arXiv subjects

Yao Chen

Publications and source records attributed to Yao Chen.

At least 19 recordsLinked to original sources

arXiv2026

Agent-R$^2$: Scaling Agents for Realistic Productivity Work via Real-Source Data Synthesis

Large language models have made rapid progress in agentic reasoning and tool use, raising expectations for agents that can take over realistic productivity work from humans. However, current agents still struggle to finish such tasks, where success requires following long-horizon professional workflows, understanding heterogeneous file contexts, and producing usable deliverables under fine-grained requirements. A key bottleneck is the lack of scalable post-training data that reflects real work intent, authentic file environments, and deliverable-level evaluation. To close this gap, we introduce Agent-R$^2$, a $\textbf{r}$eal-source data synthesis framework that scales training data for $\textbf{r}$ealistic productivity work. Instead of relying on manually specified workflows or LLM-generated files, Agent-R$^2$ grounds synthesis throughout the pipeline in real occupational seeds, corpus-derived workflow evidence, and authentic web-sourced files. From 10k web-corpus candidates, Agent-R$^2$ distills 4k task instances that pair real multi-format attachments with fine-grained, deliverable-level rubrics. On this data, we train Agent-R$^2$-4B via supervised fine-tuning and rubric-based reinforcement learning, improving both end-to-end productivity performance and file-grounded reasoning. Trained on only 4k instances, Agent-R$^2$-4B surpasses Qwen3-235B-A22B-Thinking-2507 on both GDPval (51.0$\%$ vs. 48.0$\%$) and OfficeQA (14.8$\%$ vs. 13.8$\%$). Further analysis shows that Agent-R$^2$ mitigates shortcut behavior, where agents produce superficially plausible deliverables before sufficiently grounding, constructing, or verifying them, while promoting sustained exploration.

cs.CL↗

arXiv2026

Fast Functional Clustering via Closed-Form Covariance Updates

Functional clustering identifies coordinated molecular profiles, but repeated numerical covariance optimization can make mixture-model fitting computationally demanding. We develop a fast functional clustering procedure for joint analysis across conditions, combining linear mean profiles on a standardized log niche coordinate with first-order structured antedependence (SAD(1)) covariance. Weighted residual statistics yield closed-form conditional covariance updates under four parameter-sharing specifications, while an innovation representation simplifies likelihood evaluation. In same-R comparisons with numerical optimization of identical conditional objectives, the closed-form solver was 91.5--201.4 times as fast, with objective agreement in all 60 targets. Under matched complete mixture fits, it achieved a 1.87--2.37-fold speedup with identical cluster assignments. Recovery simulations showed higher module classification accuracy and lower mean-curve error than $K$-means and a Gaussian mixture with shared full covariance when the mean structure was correctly specified. SAD(1) provided an additional recovery gain under strong residual correlation, whereas curved-mean controls identified the importance of an appropriate mean family. Applications to single-condition gut microbiota, paired leaf--root microbiota and three-visit human gut multi-omics illustrate the framework across different numbers of conditions. Closed-form covariance estimation makes joint functional clustering substantially faster while retaining the fitted model's likelihood and assignments.

stat.CO↗

arXiv2026

PIMCQG: Co-Designing Billion-Scale Graph-Based Approximate Nearest Neighbor Search for Processing-in-Memory

Graph-based Approximate Nearest Neighbor Search (ANNS) has emerged as the dominant approach for large-scale AI applications, offering superior accuracy-efficiency trade-offs over alternative methods. However, graph-based ANNS is inherently memory-bound due to frequent and irregular graph traversal. This constrains CPU throughput due to limited memory bandwidth, while GPUs lack sufficient HBM capacity for billion-scale indexes. Processing-in-Memory (PIM) offers massive internal bandwidth by co-locating computation with data, making it a natural fit for irregular, memory-bound workloads. However, directly mapping graph-based ANNS onto PIM exposes severe algorithm-architecture mismatches: restricted per-PU local memory, costly inter-PU communication, heavy coordination overhead, and weak compute capability, forcing existing PIM-based ANNS to rely on cluster-based algorithms that plateau at low recall. We propose PIMCQG, an algorithm-hardware co-design framework through three optimizations: (1) a compact index reducing PIM-resident footprint by 14.5x; (2) an asynchronous pipelined scheduler fully utilizing host-PIM bandwidth; and (3) a multiplication-free distance kernel with less than 0.08% recall loss. Evaluations on three billion-scale datasets show up to 20x/17.1x throughput over CPU/GPU baselines, 129x over prior PIM accelerators at high recall, and scale well across multi-node and emerging PIM architectures.

cs.AR↗

arXiv2026

AlphaRLM: Aligning Language Models for Literature-Grounded Symbolic Factor Discovery

Formulaic alpha mining seeks to discover executable expressions that capture predictive market signals in a vast, noisy search space. Recent LLM-based approaches use prompting, iterative search, or seed-factor evolution, leaving direct factor construction through policy learning less explored. To address this, we introduce AlphaRLM, a reinforcement-learning framework that trains LLMs to construct executable symbolic factors from financial literature without input factor expressions. AlphaRLM incorporates Literature-Guided Operator Assembly (LEGO), which conditions factor construction on literature-derived ideas and executable operators, providing semantic priors for exploration. It further aligns the generation policy with market feedback through executable-validity and grouped-return-separation rewards that capture non-monotonic structures across factor quantiles. Extensive experiments on CSI 300, CSI 1000, and S&P 500 show that AlphaRLM outperforms strong factor-mining and LLM-based baselines in annualized excess return. Notably, the 8B model achieves strong cross-market performance, highlighting the promise of literature-conditioned, market-aligned factor generation.

cs.AI↗

arXiv2026

Detection of fine-structured radio bursts during a superflare on EQ Peg

Superflares are frequently observed on magnetically active stars, particularly on active M-dwarfs, thereby influencing the habitability of nearby exoplanets. However, the causes of superflares remain unclear. Here, combining FAST and TESS, we detected a finely structured decimetric radio burst appearing as groups of spikes during a white-light superflare (total bolometric energy $\sim 10^{33} \ \mathrm{erg}$) on the M-dwarf binary EQ Peg on September 8th, 2022. The most significant ones appeared near the peak time of the optical flare and lasted for about $100 \ \mathrm{s}$. The radio spikes present extremely high brightness temperatures ($\gtrsim 10^{13} \ \mathrm{K}$) and high degrees of circular polarization ($\sim 0.84 \pm 0.16$), indicating the origin of a coherent emission process, most likely the electron cyclotron maser. Based on Zeeman-Doppler Imaging measurements of EQ Peg, we further locate the sources of the radio spikes, including their heights and latitudes, under the assumption of a large-scale dipolar magnetic field configuration. The results provide strong evidence for finely structured decimetric radio bursts generated by stellar flares and shed light on the origin of superflares.

astro-ph.SR↗

arXiv2026

HLS-Seek: QoR-Aware Code Generation for High-Level Synthesis via Proxy Comparative Reward Reinforcement Learning

High-Level Synthesis (HLS) compiles algorithmic C/C++ descriptions into hardware, with Quality of Results (QoR)---latency and resource utilization---critically governed by pragma configurations and code structure. Existing natural-language-to-HLS (NL-to-HLS) training approaches prioritize functional correctness while largely ignoring QoR. We observe that reinforcement learning (RL) for HLS does not require absolute synthesis results---only relative comparisons between candidates. Based on this insight, we propose \textbf{HLS-Seek}, a QoR-aware NL-to-HLS framework that avoids full synthesis-in-the-loop RL via a comparative proxy reward model achieving 99.53\% Pareto-dominance accuracy. To prevent reward hacking, we introduce \textit{uncertainty-aware Monte Carlo (MC) dropout switching} that selectively invokes real Vitis HLS synthesis for low-confidence candidates and online updates the proxy, creating a self-improving reward system. HLS-Seek achieves 84.7\% syntax correctness pass@1 and 81.4\% functional correctness pass@5 on HLS-Eval~\cite{abikaram2025hlseval} with only 7B parameters, surpassing GPT-5.1 on functional pass@5, while achieving 8.5$\times$ faster training than real-reward RL. On QoR evaluation, HLS-Seek achieves the lowest latency on 19/30 kernels and Pareto-dominates HLS-specific baselines on 9 kernels.

cs.LG↗

arXiv2026

Memory Augmentation Unlocks Efficient Chain-of-Thought Reasoning

Large language models often rely on Chain-of-Thought (CoT) reasoning to solve complex tasks, but verbose reasoning traces introduce substantial inference overhead. CoT compression shortens generation, yet aggressive compression may disrupt logical coherence and degrade performance. We formalize this trade-off as the Context-Generation Substitution Law, where explicit reasoning context substitutes for part of decode-time generation. Based on this principle, we propose Memory-Augmented Compression, a training-free framework that constructs reusable reasoning memories from historical traces and retrieves them as prefill-side scaffolds. Rather than using raw demonstrations, these memories summarize reusable reasoning patterns, key constraints, and critical operations to compensate for information lost during compression. Experiments show that Memory consistently improves prompt-based Chain-of-Draft (CoD) compression across mathematical reasoning, complex reasoning, and science question answering tasks, yielding accuracy gains of 21.4, 28.0, 29.5, and 6.61 points over CoD on GSM8K, MATH, BBH, and MMLU-Sci, while achieving a 1.14-1.49x latency speedup latency speedup over standard CoT. Memory is also compatible with token-level, reasoning-trace-level, and inference-state compression mechanisms.

cs.CL↗

arXiv2026

Empirical Simulation of Survival and Mixed-Type Data for Clinical Trial Design

Simulating realistic time-to-event data is essential for planning and evaluating complex clinical trial designs. Conventional approaches often sample event times from parametric families, such as Weibull or log-normal distributions, which restrict hazard shapes and may poorly represent observed survival data. We propose an empirical copula-based framework for simulating multivariate data containing continuous, binary, count, and right-censored time-to-event variables. The method completes censored historical survival data using a two-zone procedure that combines conditional Kaplan-Meier imputation with a parametric tail. It matches a target survival distribution through a log-scale location-scale transformation and a power distortion of the empirical percentile function, while preserving historical dependence through a Gaussian copula fitted to rank correlations. In an oncology trial of previously treated non-small-cell lung cancer, the method reconstructs overall survival and progression-free survival curves for the experimental arm using control-arm data and a small set of target percentiles. Simulations preserve rank correlations among baseline covariates and the dependence between progression-free and overall survival, with censored Kendall's tau of 0.522 compared with 0.549 in the observed data. The method is implemented in the R package EmpiricalSim.

stat.ME↗

arXiv2026

Alfvénic Motions in a Stratified Open Flux Tube: Transition from Propagating to Locally Standing Motions and Implications for the Kelvin-Helmholtz Instability

Standing transverse waves in closed coronal structures have been widely studied as a possible route to energy dissipation, with resonant absorption transferring kink wave energy to localized Alfvénic motions and the Kelvin-Helmholtz instability (KHI) accelerating the formation of small dissipative scales. However, it remains unclear whether the same mechanism applies to the open corona, given the long-standing consensus that the KHI tends to be prohibited for propagating Alfvénic waves. Within the framework of magnetohydrodynamics (MHD), we perform three-dimensional MHD simulations of boundary-driven kink waves in a gravitationally stratified open flux tube extending from the chromosphere into the corona. We find that propagating waves in open magnetic structures can also drive the system toward a turbulent state, with KH vortices clearly identifiable across the flux tube. This occurs because resonant absorption transfers energy from the propagating kink waves to azimuthal Alfvénic motions near the tube boundary, and wave reflection off the gradient of the Alfvén speed subsequently enables these boundary motions to acquire a locally standing character. Our results provide a possible answer to the long-standing question of whether and how propagating waves in open magnetic structures can generate nonlinear turbulent fine structures despite their globally propagating nature.

astro-ph.SR↗

arXiv2026

CW-Ghost: Search-Free Granularity Selection for Helper-Thread Prefetching via Capacity Windows

Helper-thread prefetching hides the latency of irregular memory accesses by executing address dependency chains ahead of the main thread. However, its effectiveness depends on the range of future iterations covered by the helper thread. A fixed coverage range cannot consistently accommodate different workloads and processors, whereas exhaustively evaluating candidate configurations incurs substantial configuration cost. This paper presents CW-Ghost, which uses a single offline profiling run to estimate the average demand cache line fill volume generated per target iteration in a target region. CW-Ghost combines this estimate with a cache capacity budget to derive a Capacity Window, which determines the iteration granularity of each prefetch chunk. In addition, bounded chunk-level synchronization limits the number of chunks by which the helper thread may run ahead of the main thread. Across 14 workload instances evaluated on Intel and AMD CPU platforms, CW-Ghost achieves geometric mean speedups of 1.54x and 1.33x, respectively, over the original programs. Compared with Ghost Threading, it improves geometric mean performance by 15.8% and 10.8%, respectively, while achieving more than 99% of the empirically optimal performance within the candidate set on both platforms. These results demonstrate that cache capacity constraints can effectively guide the selection of granularity for helper-thread prefetching.

cs.DC↗

arXiv2026

Dual-Perspective Microwave and Hard X-ray Constraints of Asymmetric Nonthermal Loops in an X-class Flare

We report dual-perspective microwave and HXR observations of an X-class flare on 2024 May 15, taking advantage of the unique geometry of a front-side view from the Earth and a back-side perspective from the Solar Orbiter (SolO). Using spatially resolved imaging spectroscopy from Siberian Radioheliograph (SRH) together with Chashan Broadband Solar millimeter spectrometer (CBSmm) and STIX data, we identify a set of nonthermal flaring loops in an asymmetric magnetic field, with microwave sources located near the loop top and HXR sources associated with the southern footpoint. Compared with HXR, the microwave emission shows an opposite ascending trend, an increasing time lag in time profile, and a distinctive ``SHH'' spectral pattern, which we attribute to energy-dependent trapping and precipitation of energetic electrons in an asymmetric magnetic configuration. Flux pulsations and their spectral and polarization signatures are consistent with intermittent particle acceleration rather than MHD wave modulation. Microwave magnetic diagnostics, corroborated by non-linear force free field (NLFFF) extrapolation, provide key constraints on the three-dimensional magnetic configuration. The dual-perspective flux profile comparison and consistent QPP signatures across wavelengths together support a self-consistent picture of energy-dependent electron trapping, precipitation, and transport in these asymmetric loops.

astro-ph.SR↗

arXiv2026

Kinetic Processes to Radio Burst: First Observational-driven Study in Coronal Loops

Understanding the origin of coherent solar radio bursts requires linking macroscopic coronal structures with the kinetic processes responsible for wave generation. We investigate emissions from slowly positively drifting bursts (SPDBs), a specific type of solar radio emission. SPDBs provide observational constraints for modeling beam-plasma interactions in coronal loops, serving as a basis for a multiscale description. We employ a three--stage numerical framework that combines nonlinear force--free field (NLFFF) magnetic extrapolation, guiding--center simulations, and fully kinetic particle--in--cell (PIC) modeling. The background plasma density is described using a hydrostatic model consistent with active--region conditions, producing a plasma--frequency gradient comparable to that inferred from the observed SPDBs spectrum. Energetic electrons injected near the loop top evolve through magnetic mirroring, pitch--angle scattering, turbulence development, and partial precipitation. Evolved velocity distribution functions (EVDFs) are sampled after approximately one bounce period and used in PIC simulations to evaluate emission properties. The results show that the evolved beam distribution of energetic electrons predominantly excites beam--Langmuir waves and fundamental plasma emission along the loop, with the emission intensity gradually decreasing from the loop top toward the footpoint. The modest initial beam velocity of energetic electrons explains the inefficient generation of harmonic plasma emission. The temporal evolution of the modeled emission reproduces key SPDB characteristics, including the ~ 4s duration and frequency drift behavior. These results suggest plasma emission explains the mechanisms behind SPDB generation and demonstrate the feasibility of a unified model connecting coronal magnetic topology, particle transport, and radio emission.

astro-ph.SR↗

arXiv2026

Advancing axion detection: Photon regeneration in high-sensitivity Penning trap experiments

The axion, a hypothetical particle proposed to solve the strong CP problem and considered a viable candidate for dark matter, has prompted extensive experimental efforts for its detection. This study presents a novel approach combining photon regeneration techniques ("light-shining-through-a-wall") with high-sensitivity Penning trap technologies to enhance the search for axions. Penning traps offer significant advantages, including precise electromagnetic field measurement, strong magnetic fields, and single-particle detection capabilities. By integrating these traps with resonantly enhanced photon regeneration using microwave cavities, our proposed method significantly increases sensitivity to axion-photon couplings. Preliminary calculations demonstrate an unprecedented achievable sensitivity, reaching an axion-photon coupling constant limit of $g_{aγγ} \le 7.10\times 10^{-8}\mathrm{GeV ^{-1}}$ in just one day, specifically targeting axion energies below 1 MHz. This experimental setup presents a robust and controlled platform, circumventing astrophysical uncertainties, and represents a substantial advancement in laboratory searches for axions and our understanding of dark matter.

hep-ph↗

arXiv2026

ConSA: Controllable Sparsity in Hybrid Attention via Learnable Allocation

Hybrid architectures combining full attention (FA) and sliding-window attention (SWA) are a promising paradigm for efficient LLM inference. However, existing methods typically rely on hand-crafted rules or simple post-hoc heuristics for FA/SWA allocation and offer limited analysis of the attention behaviors underlying these designs. We propose Controllable Sparsity in Hybrid Attention (ConSA), a framework that learns optimal FA/SWA assignment under a user-specified sparsity target. ConSA employs L0 regularization to learn binary masks selecting between FA and SWA for each attention unit, while an augmented Lagrangian constraint enforces the target sparsity at either layer or KV-head granularity. We evaluate ConSA on two LLMs at the 0.6B and 1.7B scales. Learned allocations consistently outperform rule-based baselines, with KV-head-wise allocation yielding clear gains over layer-wise allocation. The learned patterns place SWA in the bottom layers and concentrate FA into contiguous middle-layer blocks, diverging from evenly interleaved patterns in rule-based methods. This structure persists across model scales, sparsity levels, and allocation granularities, revealing a fine-grained spectrum of intrinsic attention behaviors that underlies the learned allocation.

cs.CL↗

arXiv2026

Approaching Shannon Bound with Lossless LLM Weight Compression

Large language models (LLMs) now scale to trillions of parameters, driving weight storage into the terabyte regime and creating an acute mismatch with GPU memory capacity. Although lossless compression is widely effective in other domains, it remains underutilized in LLM systems. Through a comprehensive entropy study across models from 1.5B to 405B parameters and numeric formats ranging from bf16 to int4 and AWQ/SQ8, we find that LLM weights contain far less intrinsic randomness than their stored bitwidth implies, their effective entropy is 2-10x lower, indicating that up to a 10x footprint reduction is theoretically achievable without altering any weight values. Leveraging this insight, we introduce a tile-level, on-the-fly lossless decompression framework based on Asymmetric Numeral Systems that aligns decoding with the GEMM tiling pattern of GPU inference. Our design achieves bit-rates within 0.01-0.1 bits of the Shannon limit across a wide range of LLM numerical formats, demonstrating that nearly all statistical redundancy is eliminated. Integrated into the SGLang serving framework with multi-GPU support, our approach increases the maximum batch size of Qwen-14B from 47 to 75, improving throughput by up to 1.2x. On Mixtral-176B, the feasible batch size increases from 20 to 95 (4.8x), yielding up to 1.6x throughput improvement. Compared to state-of-the-art lossless compression approaches NeuZip and DFloat11, our design further improves throughput by up to 11x.

cs.AR↗

arXiv2026

AutoNumerics-Zero: Automated Discovery of State-of-the-Art Mathematical Functions

Transcendental functions, such as the exponential, are central to scientific computing, yet they cannot be natively calculated by digital hardware. Instead, computers must approximate these functions by combining basic operations, such as $\{+, -, \times, ÷\}$, using methods like Taylor series. These methods were developed over centuries by mathematicians, who focused on approaches that could attain arbitrary accuracy. However, computers can handle most applications by using only finite-precision types, like float32, where any accuracy beyond the type's precision is effectively discarded. We explore, therefore, whether forgoing arbitrary accuracy can lead to the discovery of more efficient approximations. The evolutionary method of symbolic regression is particularly suitable, as it can search for arbitrary operation combinations and can optimize non-differentiable objectives, such as the number of operations used. Our results show that evolution can discover computer programs that outperform established methods in this setting, despite having no prior mathematical knowledge beyond the calculation of the basic operations. Starting from empty code, symbolic regression constructs programs representing novel mathematical expressions. In particular, we discovered a 10-operation program that approximates the exponential function to 14 significant figures, exceeding the accuracy of previously known approximations of this size by more than 6 orders of magnitude.

cs.NE↗

arXiv2026

Structure-Guided Adaptive Propagation for Protein-Protein Interaction Site Prediction

Accurate prediction of protein-protein interaction sites (PPIS) is essential for understanding cellular processes, disease mechanisms, and therapeutic target discovery. Graph-based deep learning has advanced PPIS prediction by incorporating residue-level structural context. However, most graph-based models still rely on fixed propagation schemes that treat all residues similarly, despite the structural and functional heterogeneity of protein interfaces. Such propagation may limit the ability to adapt information diffusion to local geometric environments, making it difficult to distinguish true interaction sites from structurally similar non-interacting neighbors. We present SGAP-PPIS, a structure-guided adaptive propagation model for PPIS prediction. Rather than using a fixed propagation mechanism, SGAP-PPIS leverages multi-scale geometric states from an equivariant graph neural network to generate residue-wise propagation coefficients. This design allows each residue to adaptively balance local feature preservation and neighborhood diffusion according to its geometric microenvironment. Experimental results show that SGAP-PPIS achieves competitive performance among the state-of-the-art methods on Test\_60. Ablation studies show that geometry-conditioned adaptive propagation, scale-aligned geometric guidance, and multi-step propagation-state representation jointly drive these improvements.

cs.AI↗

arXiv2026

MADS: Model-Aware Diverse Core Set Selection for Instruction Tuning

Instruction fine-tuning is employed to enhance the instruction-following ability of large language models (LLMs). As the amount of instruction fine-tuning data increases, selecting the optimal core set becomes particularly important. However, ensuring the diversity of the core set remains a significant challenge. Existing methods predominantly distinguish different training data based on the text features themselves, decoupled from LLMs' own understanding and representation of the data. To address this issue, we propose a Model-Aware Diverse Core Set Selection method, which distinguishes data features based on the neural activation states during LLM inference. This approach serves as an efficient instantiation of coverage-based selection using model-intrinsic activation features to ensure the diversity in the core set. We extensively evaluate our method on six benchmarks that cover five distinct tasks. In our method, the core set selected by the 3B-parameter LLM performs effectively when utilized to fine-tune larger models with 7B, 8B, and 13B parameters. Experimental results on the Alpaca-GPT4 dataset, which comprises 52K instruction-response pairs, show that the core set, sized at 15\% of the original dataset and selected by Llama-3.2-3B-Instruct, achieves an average improvement of 2.5\% when fine-tuning four larger base models compared with training on the full dataset. The experimental results demonstrate that our method enhances model performance on multiple downstream tasks while reducing data requirements.

cs.CL↗