Searcharxiv⌕ Search

arXiv subjects

Yanfeng Chen

Publications and source records attributed to Yanfeng Chen.

15 recordsLinked to original sources
arXiv2026

Dynamic Important Example Mining for Reinforcement Finetuning

Reinforcement fine-tuning (RFT) is increasingly used to strengthen the reasoning abilities of large models, yet its effectiveness is bound by how training data are selected and used. Most data-centric RFT methods rely on static or heuristic sample selection, implicitly assuming a sample's value is fixed over training. This overlooks the non-stationary dynamics of policy learning and can lead to suboptimal updates. We propose Dynamic Important Example Mining (DIEM), a principled and fully automated framework that makes data utilization adaptive throughout RFT. DIEM integrates two components into each optimization step: (i) a gradient-alignment importance estimator that efficiently approximates each sample's marginal contribution to policy improvement; and (ii) a constrained batch reweighting scheme that maximizes aggregate utility while preserving the update's gradient magnitude to stabilize optimization. Across several reasoning benchmarks, DIEM consistently outperforms strong static and dynamic baselines. The code will be released via https://github.com/hrtan/DIEM.

cs.AI↗
arXiv2026

FastMix: Fast Data Mixture Optimization via Gradient Descent

While large and diverse datasets have driven recent advances in large models, identifying the optimal data mixture for pre-training and post-training remains a significant open problem. We address this challenge with FASTMIX, a novel framework that automates data mixture discovery while training only a single proxy model. Instead of relying on predefined heuristics or resource-intensive simulations, FASTMIX jointly optimizes mixture coefficients and model parameters, substantially improving efficiency and scalability over prior approaches. At the core of FASTMIX is a reformulation of mixture selection as a bilevel optimization problem. Under this reformulation, we show that optimizing mixture ratios is mathematically equivalent to assigning per-source loss weights under uniform source sampling. This embeds the mixture coefficients directly into the differentiable iterative optimization objective, enabling efficient, gradient-based optimization of both mixture and model. To solve the optimization problem, FASTMIX implements an approximate iterative optimization procedure, alternating between (i) updating model parameters on data sampled according to current mixture ratios (inner loop) and (ii) updating mixture ratios based on validation feedback (outer loop). Across pre- and post-training, FASTMIX outperforms baselines while drastically reducing search cost. Code (https://github.com/hrtan/fastmix)

cs.LG↗
arXiv2026

Hy-MT2: A Family of Fast, Efficient and Powerful Multilingual Translation Models in the Wild

Hy-MT2 is a family of fast-thinking multilingual translation models designed for complex real-world scenarios. It includes three model sizes: 1.8B, 7B, and 30B-A3B (MoE), all of which support translation among 33 languages and effectively follow translation instructions in multiple languages. Multi-dimensional evaluations show that Hy-MT2 delivers outstanding performance across general, real-world business, domain-specific, and instruction-following translation tasks. The 7B and 30B models outperform open-source models such as DeepSeek-V4-Pro and Kimi K2.6 in fast-thinking mode, while the lightweight 1.8B model also surpasses mainstream commercial APIs from providers such as Microsoft and Doubao overall. Moreover, when paired with AngelSlim's 1.25-bit extreme quantization for on-device deployment, the lightweight 1.8B model requires only 440 MB of storage and achieves a 1.5x inference speedup.

cs.CL↗
arXiv2026

Physics-Informed Generative Solver: Bridging Data-Driven Priors and Conservation Laws for Stable Spatiotemporal Field Reconstruction

Reconstructing continuous physical fields from sparse measurements is a central inverse problem, but data-driven generative models can produce states that violate governing dynamics. We introduce a physics-informed generative solver that separates stable prior learning from inference-time enforcement of conservation laws. Martingale-Regularized Score Matching regularizes score pretraining with a Score Fokker-Planck constraint, yielding a dynamically stable prior. Physics-Informed Implicit Score Sampling then guides denoising trajectories by gradients of physical residuals, projecting samples toward admissible manifolds without retraining. In acoustics, the method co-generates pressure and particle velocity from sparse sensors, enabling dense virtual arrays that suppress spatial aliasing. The same framework generalizes to real-world ERA5 meteorological fields under extreme sparsity. Together, this work establishes a rigorous and generalizable paradigm for solving high-dimensional inverse problems, bridging the gap between generative artificial intelligence and first-principles science.

cs.LG↗
arXiv2026

Topological phononics

Topological phononics extends the foundational concepts of topological condensed matter physics to the realm of lattice vibrations and classical mechanical waves, unlocking robust, defect-immune states and phenomena beyond the reach of conventional phononic engineering. This review provides a unified, systematic framework for understanding topological phonons across natural and artificial systems, spanning solid-state materials, acoustic/mechanical metamaterials, and non-Hermitian platforms. We cover the core theoretical principles -- from Berry curvature and symmetry-protected topological invariants to bulk-boundary correspondence -- alongside experimental advances in probing topological phonon states via inelastic scattering and momentum-resolved techniques for solid-state phonons as well as pump-probe measurements in acoustic/mechanical metamaterials. Key topics include Weyl/Dirac/nodal-line phonons in crystalline solids, symmetry-engineered topological phases in metamaterials, non-Hermitian effects (exceptional points, skin effect), and emergent directions such as Floquet engineering, synthetic dimensions, and real-space topological textures (skyrmions, merons). We also highlight technological applications in robust waveguides, on-chip surface-acoustic-wave devices, and acoustofluidics, while outlining future challenges and opportunities in quantum phononics, nonlinear topological phenomena, and interdisciplinary integration with photonics and electronics. This review serves as a comprehensive guide across physics, materials science, and engineering, bridging fundamental theory with cutting-edge experiments and innovations in topological phononics.

cond-mat.mtrl-sci↗
arXiv2025

Hunyuan-TurboS: Advancing Large Language Models through Mamba-Transformer Synergy and Adaptive Chain-of-Thought

As Large Language Models (LLMs) rapidly advance, we introduce Hunyuan-TurboS, a novel large hybrid Transformer-Mamba Mixture of Experts (MoE) model. It synergistically combines Mamba's long-sequence processing efficiency with Transformer's superior contextual understanding. Hunyuan-TurboS features an adaptive long-short chain-of-thought (CoT) mechanism, dynamically switching between rapid responses for simple queries and deep "thinking" modes for complex problems, optimizing computational resources. Architecturally, this 56B activated (560B total) parameter model employs 128 layers (Mamba2, Attention, FFN) with an innovative AMF/MF block pattern. Faster Mamba2 ensures linear complexity, Grouped-Query Attention minimizes KV cache, and FFNs use an MoE structure. Pre-trained on 16T high-quality tokens, it supports a 256K context length and is the first industry-deployed large-scale Mamba model. Our comprehensive post-training strategy enhances capabilities via Supervised Fine-Tuning (3M instructions), a novel Adaptive Long-short CoT Fusion method, Multi-round Deliberation Learning for iterative improvement, and a two-stage Large-scale Reinforcement Learning process targeting STEM and general instruction-following. Evaluations show strong performance: overall top 7 rank on LMSYS Chatbot Arena with a score of 1356, outperforming leading models like Gemini-2.0-Flash-001 (1352) and o4-mini-2025-04-16 (1345). TurboS also achieves an average of 77.9% across 23 automated benchmarks. Hunyuan-TurboS balances high performance and efficiency, offering substantial capabilities at lower inference costs than many reasoning models, establishing a new paradigm for efficient large-scale pre-trained models.

cs.CL↗
arXiv2024

Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters by Tencent

In this paper, we introduce Hunyuan-Large, which is currently the largest open-source Transformer-based mixture of experts model, with a total of 389 billion parameters and 52 billion activation parameters, capable of handling up to 256K tokens. We conduct a thorough evaluation of Hunyuan-Large's superior performance across various benchmarks including language understanding and generation, logical reasoning, mathematical problem-solving, coding, long-context, and aggregated tasks, where it outperforms LLama3.1-70B and exhibits comparable performance when compared to the significantly larger LLama3.1-405B model. Key practice of Hunyuan-Large include large-scale synthetic data that is orders larger than in previous literature, a mixed expert routing strategy, a key-value cache compression technique, and an expert-specific learning rate strategy. Additionally, we also investigate the scaling laws and learning rate schedule of mixture of experts models, providing valuable insights and guidances for future model development and optimization. The code and checkpoints of Hunyuan-Large are released to facilitate future innovations and applications. Codes: https://github.com/Tencent/Hunyuan-Large Models: https://huggingface.co/tencent/Tencent-Hunyuan-Large

cs.CL↗
arXiv2023

Disentangled higher-orbital bands and chiral symmetric topology in confined Mie resonance photonic crystals

Topological phases based on tight-binding models have been extensively studied in recent decades. By mimicking the linear combination of atomic orbitals in tight-binding models based on the evanescent couplings between resonators in classical waves, numerous experimental demonstrations of topological phases have been successfully conducted. However, in dielectric photonic crystals, the Mie resonances' states decay too slowly as $1/r$ when $r$ $\to$ $\infty$, leading to intrinsically different physical properties between tight-binding models and dielectric photonic crystals. Here, we propose a confined Mie resonance photonic crystal by embedding perfect electric conductors in between dielectric rods, leading to a perfectly matched band structure as the tight-binding models with nearest-neighbour couplings. As a consequence, disentangled band structure spanned by higher atomic orbitals is observed. Moreover, we also achieve a three-dimensional photonic crystal with a complete photonic bandgap and third-order topology based on our design. Our implementation provides a versatile platform for studying exotic higher-orbital bands and achieving tight-binding-like 3D topological photonic crystals.

cond-mat.mes-hall↗
arXiv2022

Bulk-LDOS Correspondence in Topological Insulators

Seeking the criterion for diagnosing topological phases in real materials has been one of the major tasks in topological physics. Currently, bulk-boundary correspondence based on spectral measurements of in gap topological boundary states and the fractional corner anomaly derived from the measurement of the fractional spectral charge are two main approaches to characterize topologically insulating phases. However, these two methods require a complete band-gap with either in-gap states or strict spatial symmetry of the overall sample which significantly limits their applications to more generalized cases. Here we propose and demonstrate an approach to link the non-trivial hierarchical bulk topology to the multidimensional partition of local-density of states (LDOS) respectively, denoted as the bulk-LDOS correspondence. Specifically, in a finite-size topologically nontrivial photonic crystal, we observe that the distribution of LDOS is divided into three partitioned regions of the sample - the two-dimensional interior bulk area (avoiding edge and corner areas), one-dimensional edge region (avoiding the corner area), and zero-dimensional corner sites. In contrast, the LDOS is distributed across the entire two-dimensional bulk area across the whole spectrum for the topologically trivial cases. Moreover, we present the universality of this criterion by validating this correspondence in both a higher-order topological insulator without a complete band gap and with disorders. Our findings provide a general way to distinguish topological insulators and unveil the unexplored features of topological directional band-gap materials without in-gap states.

cond-mat.mes-hall↗
arXiv2021

Classical higher-order topological insulators

Topological states nurtures the emergence of devices with unprecedented functions in photonics, plasmonics, acoustics and phononics. As one of the recently discovered members, higher-order topological insulators (HOTIs) have been increasingly explored, featuring lower-dimensional topological boundary states, leading to rich mechanisms for topological manipulation, guiding and trapping of classical waves. Here, we provide an overview of current developments of HOTIs in classical waves including basic principles, unique physical properties, various experimental realizations, novel phenomena and potential applications. Based on these discussions, we remark on the trends and challenges in this field and the impacts of higher-order topology on other research fields.

cond-mat.mtrl-sci↗
arXiv2017

Review of Spin Orbit Coupled Semimetal SrIrO3 in thin film form

Spin orbit coupling provides a mechanism to lock the momentum of electron to its spin degree, recent years was revealed to be essential in arousing many novel physical behaviors. SrIrO3 is a typical metallic member of the strong spin orbit coupling iridate family. Its orthorhombic phase was confirmed as a particular spin orbit coupling assistant electron correlated semimetal with small electron and hole pockets, and was supposed to host versatile topological phases, with prospect of opening a new topological matter field based on oxides. The existing experiments have demonstrated that orthorhombic SrIrO3 can be easily synthesized at two dimensional scale films under the substrate lattice constraint, and the films display Fermi-liquid behavior in high temperature and generally two dimensional weak localization resulted metal insulator transition. The properties of orthorhombic SrIrO3 film are sensitive to the rotation and tilting angle, as well as the interlayer coupling of the IrO6 octahedras, consequently can be tuned through substrate strain engineering and size scale. For example, the film was approached a state similar to Sr2IrO4 at the ultrathin limit to several unit cell, becoming a canted antiferromagnetic semiconductor/insulator. The existing knowledges suggest urgent demands of researches on the superlattices constructed with orthorhombic SrIrO3, for further understanding the evolution mechanism of the electron structure, and so the relevant magnetic state and topological phases in the orthorhombic SrIrO3 and its family.

cond-mat.mtrl-sci↗
arXiv2016

Chiral Ordering Spin Associated Glass like State in SrRuO3SrIrO3 Superlattice

Heterostructure interface provides a powerful platform to observe rich emergent phenomena, such as interfacial superconductivity, nontrivial topological surface state. Here SrRuO3/SrIrO3 superlattices were epitaxially synthesized. The magnetic and electrical properties of these superlattices were characterized. Broad cusps in the zero field cooling magnetization curves and near stable residual magnetization below the broad cusps, as well as two steps magnetization hysteresis loops are observed. The magnetization relaxes following a modified Stretched function model indicating coexistence of spin glass and ferromagnetic ordering in the superlattices. Topological Hall effect was demonstrated at low temperature and weakened with the increase of SrIrO3 layer thickness. These results suggest that chiral ordering spin texture were generated at the interfaces due to the interfacial Dzyaloshinskii-Moriya (DM) interaction, which generates the spin glass behaviors. The present work demonstrates that SrIrO3 can effectively induce interface DM interactions in heterostructures, it would pave light on the new research directions of strong spin orbit interaction oxides, from the viewpoints of both basic science and prospective spintronics devices applications.

cond-mat.mtrl-sci↗
arXiv2016

Broadband Photovoltaic Detectors based on an Atomically Thin Heterostructure

Van der Waals junctions of two-dimensional materials with an atomically sharp interface open up unprecedented opportunities to design and study functional heterostructures. Semiconducting transition metal dichalcogenides have shown tremendous potential for future applications due to their unique electronic properties and strong light-matter interaction. However, many important optoelectronic applications, such as broadband photodetection, are severely hindered by their limited spectral range and reduced light absorption. Here, we present a p-g-n heterostructure formed by sandwiching graphene with a gapless bandstructure and wide absorption spectrum in an atomically thin p-n junction to overcome these major limitations. We have successfully demonstrated a MoS2-graphene-WSe2 heterostructure for broadband photodetection in the visible to short-wavelength infrared range at room temperature that exhibits competitive device performance, including a specific detectivity of up to 1011 Jones in the near-infrared region. Our results pave the way toward the implementation of atomically thin heterostructures for broadband and sensitive optoelectronic applications.

cond-mat.mtrl-sci↗