arXiv subjects

Di Wu

Publications and source records attributed to Di Wu.

At least 181 records · Page 10Linked to original sources

arXiv2026

LASER: An Efficient Target-Aware Segmented Attention Framework for End-to-End Long Sequence Modeling

Modeling ultra-long user behavior sequences is pivotal for capturing evolving and lifelong interests in modern recommendation systems. However, deploying such models in real-time industrial environments faces a strict "Latency Wall", constrained by two distinct bottlenecks: the high I/O latency of retrieving massive user histories and the quadratic computational complexity of standard attention mechanisms. To break these bottlenecks, we present LASER, a full-stack optimization framework developed and deployed at Xiaohongshu (RedNote). Our approach tackles the challenges through two complementary innovations: (1) System efficiency: We introduce SeqVault, a unified schema-aware serving infrastructure for long user histories. By implementing a hybrid DRAM-SSD indexing strategy, SeqVault reduces retrieval latency by 50% and CPU usage by 75%, ensuring millisecond-level access to full real-time and life-cycle user histories. (2) Algorithmic efficiency: We propose a Segmented Target Attention (STA) mechanism to address the computational overhead. Motivated by the inherent sparsity of user interests, STA employs a sigmoid-based gating strategy that acts as a silence mechanism to filter out noisy items. Subsequently, a lightweight Global Stacked Target Attention (GSTA) module refines these compressed segments to capture cross-segment dependencies without incurring high computational costs. This design performs effective sequence compression, reducing the complexity of long-sequence modeling while preserving critical signals. Extensive offline evaluations demonstrate that LASER consistently outperforms state-of-the-art baselines. In large-scale online A/B testing serving over 100 million daily active users, LASER achieved a 2.36% lift in ADVV and a 2.08% lift in revenue, demonstrating its scalability and significant commercial impact.

cs.IR↗

arXiv2026

Probing small-scale primordial power spectra with induced gravitational waves

Large-scale primordial perturbations have been well constrained by current cosmological observations, but the properties of small-scale perturbations remain elusive. This study focuses on second-order induced gravitational waves generated by large-amplitude primordial scalar and tensor perturbations on small scales. In this case, the induced gravitational waves include contributions from three types of source terms: scalar-scalar, tensor-scalar, and tensor-tensor. To distinguish them from second-order scalar induced gravitational waves (SIGWs), we refer to those generated by these three source terms as tensor-scalar induced gravitational waves (TSIGWs). We provide the analytical expressions for the kernel functions and the corresponding energy density spectra of second-order TSIGWs. By combining observations of stochastic gravitational wave background (SGWB) across different scales, TSIGWs can be used to constrain small-scale primordial curvature perturbations and primordial gravitational waves. Furthermore, we discuss the feasibility of TSIGWs dominating the current pulsar timing array (PTA) observations under various primordial power spectra scenarios. Our results indicate that TSIGWs generated by monochromatic primordial power spectra might be more likely to dominate the current PTA observations.

astro-ph.CO↗

arXiv2026

Beyond Free-Stream Preservation: Transport Polynomial Exactness for Moving-Mesh Methods under Arbitrary Mesh Motion

High-order moving-mesh methods can effectively reduce numerical diffusion, but their formal accuracy typically relies on the regularity of the mesh velocity. This dependency creates a fundamental conflict in the numerical solution of hyperbolic conservation laws, where solution-driven adaptation may induce nonsmooth mesh motion, thereby degrading convergence order. We introduce \emph{transport polynomial exactness} (TPE($k$)), a mesh-motion-independent criterion that generalizes classical free-stream preservation (TPE(0)) to the exact advection of degree-$k$ polynomials. We show that the classical geometric conservation law (GCL) is insufficient to ensure TPE($k$) for $k \ge 1$ due to mismatches in higher-order geometric moments. To resolve this, we propose \emph{evolved geometric moments} (EGMs), obtained by solving auxiliary transport equations discretized compatibly with the physical variables. We rigorously prove that second-degree EGMs evolved via the third-order strong stability preserving Runge--Kutta (SSPRK3) method coincide with the exact geometric moments. This exactness arises from a \emph{superconvergence} mechanism wherein SSPRK3 reduces to Simpson's rule for EGM evolution. Leveraging this result, we construct a third-order conservative finite-volume rezoning moving-mesh scheme. The scheme satisfies the TPE(2) property for \emph{arbitrary mesh motion} and \emph{any pseudo-time step size}, thereby naturally accommodating spatiotemporally discontinuous mesh velocity. Crucially, this \emph{breaks the efficiency bottleneck} in the conventional advection-based remapping step and reduces the required pseudo-time levels from $\mathcal{O}(h^{-1})$ to $\mathcal{O}(1)$ under bounded but discontinuous mesh velocity. Numerical experiments verify exact quadratic transport and stable third-order convergence under extreme mesh deformation, demonstrating substantial efficiency gains.

math.NA↗

arXiv2026

A benchmark for joint dialogue satisfaction, emotion recognition, and emotion state transition prediction

User satisfaction is closely related to enterprises, as it not only directly reflects users' subjective evaluation of service quality or products, but also affects customer loyalty and long-term business revenue. Monitoring and understanding user emotions during interactions helps predict and improve satisfaction. However, relevant Chinese datasets are limited, and user emotions are dynamic; relying on single-turn dialogue cannot fully track emotional changes across multiple turns, which may affect satisfaction prediction. To address this, we constructed a multi-task, multi-label Chinese dialogue dataset that supports satisfaction recognition, as well as emotion recognition and emotional state transition prediction, providing new resources for studying emotion and satisfaction in dialogue systems.

cs.CL↗

arXiv2026

REArtGS++: Generalizable Articulation Reconstruction with Temporal Geometry Constraint via Planar Gaussian Splatting

Articulated objects are pervasive in daily environments, such as drawers and refrigerators. Towards their part-level surface reconstruction and joint parameter estimation, REArtGS introduces a category-agnostic approach using multi-view RGB images at two different states. However, we observe that REArtGS still struggles with screw-joint or multi-part objects and lacks geometric constraints for unseen states. In this paper, we propose REArtGS++, a novel method towards generalizable articulated object reconstruction with temporal geometry constraint and planar Gaussian splatting. We first model a decoupled screw motion for each joint without type prior, and jointly optimize part-aware Gaussians with joint parameters through part motion blending. To introduce time-continuous geometric constraint for articulated modeling, we encourage Gaussians to be planar and propose a temporally consistent regularization between planar normal and depth through Taylor first-order expansion. Extensive experiments on both synthetic and real-world articulated objects demonstrate our superiority in generalizable part-level surface reconstruction and joint parameter estimation, compared to existing approaches. Project Site: https://sites.google.com/view/reartgs2/home.

cs.CV↗

arXiv2026

Mugi: Value Level Parallelism For Efficient LLMs

Value level parallelism (VLP) has been proposed to improve the efficiency of large-batch, low-precision general matrix multiply (GEMM) between symmetric activations and weights. In transformer based large language models (LLMs), there exist more sophisticated operations beyond activation-weight GEMM. In this paper, we explore how VLP benefits LLMs. First, we generalize VLP for nonlinear approximations, outperforming existing nonlinear approximations in end-to-end LLM accuracy, performance, and efficiency. Our VLP approximation follows a value-centric approach, where important values are assigned with greater accuracy. Second, we optimize VLP for small-batch GEMMs with asymmetric inputs efficiently, which leverages timely LLM optimizations, including weight-only quantization, key-value (KV) cache quantization, and group query attention. Finally, we design a new VLP architecture, Mugi, to encapsulate the innovations above and support full LLM workloads, while providing better performance, efficiency and sustainability. Our experimental results show that Mugi can offer significant improvements on throughput and energy efficiency, up to $45\times$ and $668\times$ for nonlinear softmax operations, and $2.07\times$ and $3.11\times$ for LLMs, and also decrease operational carbon for LLM operation by $1.45\times$ and embodied carbon by $1.48\times$.

cs.LG↗

arXiv2026

DiffRhythm 2: Efficient and High Fidelity Song Generation via Block Flow Matching

Generating full-length, high-quality songs is challenging, as it requires maintaining long-term coherence both across text and music modalities and within the music modality itself. Existing non-autoregressive (NAR) frameworks, while capable of producing high-quality songs, often struggle with the alignment between lyrics and vocal. Concurrently, catering to diverse musical preferences necessitates reinforcement learning from human feedback (RLHF). However, existing methods often rely on merging multiple models during multi-preference optimization, which results in significant performance degradation. To address these challenges, we introduce DiffRhythm 2, an end-to-end framework designed for high-fidelity, controllable song generation. To tackle the lyric alignment problem, DiffRhythm 2 employs a semi-autoregressive architecture based on block flow matching. This design enables faithful alignment of lyrics to singing vocals without relying on external labels and constraints, all while preserving the high generation quality and efficiency of NAR models. To make this framework computationally tractable for long sequences, we implement a music variational autoencoder (VAE) that achieves a low frame rate of 5 Hz while still enabling high-fidelity audio reconstruction. In addition, to overcome the limitations of multi-preference optimization in RLHF, we propose cross-pair preference optimization. This method effectively mitigates the performance drop typically associated with model merging, allowing for more robust optimization across diverse human preferences. We further enhance musicality and structural coherence by introducing stochastic block representation alignment loss.

eess.AS↗

arXiv2026

Deep-Learning-Based Control of a Decoupled Two-Segment Continuum Robot for Endoscopic Submucosal Dissection

Manual endoscopic submucosal dissection (ESD) is technically demanding, and existing single-segment robotic tools offer limited dexterity. These limitations motivate the development of more advanced solutions. To address this, DESectBot, a novel dual segment continuum robot with a decoupled structure and integrated surgical forceps, enabling 6 degrees of freedom (DoFs) tip dexterity for improved lesion targeting in ESD, was developed in this work. Deep learning controllers based on gated recurrent units (GRUs) for simultaneous tip position and orientation control, effectively handling the nonlinear coupling between continuum segments, were proposed. The GRU controller was benchmarked against Jacobian based inverse kinematics, model predictive control (MPC), a feedforward neural network (FNN), and a long short-term memory (LSTM) network. In nested-rectangle and Lissajous trajectory tracking tasks, the GRU achieved the lowest position/orientation RMSEs: 1.11 mm/ 4.62° and 0.81 mm/ 2.59°, respectively. For orientation control at a fixed position (four target poses), the GRU attained a mean RMSE of 0.14 mm and 0.72°, outperforming all alternatives. In a peg transfer task, the GRU achieved a 100% success rate (120 success/120 attempts) with an average transfer time of 11.8s, the STD significantly outperforms novice-controlled systems. Additionally, an ex vivo ESD demonstration grasping, elevating, and resecting tissue as the scalpel completed the cut confirmed that DESectBot provides sufficient stiffness to divide thick gastric mucosa and an operative workspace adequate for large lesions.These results confirm that GRU-based control significantly enhances precision, reliability, and usability in ESD surgical training scenarios.

cs.RO↗

arXiv2026

You Only Forward Once: An Efficient Compositional Judging Paradigm

Multimodal large language models (MLLMs) show strong potential as judges. However, existing approaches face a fundamental trade-off: adapting MLLMs to output a single score misaligns with the generative nature of MLLMs and limits fine-grained requirement understanding, whereas autoregressively generating judging analyses is prohibitively slow in high-throughput settings. Observing that judgment reduces to verifying whether inputs satisfy a set of structured requirements, we propose YOFO, a template-conditioned method that judges all requirements in a single forward pass. Built on an autoregressive model, YOFO accepts a structured requirement template and, in one inference step, produces a binary yes/no decision for each requirement by reading the logits of the final token associated with that requirement. This design yields orders-of-magnitude speedups while preserving interpretability. Extensive experiments show that YOFO not only achieves state-of-the-art results on standard recommendation datasets, but also supports dependency-aware analysis -- where subsequent judgments are conditioned on previous ones -- and further benefits from post-hoc CoT.

cs.AI↗

arXiv2026

Exploration of Unary Arithmetic-Based Matrix Multiply Units for Low Precision DL Accelerators

General matrix multiplication (GEMM) is a fundamental operation in deep learning (DL). With DL moving increasingly toward low precision, recent works have proposed novel unary GEMM designs as an alternative to conventional binary GEMM hardware. A rigorous evaluation of recent unary and binary GEMM designs is needed to assess the potential of unary hardware for future DL compute. This paper focuses on unary GEMM designs for integer-based DL inference and performs a detailed evaluation of three latest unary design proposals, namely, uGEMM, tuGEMM and tubGEMM, by comparing them to a conventional binary GEMM. Rigorous post-synthesis evaluations beyond prior works are performed across varying bit-widths and matrix sizes to assess the designs' tradeoffs and determine optimal sweetspots. Further, we perform weight sparsity analysis across eight pretrained convolutional neural networks (CNNs) and the LLaMA2 large language model (LLM). In this work, we demonstrate how unary GEMM can be effectively used for energy-efficient compute in future edge AI accelerators.

cs.AR↗

arXiv2026

Structure-Aware Decoding Mechanisms for Complex Entity Extraction with Large-Scale Language Models

This paper proposes a structure-aware decoding method based on large language models to address the difficulty of traditional approaches in maintaining both semantic integrity and structural consistency in nested and overlapping entity extraction tasks. The method introduces a candidate span generation mechanism and structured attention modeling to achieve unified modeling of entity boundaries, hierarchical relationships, and cross-dependencies. The model first uses a pretrained language model to obtain context-aware semantic representations, then captures multi-granular entity span features through candidate representation combinations, and introduces hierarchical structural constraints during decoding to ensure consistency between semantics and structure. To enhance stability in complex scenarios, the model jointly optimizes classification loss and structural consistency loss, maintaining high recognition accuracy under multi-entity co-occurrence and long-sentence dependency conditions. Experiments conducted on the ACE 2005 dataset demonstrate significant improvements in Accuracy, Precision, Recall, and F1-Score, particularly in nested and overlapping entity recognition, where the model shows stronger boundary localization and structural modeling capability. This study verifies the effectiveness of structure-aware decoding in complex semantic extraction tasks, provides a new perspective for developing language models with hierarchical understanding, and establishes a methodological foundation for high-precision information extraction.

cs.CL↗

arXiv2026

What Does Neuro Mean to Cardio? Investigating the Role of Clinical Specialty Data in Medical LLMs

In this paper, we introduce S-MedQA, an English medical question-answering (QA) dataset designed for benchmarking large language models (LLMs) in fine-grained clinical specialties. S-MedQA consists of over 24k examples, covering 15 medical specialties, with QA pairs that can have multiple specialty annotations, such as when a question is cross-disciplinary. The dataset is constructed using both machine and expert verification to maximize data availability and reliability. We use S-MedQA to investigate the role of clinical specialties in the knowledge-intensive scenario of medical QA. Our results show that training on data from a clinical specialty does not necessarily lead to the best performance on that specialty. Additionally, regardless of the specialty the LLM was fine-tuned on, token probabilities of clinically relevant terms consistently increase across all specialties. Based on these findings, we hypothesize that improvement gains, at least in our settings, are derived primarily from domain shifting (e.g., general to medical) rather than from injecting specialty-specific knowledge. This suggests a need to rethink the role of fine-tuning data in the medical domain. To encourage further advancements in the clinical NLP field, we release S-MedQA along with all the code required to reproduce our experiments for the research community.

cs.CL↗

arXiv2026

CSST Strong Lensing Preparation: Cosmological constraints from double-source-plane strong lensing systems in era of CSST

Double source plane strong lensing (DSPL) systems offer a robust, independent probe of cosmological parameters. The Chinese Space Station Telescope (CSST) is expected to discover hundreds of DSPLs, yet the survey modes and system configurations that best enable cosmological inference remain uncertain. To investigate the impact of varying signal-to-noise ratios (SNR) and Einstein radius ratios of DSPLs (denoted as $β^{-1}$ parameters) on cosmographic inference under different CSST survey modes (Wide Field (WF), Deep Field (DF), and Ultra-Deep Field (UDF)), we simulate and model mock lenses with Singular Isothermal Ellipsoid (SIE) mass profiles and Sérsic sources whose image properties are tailored to CSST specifications. Assuming a flat $w$CDM universe with fiducial values $Ω_{\rm m} = 0.30966$ and $w = -1$, and uniform priors of $Ω_{\rm m} \in [0, 1]$ and $w \in [-2, -1/3$), we find that the constraining power on cosmological parameters for a given DSPL system increases significantly with survey depth. For a representative DSPL system with two prominent arcs and a moderate $β^{-1}=1.17$, the constraints on ($w, Ω_{\rm m}$) improve from ($-1.28_{-1.00}^{+0.64}, 0.50_{-0.32}^{+0.28}$) in the WF to ($-1.59_{-0.32}^{+0.63}, 0.42_{-0.06}^{+0.15}$) in the UDF. Furthermore, we find that systems with smaller $β$ values yield tighter cosmographic constraints. We conclude that DSPL systems identified in UDF observations, particularly those with small $β$, are the most promising candidates for early-stage cosmological studies with CSST.

astro-ph.CO↗

arXiv2026

STAR-S: Improving Safety Alignment through Self-Taught Reasoning on Safety Rules

Defending against jailbreak attacks is crucial for the safe deployment of Large Language Models (LLMs). Recent research has attempted to improve safety by training models to reason over safety rules before responding. However, a key issue lies in determining what form of safety reasoning effectively defends against jailbreak attacks, which is difficult to explicitly design or directly obtain. To address this, we propose \textbf{STAR-S} (\textbf{S}elf-\textbf{TA}ught \textbf{R}easoning based on \textbf{S}afety rules), a framework that integrates the learning of safety rule reasoning into a self-taught loop. The core of STAR-S involves eliciting reasoning and reflection guided by safety rules, then leveraging fine-tuning to enhance safety reasoning. Repeating this process creates a synergistic cycle. Improvements in the model's reasoning and interpretation of safety rules allow it to produce better reasoning data under safety rule prompts, which is then utilized for further training. Experiments show that STAR-S effectively defends against jailbreak attacks, outperforming baselines. Code is available at: https://github.com/pikepokenew/STAR_S.git.

cs.AI↗

arXiv2026

Self-Routing RAG: Binding Selective Retrieval with Knowledge Verbalization

Selective retrieval aims to make retrieval-augmented generation (RAG) more efficient and reliable by skipping retrieval when an LLM's parametric knowledge suffices. Despite promising results, existing methods are constrained by a binary design choice: either retrieve from a single external source or skip retrieval and let the LLM directly produce the final answer. We argue that this fallback underestimates the model's knowledge and obscures the more general multi-source decision problem that arises in practical systems. We propose Self-Routing RAG (SR-RAG), which casts selective retrieval as knowledge source selection and treats the LLM itself as a first-class knowledge source. SR-RAG learns to select an appropriate knowledge source, optionally verbalize parametric knowledge, and answer using the selected source, all within a single left-to-right generation pass. SR-RAG further augments source selection by combining LLM-based uncertainty with a flexible external policy datastore to improve decision calibration. Across four benchmarks and three 7B-class LLMs, SR-RAG outperforms a strong selective retrieval baseline by 8.5%/2.1%/4.7% while performing 26%/40%/21% fewer retrievals, and it achieves favorable accuracy-latency trade-offs without dataset-specific threshold tuning.

cs.CL↗

arXiv2025

Large lepton asymmetry from axion inflation and helium abundance hinted by ACT

The generation of helical magnetic fields and the associated chiral asymmetry via the chiral anomaly is a generic feature in pseudoscalar inflation. In the presence of a Chern--Simons coupling between the inflaton and a U(1) gauge field, the homogeneous evolution of the inflaton induces a tachyonic instability in one circular polarization of the gauge field, resulting in the production of helical magnetic fields. In this work, we show that, in the case of a gauged lepton flavor symmetry, U(1)$_{L_i-L_j}$, this mechanism can lead to the generation of a sizable lepton asymmetry. In a simple setup, however, the resulting lepton asymmetry is typically too small to have an observational consequences, even setting aside constraints from baryon overproduction via sphaleron processes, due to the backreaction of the produced gauge fields and fermions on the inflationary dynamics. We demonstrate that this limitation can be overcome by implementing a mechanism to suppress fermion production during inflation. As a result, a much larger lepton asymmetry can be generated from the subsequent decay of magnetic helicity. Remarkably, for the gauged U(1)$_{L_μ-L_τ}$ symmetry, the generated asymmetry can be sufficiently large to suppress the primordial helium abundance, as may be inferred from recent cosmic microwave background observations by ACT.

hep-ph↗

arXiv2025

StereoMV2D: A Sparse Temporal Stereo-Enhanced Framework for Robust Multi-View 3D Object Detection

Multi-view 3D object detection is a fundamental task in autonomous driving perception, where achieving a balance between detection accuracy and computational efficiency remains crucial. Sparse query-based 3D detectors efficiently aggregate object-relevant features from multi-view images through a set of learnable queries, offering a concise and end-to-end detection paradigm. Building on this foundation, MV2D leverages 2D detection results to provide high-quality object priors for query initialization, enabling higher precision and recall. However, the inherent depth ambiguity in single-frame 2D detections still limits the accuracy of 3D query generation. To address this issue, we propose StereoMV2D, a unified framework that integrates temporal stereo modeling into the 2D detection-guided multi-view 3D detector. By exploiting cross-temporal disparities of the same object across adjacent frames, StereoMV2D enhances depth perception and refines the query priors, while performing all computations efficiently within 2D regions of interest (RoIs). Furthermore, a dynamic confidence gating mechanism adaptively evaluates the reliability of temporal stereo cues through learning statistical patterns derived from the inter-frame matching matrix together with appearance consistency, ensuring robust detection under object appearance and occlusion. Extensive experiments on the nuScenes and Argoverse 2 datasets demonstrate that StereoMV2D achieves superior detection performance without incurring significant computational overhead. Code will be available at https://github.com/Uddd821/StereoMV2D.

cs.CV↗

arXiv2025

Meta-Reinforcement Learning for Building Energy Management System

The building sector is one of the largest contributors to global energy consumption. Improving its energy efficiency is essential for reducing operational costs and greenhouse gas emissions. Energy management systems (EMS) play a key role in monitoring and controlling building appliances efficiently and reliably. With the increasing integration of renewable energy, intelligent EMS solutions have received growing attention. Reinforcement learning (RL) has recently been explored for this purpose and shows strong potential. However, most RL-based EMS methods require a large number of training steps to learn effective control policies, especially when adapting to unseen buildings, which limits their practical deployment. This paper introduces MetaEMS, a meta-reinforcement learning framework for EMS. MetaEMS improves learning efficiency by transferring knowledge from previously solved tasks to new ones through group-level and building-level adaptation, enabling fast adaptation and effective control across diverse building environments. Experimental results demonstrate that MetaEMS adapts more rapidly to unseen buildings and consistently outperforms baseline methods across various scenarios.

cs.AI↗