arXiv subjects

Zehao Chen

Publications and source records attributed to Zehao Chen.

16 recordsLinked to original sources

arXiv2026

WDL-OPD: Weak-Driven On-Policy Distillation via Mixture-Constrained Co-Training

On-policy distillation (OPD) aligns a student with a teacher on trajectories sampled from the student itself, reducing the train-test state mismatch of offline distillation. The same feedback loop can nevertheless be unstable: each update changes both the policy and the states on which the next update is computed. We introduce WDL-OPD, a mixture-constrained co-training method with two trainable policies. An anchor policy generates every rollout, an auxiliary policy evaluates the same visited states, and a geometric mixture of their token distributions is matched to a frozen teacher by reverse KL. Both policies receive gradient. We show that freezing the auxiliary recovers an anchor-plus-contrast proxy target closely related to OPD$^2$ and W2S-OPD, whereas joint training creates branch-level degrees of freedom that a static delta cannot express. In recorded Qwen3 experiments at 1.7B and 4B scale, WDL-OPD produces the strongest student checkpoint in each of four scale-domain settings. It raises MATH500 accuracy from 0.630 to 0.685 at 4B and from 0.521 to 0.585 at 1.7B. In code generation, seven single-policy OPD configurations exhibit entropy growth or trajectory degradation, while co-training reaches independently re-evaluated development scores of 0.637 and 0.375. Because several comparisons differ in curriculum or initialization, these results support a stabilization hypothesis rather than a universal causal claim. We provide the exact training algorithm, failure evidence, and the controlled comparison matrix needed to test that hypothesis.

cs.LG↗

arXiv2026

RAVEN: A Regime-Aware Variable-context Expert Network for Financial Time Series Forecasting

Financial time series forecasting presents structural challenges absent from standard benchmarks. Log-returns are non-stationary, exhibit exceptionally low signal-to-noise (SNR) ratios, and are governed by regime-dependent temporal dependencies. We identify a key limitation of state-of-the-art (SOTA) time series models in financial settings. A fixed context window is mismatched to the time-varying optimal look-back of non-stationary price processes. We propose the Regime-Aware Variable-context Expert Network (RAVEN), a Mixture-of-Experts framework designed to adaptively determine the temporal context for each input sample. Instead of relying on a fixed look-back horizon, RAVEN constructs a hierarchy of nested contiguous windows whose lengths are determined by the data itself. Specifically, RAVEN scores patches by learned importance in reverse chronological order and applies the Cumulative Importance Thresholding (CIT) mechanism to derive nested prefix windows, each routed to a scale-specialized expert. A Global Compressed Representation (GCR) branch runs in parallel over the full context, preserving global temporal coherence that local experts cannot guarantee. Because the nested routing induces structured overlap among expert inputs, we introduce a Correlation-Aware Weighting (CAW) to align variable-length expert outputs and penalize pairwise cosine similarity prior to aggregation. Experiments on cumulative log-return prediction (HS300, S&P500) and fund sales forecasting demonstrate that RAVEN achieves SOTA performances, improves Pearson correlation by 9.2% on HS300 and 20.2% on S&P500, and reduces MSE by 18.2% on fund sales forecasting, while achieving the best results in 14 of 16 metrics on four PEMS traffic benchmarks.

cs.LG↗

arXiv2026

LASAR: Latent Adaptive Semantic Aligned Reasoning for Generative Recommendation

Large Language Models (LLMs) have demonstrated powerful reasoning capabilities through Chain-of-Thought (CoT) in various tasks, yet the inefficiency of token-by-token generation hinders real-world deployment in latency-sensitive recommender systems. Latent reasoning has emerged as an effective paradigm in LLMs, performing multi-step inference in a continuous hidden-state space to achieve stronger reasoning at lower cost. However, this paradigm remains underexplored in mainstream generative recommendation. Achieving this reveals three key challenges: (1) the gap between prior-less Semantic ID (SID) symbols and continuous latent reasoning, as SIDs lack pre-trained semantics, hindering joint optimization; (2) representation drift due to a lack of reasoning chain supervision; and (3) the suboptimality of applying a globally fixed reasoning depth. To address these, we propose LASAR (Latent Adaptive Semantic Aligned Reasoning), an SFT-then-RL framework. First, we bridge this gap via two-stage training: Stage 1 grounds SID semantics before Stage 2 introduces latent reasoning, ensuring efficient convergence. Second, we mitigate representation drift through explicit CoT semantic alignment. Step-wise bidirectional KL divergence constrains the latent reasoning trajectory using hidden-state anchors extracted from CoT text, while a Policy Head predicts per-sample reasoning depth. Third, during the GRPO-based RL phase, terminal-only KL alignment accommodates variable-length reasoning, and REINFORCE optimizes the Policy Head to dynamically allocate steps. This nearly halves the average latent step count while simultaneously improving recommendation quality. Experiments on three real-world datasets show that LASAR achieves the best overall performance across the evaluated settings. It adds limited inference latency and is roughly 20x faster than generating explicit CoT text.

cs.IR↗

arXiv2026

Efficient Generation of Neutrons Based on Ultrashort Laser-driven Direct Acceleration in Microwire-Array Targets

We report on an experimental demonstration of efficient neutron generation based on direct laser acceleration in microwire-array targets irradiated by ultrashort (tens of femtoseconds) laser pulses. The optimal array period was identified, at which the maximum proton energy and the number of protons with energies exceeding $1~\mathrm{MeV}$ were significantly increased. Using a $1~\mathrm{PW}$, $\sim25~\mathrm{fs}$ laser at a moderate intensity of $\sim10^{20}~\mathrm{W/cm^2}$, a high neutron yield of up to $(8.33\pm0.84)\times10^{6}~\mathrm{n/sr/J}$ was detected from the LiD converter via $^7\mathrm{Li}(p,n)$ and $\mathrm{D}(p,n+p)$ nuclear reactions. Self-consistent integrated simulations reproduced the experimental results and predicted that with a Be converter, a forward pulsed neutron source with an unprecedented yield per joule of $3.67\times10^{7}~\mathrm{n/sr/J}$ can be obtained under identical laser conditions. This type of neutron source is favorable for applications that require a high repetition rate utilizing compact and economical laser systems.

physics.plasm-ph↗

arXiv2026

Fusian: Multi-LoRA Fusion for Fine-Grained Continuous MBTI Personality Control in Large Language Models

Large Language Models (LLMs) have demonstrated impressive capabilities in simulating diverse human behaviors and personalities. However, existing methods for personality control, which include prompt engineering and standard Supervised Fine-Tuning (SFT), typically treat personality traits as discrete categories (e.g., "Extroverted" vs. "Introverted"), lacking the ability to precisely control the intensity of a trait on a continuous spectrum. In this paper, we introduce Fusian, a novel framework for fine-grained, continuous personality control in LLMs. Fusian operates in two stages: (1) Trajectory Collection, where we capture the dynamic evolution of personality adoption during SFT by saving a sequence of LoRA adapters, effectively mapping the continuous manifold of a trait; and (2) RL-based Dynamic Fusion, where we train a policy network using Reinforcement Learning to dynamically compute mixing weights for these frozen adapters. By sampling from a Dirichlet distribution parameterized by the policy network, Fusian fuses multiple adapters to align the model's output with a specific numerical target intensity. Experiments on the Qwen3-14B model demonstrate that Fusian achieves high precision in personality control, significantly outperforming baseline methods in aligning with user-specified trait intensities.

cs.CL↗

arXiv2026

Rotational anisotropy Raman spectrometer for high-sensitivity crystallographic symmetry analysis

Raman spectroscopy stands as a cornerstone technique for probing collective excitations and emergent quantum phases in solids. While polarization-resolved Raman scattering has been widely used to extract symmetry information of eigenmodes, its conventional geometry suffers from significant limitations: it accesses only a subset of Raman tensor elements, enforces {\pi}-periodic intensity patterns that obscure intrinsic crystalline symmetries, and lacks sensitivity to wavevector-dependent anisotropy. To overcome these constraints, here we introduce rotational-anisotropy Raman spectroscopy (RA-Raman). By measuring scattering intensity during full azimuthal rotation of the optical scattering plane at oblique incidence, this geometry enables complete reconstruction of the Raman tensor and reveals rich rotational anisotropy patterns essential for accessing subtle symmetry information elusive to conventional methods. We developed a prototype instrument to validate this approach experimentally. For centrosymmetric crystals, RA-Raman unambiguously identifies phonon symmetry representations and determines crystallographic axes. In noncentrosymmetric crystals, it resolves directional anisotropy and angular dispersion of phonon-polaritons, enabling quantitative determination of the Faust-Henry coefficient and the separation of deformation-potential and electro-optic scattering contributions. By demonstrating unprecedented symmetry-resolving power in standard benchmark crystals, we establish RA-Raman as a powerful tool with far-reaching potential to discover and characterize symmetry-breaking phases and topological excitations in quantum materials.

cond-mat.mtrl-sci↗

arXiv2026

Weak-Driven Learning: How Weak Agents make Strong Agents Stronger

As post-training optimization becomes central to improving large language models, we observe a persistent saturation bottleneck: once models grow highly confident, further training yields diminishing returns. While existing methods continue to reinforce target predictions, we find that informative supervision signals remain latent in models' own historical weak states. Motivated by this observation, we propose WMSS (Weak Agents Can Make Strong Agents Stronger), a post-training paradigm that leverages weak checkpoints to guide continued optimization. By identifying recoverable learning gaps via entropy dynamics and reinforcing them through compensatory learning, WMSS enables strong agents to improve beyond conventional post-training saturation. Experiments on mathematical reasoning and code generation datasets show that agents trained with our approach achieve effective performance improvements, while incurring zero additional inference cost.

cs.AI↗

arXiv2026

Does Your Reasoning Model Implicitly Know When to Stop Thinking?

Recent advancements in large reasoning models (LRMs) have greatly improved their capabilities on complex reasoning tasks through Long Chains of Thought (CoTs). However, this approach often results in substantial redundancy, impairing computational efficiency and causing significant delays in real-time applications. Recent studies show that longer reasoning chains are frequently uncorrelated with correctness and can even be detrimental to accuracy. In a further in-depth analysis of this phenomenon, we surprisingly uncover and empirically verify that LRMs implicitly know the appropriate time to stop thinking, while this capability is obscured by current sampling paradigms. Motivated by this, we introduce SAGE (Self-Aware Guided Efficient Reasoning), a novel sampling paradigm that unleashes this efficient reasoning potential. Furthermore, integrating SAGE as mixed sampling into group-based reinforcement learning (SAGE-RL) enables SAGE-RL to effectively incorporate SAGE-discovered efficient reasoning patterns into standard pass@1 inference, markedly enhancing both the reasoning accuracy and efficiency of LRMs across multiple challenging mathematical benchmarks.

cs.AI↗

arXiv2025

LLMBoost: Make Large Language Models Stronger with Boosting

Ensemble learning of LLMs has emerged as a promising alternative to enhance performance, but existing approaches typically treat models as black boxes, combining the inputs or final outputs while overlooking the rich internal representations and interactions across models.In this work, we introduce LLMBoost, a novel ensemble fine-tuning framework that breaks this barrier by explicitly leveraging intermediate states of LLMs. Inspired by the boosting paradigm, LLMBoost incorporates three key innovations. First, a cross-model attention mechanism enables successor models to access and fuse hidden states from predecessors, facilitating hierarchical error correction and knowledge transfer. Second, a chain training paradigm progressively fine-tunes connected models with an error-suppression objective, ensuring that each model rectifies the mispredictions of its predecessor with minimal additional computation. Third, a near-parallel inference paradigm design pipelines hidden states across models layer by layer, achieving inference efficiency approaching single-model decoding. We further establish the theoretical foundations of LLMBoost, proving that sequential integration guarantees monotonic improvements under bounded correction assumptions. Extensive experiments on commonsense reasoning and arithmetic reasoning tasks demonstrate that LLMBoost consistently boosts accuracy while reducing inference latency.

cs.LG↗

arXiv2025

StoryBox: Collaborative Multi-Agent Simulation for Hybrid Bottom-Up Long-Form Story Generation Using Large Language Models

Human writers often begin their stories with an overarching mental scene, where they envision the interactions between characters and their environment. Inspired by this creative process, we propose a novel approach to long-form story generation, termed hybrid bottom-up long-form story generation, using multi-agent simulations. In our method, agents interact within a dynamic sandbox environment, where their behaviors and interactions with one another and the environment generate emergent events. These events form the foundation for the story, enabling organic character development and plot progression. Unlike traditional top-down approaches that impose rigid structures, our hybrid bottom-up approach allows for the natural unfolding of events, fostering more spontaneous and engaging storytelling. The system is capable of generating stories exceeding 10,000 words while maintaining coherence and consistency, addressing some of the key challenges faced by current story generation models. We achieve state-of-the-art performance across several metrics. This approach offers a scalable and innovative solution for creating dynamic, immersive long-form stories that evolve organically from agent-driven interactions.

cs.CL↗

arXiv2025

Latent Feature-Guided Conditional Diffusion for Generative Image Semantic Communication

Semantic communication is proposed and expected to improve the efficiency of massive data transmission over sixth generation (6G) networks. However, existing image semantic communication schemes are primarily focused on optimizing pixel-level metrics, while neglecting the crucial aspect of region of interest (ROI) preservation. To address this issue, we propose an ROI-aware latent representation-oriented image semantic communication (LRISC) system. In particular, we first map the source image to latent features in a high-dimensional semantic space, these latent features are then fused with ROI mask through a feature-weighting mechanism. Subsequently, these features are encoded using a joint source and channel coding (JSCC) scheme with adaptive rate for efficient transmission over a wireless channel. At the receiver, a conditional diffusion model is developed by using the received latent features as conditional guidance to steer the reverse diffusion process, progressively reconstructing high-fidelity images while preserving semantic consistency. Moreover, we introduce a channel signal-to-noise ratio (SNR) adaptation mechanism, allowing one model to work across various channel states. Experiments show that the proposed method significantly outperforms existing methods, in terms of learned perceptual image patch similarity (LPIPS) and robustness against channel noise, with an average LPIPS reduction of 43.3% compared to DeepJSCC, while guaranteeing the semantic consistency.

cs.MM↗

arXiv2025

Knowledge-Driven Hybrid SSD Management Enhanced by Fine-Tuned LLMs

A hybrid Solid-State Drives (SSDs) integrates different modes of flash cells (e.g., single-level cell (SLC) and Quad-Level Cell (QLC)) and enables them to convert between each other, achieving both high performance and storage capacity. However, this hybrid design introduces a significantly larger design space than traditional SSDs with additional design factors such as flash conversion and data migration across different flash modes, leading to higher optimization complexity. Efficient management of such complexity requires deep hybrid SSD knowledge and dynamic adjustment mechanisms. Large language models (LLMs) offer a promising solution through their contextual reasoning and adaptive coordination capabilities. In this work, we explore the potential of using LLMs in understanding and efficiently managing hybrid SSD design space. We find that leveraging LLMs for knowledge-guided optimization of management parameters enables substantial performance gains. Building on these insights, we propose LLM-hybridSSD, an integrated optimization framework that formulates hybrid SSD management as a parameter-tuning problem, employs an LLM-based tuner for adaptive configuration, and applies reinforcement learning-based fine-tuning to align local lightweight models with domain-specific knowledge. Experimental results show an average 58.92% increase in throughput and a 28.56\% reduction in write amplification (WA) compared with state-of-the-art schemes under different real-world workloads.

cs.AR↗

arXiv2024

SVGBuilder: Component-Based Colored SVG Generation with Text-Guided Autoregressive Transformers

Scalable Vector Graphics (SVG) are essential XML-based formats for versatile graphics, offering resolution independence and scalability. Unlike raster images, SVGs use geometric shapes and support interactivity, animation, and manipulation via CSS and JavaScript. Current SVG generation methods face challenges related to high computational costs and complexity. In contrast, human designers use component-based tools for efficient SVG creation. Inspired by this, SVGBuilder introduces a component-based, autoregressive model for generating high-quality colored SVGs from textual input. It significantly reduces computational overhead and improves efficiency compared to traditional methods. Our model generates SVGs up to 604 times faster than optimization-based approaches. To address the limitations of existing SVG datasets and support our research, we introduce ColorSVG-100K, the first large-scale dataset of colored SVGs, comprising 100,000 graphics. This dataset fills the gap in color information for SVG generation models and enhances diversity in model training. Evaluation against state-of-the-art models demonstrates SVGBuilder's superior performance in practical applications, highlighting its efficiency and quality in generating complex SVG graphics.

cs.CV↗

arXiv2021

Reputation dependent pricing strategy: analysis based on a Chinese C2C marketplace

Most online markets establish reputation systems to assist building trust between sellers and buyers. Sellers' reputations not only provide guidelines for buyers but may also inform sellers their optimal pricing strategy. In this research, we assumed two types of buyer: informed buyers and uninformed buyers. Informed buyers know more about the reputation about the seller but may incur a search cost. Then we developed a benchmark model and a competition model. We found that high reputation sellers and low reputation sellers adapt different pricing strategy depending on the informativeness of buyers and the competition among sellers. With a large proportion of informed buyers, high reputation sellers may charge lower price than low reputation sellers, which exists a negative price premium effect, in contrast to conclusions of some previous studies. Empirical findings were in consistence with our theoretical models. We collected data of five categories of products, televisions, laptops, cosmetics, shoes, and beverages, from Taobao, a leading C2C Chinese online market. Negative price premium effect was observed for TVs, laptops, and cosmetics; price premium effect was observed for beverages; no significant trend was observed for shoes. We infer product value and market complexity are the main factors of buyer informativeness.

econ.GN↗

arXiv2020

Attentive Semantic Exploring for Manipulated Face Detection

Face manipulation methods develop rapidly in recent years, whose potential risk to society accounts for the emerging of researches on detection methods. However, due to the diversity of manipulation methods and the high quality of fake images, detection methods suffer from a lack of generalization ability. To solve the problem, we find that segmenting images into semantic fragments could be effective, as discriminative defects and distortions are closely related to such fragments. Besides, to highlight discriminative regions in fragments and to measure contribution to the final prediction of each fragment is efficient for the improvement of generalization ability. Therefore, we propose a novel manipulated face detection method based on Multilevel Facial Semantic Segmentation and Cascade Attention Mechanism. To evaluate our method, we reconstruct two datasets: GGFI and FFMI, and also collect two open-source datasets. Experiments on four datasets verify the advantages of our approach against other state-of-the-arts, especially its generalization ability.

cs.CV↗

arXiv2011

Mean--variance portfolio optimization when means and covariances are unknown

Markowitz's celebrated mean--variance portfolio optimization theory assumes that the means and covariances of the underlying asset returns are known. In practice, they are unknown and have to be estimated from historical data. Plugging the estimates into the efficient frontier that assumes known parameters has led to portfolios that may perform poorly and have counter-intuitive asset allocation weights; this has been referred to as the "Markowitz optimization enigma." After reviewing different approaches in the literature to address these difficulties, we explain the root cause of the enigma and propose a new approach to resolve it. Not only is the new approach shown to provide substantial improvements over previous methods, but it also allows flexible modeling to incorporate dynamic features and fundamental analysis of the training sample of historical data, as illustrated in simulation and empirical studies.

stat.AP↗