. Training such models at a trillion-parameter scale introduces unprecedented challenges, including train-inference misalignment, inefficiencies in rollout processing, and bottlenecks in the RL system. To address these, we pioneer three interconnected innovations: (1) IcePop stabilizes RL training via token-level discrepancy masking and clipping, resolving instability from training-inference mismatches; (2) C3PO++ improves resource utilization for long rollouts under a token budget by dynamically partitioning them, thereby obtaining high time efficiency; and (3) ASystem, a high-performance RL framework designed to overcome the systemic bottlenecks that impede trillion-parameter model training. Ring-1T delivers breakthrough results across critical benchmarks: 93.4 on AIME-2025, 86.72 on HMMT-2025, 2088 on CodeForces, and 55.94 on ARC-AGI-1. Notably, it attains a silver medal-level result on the IMO-2025, underscoring its exceptional reasoning capabilities. By releasing the complete 1T parameter MoE model to the community, we provide the research community with direct access to cutting-edge reasoning capabilities. This contribution marks a significant milestone in democratizing large-scale reasoning intelligence and establishes a new baseline for open-source model performance.

cs.CL↗

arXiv2025

Optimal Transport Learning: Balancing Value Optimization and Fairness in Individualized Treatment Rules

Individualized treatment rules (ITRs) have gained significant attention due to their wide-ranging applications in fields such as precision medicine, ridesharing, and advertising recommendations. However, when ITRs are influenced by sensitive attributes such as race, gender, or age, they can lead to outcomes where certain groups are unfairly advantaged or disadvantaged. To address this gap, we propose a flexible approach based on the optimal transport theory, which is capable of transforming any optimal ITR into a fair ITR that ensures demographic parity. Recognizing the potential loss of value under fairness constraints, we introduce an ``improved trade-off ITR," designed to balance value optimization and fairness while accommodating varying levels of fairness through parameter adjustment. To maximize the value of the improved trade-off ITR under specific fairness levels, we propose a smoothed fairness constraint for estimating the adjustable parameter. Additionally, we establish a theoretical upper bound on the value loss for the improved trade-off ITR. We demonstrate performance of the proposed method through extensive simulation studies and application to the Next 36 entrepreneurial program dataset.

stat.ML↗

arXiv2025

A Two-armed Bandit Framework for A/B Testing

A/B testing is widely used in modern technology companies for policy evaluation and product deployment, with the goal of comparing the outcomes under a newly-developed policy against a standard control. Various causal inference and reinforcement learning methods developed in the literature are applicable to A/B testing. This paper introduces a two-armed bandit framework designed to improve the power of existing approaches. The proposed procedure consists of three main steps: (i) employing doubly robust estimation to generate pseudo-outcomes, (ii) utilizing a two-armed bandit framework to construct the test statistic, and (iii) applying a permutation-based method to compute the $p$-value. We demonstrate the efficacy of the proposed method through asymptotic theories, numerical experiments and real-world data from a ridesharing company, showing its superior performance in comparison to existing methods.

stat.ML↗

arXiv2025

Strategic A/B testing via Maximum Probability-driven Two-armed Bandit

Detecting a minor average treatment effect is a major challenge in large-scale applications, where even minimal improvements can have a significant economic impact. Traditional methods, reliant on normal distribution-based or expanded statistics, often fail to identify such minor effects because of their inability to handle small discrepancies with sufficient sensitivity. This work leverages a counterfactual outcome framework and proposes a maximum probability-driven two-armed bandit (TAB) process by weighting the mean volatility statistic, which controls Type I error. The implementation of permutation methods further enhances the robustness and efficacy. The established strategic central limit theorem (SCLT) demonstrates that our approach yields a more concentrated distribution under the null hypothesis and a less concentrated one under the alternative hypothesis, greatly improving statistical power. The experimental results indicate a significant improvement in the A/B testing, highlighting the potential to reduce experimental costs while maintaining high statistical power.

stat.ML↗

arXiv2025

Ring-lite: Scalable Reasoning via C3PO-Stabilized Reinforcement Learning for LLMs

We present Ring-lite, a Mixture-of-Experts (MoE)-based large language model optimized via reinforcement learning (RL) to achieve efficient and robust reasoning capabilities. Built upon the publicly available Ling-lite model, a 16.8 billion parameter model with 2.75 billion activated parameters, our approach matches the performance of state-of-the-art (SOTA) small-scale reasoning models on challenging benchmarks (e.g., AIME, LiveCodeBench, GPQA-Diamond) while activating only one-third of the parameters required by comparable models. To accomplish this, we introduce a joint training pipeline integrating distillation with RL, revealing undocumented challenges in MoE RL training. First, we identify optimization instability during RL training, and we propose Constrained Contextual Computation Policy Optimization(C3PO), a novel approach that enhances training stability and improves computational throughput via algorithm-system co-design methodology. Second, we empirically demonstrate that selecting distillation checkpoints based on entropy loss for RL training, rather than validation metrics, yields superior performance-efficiency trade-offs in subsequent RL training. Finally, we develop a two-stage training paradigm to harmonize multi-domain data integration, addressing domain conflicts that arise in training with mixed dataset. We will release the model, dataset, and code.

cs.CL↗

arXiv2025

Demographic Parity-aware Individualized Treatment Rules

There has been growing interest in developing optimal individualized treatment rules (ITRs) in various fields, such as precision medicine, business decision-making, and social welfare distribution. The application of ITRs within a societal context raises substantial concerns regarding potential discrimination over sensitive attributes such as age, gender, or race. To address this concern directly, we introduce the concept of demographic parity in ITRs. However, estimating an optimal ITR that satisfies the demographic parity requires solving a non-convex constrained optimization problem. To overcome these computational challenges, we employ tailored fairness proxies inspired by demographic parity and transform it into a convex quadratic programming problem. Additionally, we establish the consistency of the proposed estimator and the risk bound. The performance of the proposed method is demonstrated through extensive simulation studies and real data analysis.

stat.ME↗

arXiv2024

Low-Dimensional Solid-State Single-Photon Emitters

Solid-state single-photon emitters (SPEs) are attracting significant attention as fundamental components in quantum computing, communication, and sensing. Low-dimensional materials-based SPEs (LD-SPEs) have drawn particular interest due to their high photon extraction efficiency, ease of integration with photonic circuits, and strong coupling with external fields. The accessible surfaces of LD materials allow for deterministic control over quantum light emission, while enhanced quantum confinement and light-matter interactions improve photon emissive properties. This review examines recent progress in LDSPEs across four key materials: zero-dimensional (0D) semiconductor quantum dots, one-dimensional (1D) nanotubes, two-dimensional (2D) materials, including hexagonal boron nitride (hBN) and transition metal dichalcogenides (TMDCs). We explore their structural and photophysical properties, along with techniques such as spectral tuning and cavity coupling that enhance SPE performance. Finally, we address future challenges and suggest strategies for optimizing LD-SPEs for practical quantum applications.

physics.optics↗

arXiv2024

Telecom-wavelength Single-photon Emitters in Multi-layer InSe

The development of robust and efficient single photon emitters (SPEs) at telecom wavelengths is critical for advancements in quantum information science. Two-dimensional (2D) materials have recently emerged as promising sources for SPEs, owing to their high photon extraction efficiency, facile coupling to external fields, and seamless integration into photonic circuits. In this study, we demonstrate the creation of SPEs emitting in the 1000 to 1550 nm near-infrared range by coupling 2D indium selenide (InSe) with strain-inducing nanopillar arrays. The emission wavelength exhibits a strong dependence on the number of layers. Hanbury Brown and Twiss experiments conducted at 10 K reveal clear photon antibunching, confirming the single-photon nature of the emissions. Density-functional-theory calculations and scanning-tunneling-microscopy analyses provide insights into the electronic structures and defect states, elucidating the origins of the SPEs. Our findings highlight the potential of multilayer 2D metal monochalcogenides for creating SPEs across a broad spectral range, paving the way for their integration into quantum communication technologies.

physics.optics↗

arXiv2023

DRGCN: Dynamic Evolving Initial Residual for Deep Graph Convolutional Networks

Graph convolutional networks (GCNs) have been proved to be very practical to handle various graph-related tasks. It has attracted considerable research interest to study deep GCNs, due to their potential superior performance compared with shallow ones. However, simply increasing network depth will, on the contrary, hurt the performance due to the over-smoothing problem. Adding residual connection is proved to be effective for learning deep convolutional neural networks (deep CNNs), it is not trivial when applied to deep GCNs. Recent works proposed an initial residual mechanism that did alleviate the over-smoothing problem in deep GCNs. However, according to our study, their algorithms are quite sensitive to different datasets. In their setting, the personalization (dynamic) and correlation (evolving) of how residual applies are ignored. To this end, we propose a novel model called Dynamic evolving initial Residual Graph Convolutional Network (DRGCN). Firstly, we use a dynamic block for each node to adaptively fetch information from the initial representation. Secondly, we use an evolving block to model the residual evolving pattern between layers. Our experimental results show that our model effectively relieves the problem of over-smoothing in deep GCNs and outperforms the state-of-the-art (SOTA) methods on various benchmark datasets. Moreover, we develop a mini-batch version of DRGCN which can be applied to large-scale data. Coupling with several fair training techniques, our model reaches new SOTA results on the large-scale ogbn-arxiv dataset of Open Graph Benchmark (OGB). Our reproducible code is available on GitHub.

cs.LG↗

arXiv2023

A phase field model for the motion of prismatic dislocation loops by both climb and self-climb

We study the sharp interface limit and well-posedness of a phase field model for self-climb of prismatic dislocation loops in periodic settings. The model is set up in a Cahn-Hilliard/Allen-Cahn framework featured with degenerate phase-dependent diffusion mobility with an additional stablizing function. Moreover, a nonlocal climb force is added to the chemical potential. We introduce a notion of weak solutions for the nonlinear model. The existence result is obtained by approximations of the proposed model with nondegenerate mobilities. Lastly, the numerical simulations are performed to validate the phase field model and the simulation results show the big difference for the prismatic dislocation loops in the evolution time and the pattern with and without self-climb contribution.

math.AP↗

arXiv2022

A smectic liquid crystal model in the periodic setting

We consider the asymptotic behavior as $\varepsilon $ goes to zero of the 2D smectics model in the periodic setting given by \begin{equation*} \mathcal{E}_{\varepsilon }( w) =\frac{1}{2}\int_{\mathbb{T}^{2}}\frac{1}{ \varepsilon }\left( \left\vert \partial_{1}\right\vert ^{-1}\left( \partial_{2}w-\partial_{1}\frac{1}{2}w^{2}\right) \right) ^{2}+\varepsilon \left( \partial_{1}w\right) ^{2}dx . \end{equation*} We show that the energy $\mathcal{E}_\varepsilon(w)$ controls suitable $L^p$ and Besov norms of $w$ and use this to demonstrate the existence of minimizers for $\mathcal{E}_\varepsilon(w)$, which has not been proved for this smectics model before, and compactness in $L^p$ for an energy-bounded sequence. We also prove an asymptotic lower bound for $\mathcal{E}_\varepsilon(w)$ as $\varepsilon \to 0$ by means of an entropy argument.

math.AP↗