arXiv subjects

Siyuan Zhu

Publications and source records attributed to Siyuan Zhu.

9 recordsLinked to original sources

arXiv2026

From Verdict to Process: Agentic Reinforcement Learning for Multi-Stage Fact Verification

Recent approaches combining Large Language Models (LLMs) with retrieval-augmented reasoning have shown promise for automated fact verification. To process complex claims, these verification pipelines typically execute multi-stage workflows that coordinate tightly coupled modules, including claim decomposition, evidence gathering, and verdict prediction. However, existing methods optimize individual stages in isolation or rely on fixed heuristics, which limits adaptive coordination among stages and can lead to suboptimal outcomes. In this work, we propose ProFact, an agentic reinforcement learning framework for end-to-end optimization of multi-stage fact verification trajectories. ProFact trains a unified policy to coordinate claim decomposition, evidence seeking, answer generation, and verdict prediction. To address the sparse and delayed supervision provided by final veracity labels, ProFact introduces process-aware rewards that provide stage-level learning signals throughout the verification process. Empirical evaluation shows that ProFact consistently outperforms strong baselines in both verification performance and inference efficiency. These results highlight the effectiveness of process-aware trajectory optimization for multi-stage fact verification.

cs.AI↗

arXiv2026

GAGPO: Generalized Advantage Grouped Policy Optimization

Reinforcement learning has become a powerful paradigm for post-training large language model agents, yet credit assignment in multi-turn environments remains a challenge. Agents often receive sparse, trajectory-level rewards only at the end of an episode, making it difficult to determine which intermediate actions contributed to success or failure. As a result, propagating delayed outcomes back to individual decision steps without relying on costly auxiliary value models remains an open problem. We propose Generalized Advantage Grouped Policy Optimization (GAGPO), a critic-free reinforcement learning method for precise, step-aligned temporal credit assignment. GAGPO constructs a non-parametric grouped value proxy from sampled rollouts and uses it to compute TD/GAE-style temporal advantages, recursively propagating outcome supervision backward through time. Combined with group-wise advantage normalization and an action-level importance ratio, GAGPO extracts stable, localized optimization signals directly from multi-turn trajectories. Experiments on ALFWorld and WebShop show that GAGPO outperforms strong reinforcement learning baselines. Further analyses demonstrate faster early-stage learning, improved interaction efficiency, and smoother optimization dynamics, suggesting that GAGPO offers a simple yet effective framework for multi-turn agentic reinforcement learning.

cs.CL↗

arXiv2026

Context-Picker: Dynamic context selection using multi-stage reinforcement learning

In long-context question answering, selecting the appropriate scope of context for a query remains a key and unresolved challenge. Insufficient context can lead to missing essential information, whereas excessive context often introduces noise and degrades answer quality. Conventional methods, such as retrieving a fixed number of passages or applying reranking, struggle to dynamically determine which context to include. This is especially problematic for factoid questions, which typically depend only on a few precise pieces of evidence. To overcome this limitation, we propose Context-Picker, a reasoning-aware framework that reframes context selection as the task of identifying a minimal sufficient evidence subset, moving beyond conventional similarity-based ranking. Context-Picker uses a human-inspired two-stage reinforcement learning schedule: stage 1 focuses on improving the recall rate of critical passages, and stage 2 prioritizes pruning redundancy to distill a compact evidence set. To resolve reward sparsity, we propose an offline evidence distillation pipeline that mines ``minimal sufficient sets" via a Leave-One-Out (LOO) procedure, providing dense and task-aligned supervision. Experiments on five long-context and multi-hop QA datasets demonstrate that our method outperforms strong RAG baselines and achieved higher answer accuracy. Ablation studies also indicate that our coarse-to-fine optimization schedule, the redundancy-aware reward shaping, along with the rationale generated by the policy, all contribute substantially to these gains.

cs.AI↗

arXiv2025

Ultrafast dynamics of three-dimensional Kane plasmons in the narrow-bandgap Hg$_{0.8}$Cd$_{0.2}$Te

We report on an ultrafast terahertz spectroscopic study on the dynamics of free carriers and the pertinent bulk plasmons in Hg$_{0.8}$Cd$_{0.2}$Te (MCT) film, a narrowband semiconductor accommodating three dimensional massless Kane fermions. The ultrabroadband terahertz source enables the investigation of the lightly doped equilibrium state in the presence of plasmon-phonon hybridization through the heavily doped excited state, primarily dominated by plasmons. Without the recourse to the resource consuming cryogenic high magnetic field spectroscopy that hinges on observable related to the interband transition, we show that the massless band dispersion can instead be conveniently perceived by the room temperature study of the intraband transition through the determination of the plasmon carrier density relationship. We found the plasma frequency in MCT scales with the cube root of carrier density, in contrast with the square root scaling in the conventional massive fermion system of parabolic band dispersion. This work also answers the curious question of whether the MCT can maintain its massless Kane fermion character in case the strict gapless condition is deviated from. The method presented herein provides a convenient approach to identifying the landscape of both massless and massive band dispersion.

cond-mat.mtrl-sci↗

arXiv2023

Ultraviolet Resonant Nanogap Antennas with Rhodium Nanocube Dimers for Enhancing Protein Intrinsic Autofluorescence

Plasmonic optical nanoantennas offer compelling solutions for enhancing light-matter interactions at the nanoscale. However, until now, their focus has been mainly limited to the visible and near-infrared regions, overlooking the immense potential of the ultraviolet (UV) range, where molecules exhibit their strongest absorption. Here, we present the realization of UV resonant nanogap antennas constructed from paired rhodium nanocubes. Rhodium emerges as a robust alternative to aluminum, offering enhanced stability in wet environments and ensuring reliable performance in the UV range. Our results showcase the nanoantenna ability to enhance the UV autofluorescence of label-free streptavidin and hemoglobin proteins. We achieve significant enhancements of the autofluorescence brightness per protein by up to 120-fold, and reach zeptoliter detection volumes enabling UV autofluorescence correlation spectroscopy (UV-FCS) at high concentrations of several tens of micromolar. We investigate the modulation of fluorescence photokinetic rates and report excellent agreement between experimental results and numerical simulations. This work expands the applicability of plasmonic nanoantennas into the deep UV range, unlocking the investigation of label-free proteins at physiological concentrations.

physics.optics↗

arXiv2022

Pointillism: Accurate 3D bounding box estimation with multi-radars

Autonomous perception requires high-quality environment sensing in the form of 3D bounding boxes of dynamic objects. The primary sensors used in automotive systems are light-based cameras and LiDARs. However, they are known to fail in adverse weather conditions. Radars can potentially solve this problem as they are barely affected by adverse weather conditions. However, specular reflections of wireless signals cause poor performance of radar point clouds. We introduce Pointillism, a system that combines data from multiple spatially separated radars with an optimal separation to mitigate these problems. We introduce a novel concept of Cross Potential Point Clouds, which uses the spatial diversity induced by multiple radars and solves the problem of noise and sparsity in radar point clouds. Furthermore, we present the design of RP-net, a novel deep learning architecture, designed explicitly for radar's sparse data distribution, to enable accurate 3D bounding box estimation. The spatial techniques designed and proposed in this paper are fundamental to radars point cloud distribution and would benefit other radar sensing applications.

cs.CV↗

arXiv2021

Joint Distribution across Representation Space for Out-of-Distribution Detection

Deep neural networks (DNNs) have become a key part of many modern software applications. After training and validating, the DNN is deployed as an irrevocable component and applied in real-world scenarios. Although most DNNs are built meticulously with huge volumes of training data, data in the real world still remain unknown to the DNN model, which leads to the crucial requirement of runtime out-of-distribution (OOD) detection. However, many existing approaches 1) need OOD data for classifier training or parameter tuning, or 2) simply combine the scores of each hidden layer as an ensemble of features for OOD detection. In this paper, we present a novel outlook on in-distribution data in a generative manner, which takes their latent features generated from each hidden layer as a joint distribution across representation spaces. Since only the in-distribution latent features are comprehensively understood in representation space, the internal difference between in-distribution and OOD data can be naturally revealed without the intervention of any OOD data. Specifically, We construct a generative model, called Latent Sequential Gaussian Mixture (LSGM), to depict how the in-distribution latent features are generated in terms of the trace of DNN inference across representation spaces. We first construct the Gaussian Mixture Model (GMM) based on in-distribution latent features for each hidden layer, and then connect GMMs via the transition probabilities of the inference traces. Experimental evaluations on popular benchmark OOD datasets and models validate the superiority of the proposed method over the state-of-the-art methods in OOD detection.

cs.LG↗

arXiv2018

Prolonged photo-carriers generated in a massive-and-anisotropic Dirac material

Transient electron-hole pairs generated in semiconductors can exhibit unconventional excitonic condensation. Anisotropy in the carrier mass is considered as the key to elongate the life time of the pairs, and hence to stabilize the condensation. Here we employ time- and angle-resolved photoemission spectroscopy to explore the dynamics of photo-generated carriers in black phosphorus. The electronic structure above the Fermi level has been successfully observed, and a massive-and-anisotropic Dirac-type dispersions are confirmed; more importantly, we directly observe that the photo-carriers generated across the direct band gap have the life time exceeding 400 ps. Our finding confirms that black phosphorus is a suitable platform for excitonic condensations, and also open an avenue for future applications in broadband mid-infrared BP-based optoelectronic devices.

cond-mat.mtrl-sci↗

arXiv2015

Ultrafast electron dynamics at the Dirac node of the topological insulator Sb$_2$Te$_3$

Topological insulators (TIs) are a new quantum state of matter. Their surfaces and interfaces act as a topological boundary to generate massless Dirac fermions with spin-helical textures. Investigation of fermion dynamics near the Dirac point is crucial for the future development of spintronic devices incorporating topological insulators. However, research so far has been unsatisfactory because of a substantial overlap with the bulk valence band and a lack of a completely unoccupied Dirac point (DP). Here, we explore the surface Dirac fermion dynamics in the TI Sb$_2$Te$_3$ by time- and angle-resolved photoemission spectroscopy (TrARPES). Sb$_2$Te$_3$ has a DP located completely above the Fermi energy ($E_F$) with an in-gap DP. The excited electrons in the upper Dirac cone stay longer than those below the Dirac point to form an inverted population. This was attributed to a reduced density of states (DOS) near the DP .

cond-mat.mtrl-sci↗