arXiv subjects

Yu Wu

Publications and source records attributed to Yu Wu.

At least 91 records · Page 5Linked to original sources

arXiv2025

Forecasting When to Forecast: Accelerating Diffusion Models with Confidence-Gated Taylor

Diffusion Transformers (DiTs) have demonstrated remarkable performance in visual generation tasks. However, their low inference speed limits their deployment in low-resource applications. Recent training-free approaches exploit the redundancy of features across timesteps by caching and reusing past representations to accelerate inference. Building on this idea, TaylorSeer instead uses cached features to predict future ones via Taylor expansion. However, its module-level prediction across all transformer blocks (e.g., attention or feedforward modules) requires storing fine-grained intermediate features, leading to notable memory and computation overhead. Moreover, it adopts a fixed caching schedule without considering the varying accuracy of predictions across timesteps, which can lead to degraded outputs when prediction fails. To address these limitations, we propose a novel approach to better leverage Taylor-based acceleration. First, we shift the Taylor prediction target from the module level to the last block level, significantly reducing the number of cached features. Furthermore, observing strong sequential dependencies among Transformer blocks, we propose to use the error between the Taylor-estimated and actual outputs of the first block as an indicator of prediction reliability. If the error is small, we trust the Taylor prediction for the last block; otherwise, we fall back to full computation, thereby enabling a dynamic caching mechanism. Empirical results show that our method achieves a better balance between speed and quality, achieving a 3.17x acceleration on FLUX, 2.36x on DiT, and 4.14x on Wan Video with negligible quality drop. The Project Page is \href{https://cg-taylor-acce.github.io/CG-Taylor/}{here.}

cs.CV↗

arXiv2025

Arrow: Adaptive Scheduling Mechanisms for Disaggregated LLM Inference Architecture

Existing large language model (LLM) serving systems typically employ Prefill-Decode disaggregated architecture to prevent computational interference between the prefill and decode phases. However, in real-world LLM serving scenarios, significant fluctuations in request input/output lengths lead to imbalanced computational loads between prefill and decode nodes under traditional static node allocation strategies, consequently preventing efficient utilization of computing resources to improve the system's goodput. To address this challenge, we design and implement Arrow, an adaptive scheduler that leverages stateless instances and latency characteristics of prefill and decode tasks to achieve efficient adaptive request and instance scheduling. Arrow dynamically adjusts the number of instances handling prefill and decode tasks based on real-time cluster performance metrics, substantially enhancing the system's capability to handle traffic spikes and load variations. Our evaluation under diverse real-world workloads shows that Arrow achieves up to $2.55 \times$ higher request serving rates compared to state-of-the-art Prefill-Decode disaggregated serving systems.

cs.DC↗

arXiv2025

Incremental Human-Object Interaction Detection with Invariant Relation Representation Learning

In open-world environments, human-object interactions (HOIs) evolve continuously, challenging conventional closed-world HOI detection models. Inspired by humans' ability to progressively acquire knowledge, we explore incremental HOI detection (IHOID) to develop agents capable of discerning human-object relations in such dynamic environments. This setup confronts not only the common issue of catastrophic forgetting in incremental learning but also distinct challenges posed by interaction drift and detecting zero-shot HOI combinations with sequentially arriving data. Therefore, we propose a novel exemplar-free incremental relation distillation (IRD) framework. IRD decouples the learning of objects and relations, and introduces two unique distillation losses for learning invariant relation features across different HOI combinations that share the same relation. Extensive experiments on HICO-DET and V-COCO datasets demonstrate the superiority of our method over state-of-the-art baselines in mitigating forgetting, strengthening robustness against interaction drift, and generalization on zero-shot HOIs. Code is available at \href{https://github.com/weiyana/ContinualHOI}{this HTTP URL}

cs.CV↗

arXiv2025

GroundedPRM: Tree-Guided and Fidelity-Aware Process Reward Modeling for Step-Level Reasoning

Process Reward Models (PRMs) aim to improve multi-step reasoning in Large Language Models (LLMs) by supervising intermediate steps and identifying errors. However, building effective PRMs remains challenging due to the lack of scalable, high-quality annotations. Existing approaches rely on costly human labeling, LLM-based self-evaluation that is prone to hallucination, or Monte Carlo (MC) estimation, which infers step quality solely from rollout outcomes and often introduces noisy, misaligned supervision due to credit misattribution. These issues result in three core limitations: noisy rewards, low factual fidelity, and misalignment with step-level reasoning objectives. To address these challenges, we introduce GroundedPRM, a tree-guided and fidelity-aware framework for automatic process supervision. To reduce reward noise and enable fine-grained credit assignment, we construct structured reasoning paths via Monte Carlo Tree Search (MCTS). To eliminate hallucinated supervision, we validate each intermediate step using an external tool, providing execution-grounded correctness signals. To combine both step-level validation and global outcome assessment, we design a hybrid reward aggregation mechanism that fuses tool-based verification with MCTS-derived feedback. Finally, we format the reward signal into a rationale-enhanced, generative structure to promote interpretability and compatibility with instruction-tuned LLMs. GroundedPRM is trained on only 40K automatically labeled samples, amounting to just 10% of the data used by the best-performing PRM trained with auto-labeled supervision. Nevertheless, it achieves up to a 26% relative improvement in average performance on ProcessBench. When used for reward-guided greedy search, GroundedPRM outperforms even PRMs trained with human-labeled supervision, offering a scalable and verifiable path toward high-quality process-level reasoning.

cs.AI↗

arXiv2025

Cryogenic growth of aluminum: structural morphology, optical properties, superconductivity and microwave dielectric loss

We explore the molecular beam epitaxy synthesis of superconducting aluminum thin films grown on c-plane sapphire substrates at cryogenic temperatures of 6 K and compare their behavior with films synthesized at room temperature. We demonstrate that cryogenic growth increases structural disorder, producing crystalline grains that modify the optical, electrical, and superconducting properties of aluminum. We observe that cryogenic deposition changes the color of aluminum from fully reflective to yellow and correlate the pseudo-dielectric function and reflectance with structural changes in the film. We find that smaller grain sizes enhance the superconductivity of aluminum, increasing its critical temperature and critical field. We then estimate the superconducting gap and coherence length of Cooper pairs in aluminum in the presence of disorder. Finally, we fabricate superconducting microwave resonators on these films and find that, independently of the growth temperature, the system is dominated by two-level system loss with similar quality factors in the high and low power regimes. We further measure a higher kinetic inductance in the cryogenically grown films.

cond-mat.supr-con↗

arXiv2025

Cogito, Ergo Ludo: An Agent that Learns to Play by Reasoning and Planning

The pursuit of artificial agents that can learn to master complex environments has led to remarkable successes, yet prevailing deep reinforcement learning methods often rely on immense experience, encoding their knowledge opaquely within neural network weights. We propose a different paradigm, one in which an agent learns to play by reasoning and planning. We introduce Cogito, ergo ludo (CEL), a novel agent architecture that leverages a Large Language Model (LLM) to build an explicit, language-based understanding of its environment's mechanics and its own strategy. Starting from a tabula rasa state with no prior knowledge (except action set), CEL operates on a cycle of interaction and reflection. After each episode, the agent analyzes its complete trajectory to perform two concurrent learning processes: Rule Induction, where it refines its explicit model of the environment's dynamics, and Strategy and Playbook Summarization, where it distills experiences into an actionable strategic playbook. We evaluate CEL on diverse grid-world tasks (i.e., Minesweeper, Frozen Lake, and Sokoban), and show that the CEL agent successfully learns to master these games by autonomously discovering their rules and developing effective policies from sparse rewards. Ablation studies confirm that the iterative process is critical for sustained learning. Our work demonstrates a path toward more general and interpretable agents that not only act effectively but also build a transparent and improving model of their world through explicit reasoning on raw experience.

cs.AI↗

arXiv2025

All Giant Graviton Two-Point Functions at Two-Loops

We present a comprehensive two-loop computation of correlation functions involving two maximal giant gravitons and two arbitrary $R$-charge single-trace half-BPS operators in $\mathcal{N}=4$ Super-Yang-Mills theory. By combining the partially-contracted giant graviton (PCGG) method with the $\mathcal{N}=2$ harmonic superspace formalism, we achieve significant simplifications in perturbative calculations. The resulting correlation functions encode rich CFT data, from which we derive sum rules for the OPE coefficients. These sum rules are in perfect agreement with integrability predictions. Furthermore, at the integrand level, we find a hidden higher dimensional symmetry present at both one- and two-loop orders. This symmetry was discovered recently at strong coupling, which generalizes its counterpart in correlation functions of single-trace operators.

hep-th↗

arXiv2025

50 mm $\times$ 50 mm Cesium Atomic Vapor Cell for Terahertz Imaging: Implementation and Application

Rydberg atomic sensors offer transformative potential for high-speed, high-sensitivity terahertz (THz) imaging. However, previous systems are hindered by restricted imaging areas, largely due to the compact dimension of atomic vapor cells and inefficient beam-shaping methodologies. We present a THz imaging system with a 50 mm $\times$ 50 mm area, enabled by a custom-engineered scaled-up atomic vapor cell and an optimized beam-shaping optical architecture. Experimental validation confirms that this system achieves near-diffraction-limited, high resolution THz imaging at 0.55 THz under ambient conditions. Furthermore, its capabilities are demonstrated through real-time visualization of the diffusion dynamics of a deionized water droplet in anhydrous ethanol. This work not only expands the boundaries of Rydberg atomic sensors but also establishes a critical foundation for advancing THz imaging technologies toward into real-world, large-scale applications.

physics.optics↗

arXiv2025

Inference-Time Scaling for Generalist Reward Modeling

Reinforcement learning (RL) has been widely adopted in post-training for large language models (LLMs) at scale. Recently, the incentivization of reasoning capabilities in LLMs from RL indicates that $\textit{proper learning methods could enable effective inference-time scalability}$. A key challenge of RL is to obtain accurate reward signals for LLMs in various domains beyond verifiable questions or artificial rules. In this work, we investigate how to improve reward modeling (RM) with more inference compute for general queries, i.e. the $\textbf{inference-time scalability of generalist RM}$, and further, how to improve the effectiveness of performance-compute scaling with proper learning methods. For the RM approach, we adopt pointwise generative reward modeling (GRM) to enable flexibility for different input types and potential for inference-time scaling. For the learning method, we propose Self-Principled Critique Tuning (SPCT) to foster scalable reward generation behaviors in GRMs through online RL, to generate principles adaptively and critiques accurately, resulting in $\textbf{DeepSeek-GRM}$ models. Furthermore, for effective inference-time scaling, we use parallel sampling to expand compute usage, and introduce a meta RM to guide voting process for better scaling performance. Empirically, we show that SPCT significantly improves the quality and scalability of GRMs, outperforming existing methods and models in various RM benchmarks without severe biases, and could achieve better performance compared to training-time scaling. DeepSeek-GRM still meets challenges in some tasks, which we believe can be addressed by future efforts in generalist reward systems. The models are released at Hugging Face and ModelScope.

cs.CL↗

arXiv2025

Cooperative Suppression Strategy for Dual Thermal Transport Channels in Crystalline Materials

We propose a novel design principle for achieving ultralow thermal conductivity in crystalline materials via a "heavy-light and soft-stiff" structural motif. By combining heavy and light atomic species with soft and stiff bonding networks, both particle-like ($κ_p$) and wave-like ($κ_c$) phonon transport channels are concurrently suppressed. First-principles calculations show that this architecture induces a hierarchical phonon spectrum: soft-bonded heavy atoms generate dense low-frequency modes that enhance scattering and reduce $κ_p$, while stiff-bonded light atoms produce sparse high-frequency optical branches that disrupt coherence and lower $κ_c$. High-throughput screening identifies Tl$_4$SiS$_4$ ($κ_p$ = 0.10, $κ_c$ = 0.06 W/mK) and Tl$_4$GeS$_4$ ($κ_p$ = 0.09, $κ_c$ = 0.06 W/mK) as representative candidates with strongly suppressed transport in both channels. A minimal 1D triatomic chain model further demonstrates the generality of this mechanism, offering a new paradigm for phonon engineering beyond the conventional $κ_p$-$κ_c$ trade-off.

cond-mat.mtrl-sci↗

arXiv2025

The Silent Assistant: NoiseQuery as Implicit Guidance for Goal-Driven Image Generation

In this work, we introduce NoiseQuery as a novel method for enhanced noise initialization in versatile goal-driven text-to-image (T2I) generation. Specifically, we propose to leverage an aligned Gaussian noise as implicit guidance to complement explicit user-defined inputs, such as text prompts, for better generation quality and controllability. Unlike existing noise optimization methods designed for specific models, our approach is grounded in a fundamental examination of the generic finite-step noise scheduler design in diffusion formulation, allowing better generalization across different diffusion-based architectures in a tuning-free manner. This model-agnostic nature allows us to construct a reusable noise library compatible with multiple T2I models and enhancement techniques, serving as a foundational layer for more effective generation. Extensive experiments demonstrate that NoiseQuery enables fine-grained control and yields significant performance boosts not only over high-level semantics but also over low-level visual attributes, which are typically difficult to specify through text alone, with seamless integration into current workflows with minimal computational overhead.

cs.CV↗

arXiv2025

Rethinking Query-based Transformer for Continual Image Segmentation

Class-incremental/Continual image segmentation (CIS) aims to train an image segmenter in stages, where the set of available categories differs at each stage. To leverage the built-in objectness of query-based transformers, which mitigates catastrophic forgetting of mask proposals, current methods often decouple mask generation from the continual learning process. This study, however, identifies two key issues with decoupled frameworks: loss of plasticity and heavy reliance on input data order. To address these, we conduct an in-depth investigation of the built-in objectness and find that highly aggregated image features provide a shortcut for queries to generate masks through simple feature alignment. Based on this, we propose SimCIS, a simple yet powerful baseline for CIS. Its core idea is to directly select image features for query assignment, ensuring "perfect alignment" to preserve objectness, while simultaneously allowing queries to select new classes to promote plasticity. To further combat catastrophic forgetting of categories, we introduce cross-stage consistency in selection and an innovative "visual query"-based replay mechanism. Experiments demonstrate that SimCIS consistently outperforms state-of-the-art methods across various segmentation tasks, settings, splits, and input data orders. All models and codes will be made publicly available at https://github.com/SooLab/SimCIS.

cs.CV↗

arXiv2025

Adaptive Part Learning for Fine-Grained Generalized Category Discovery: A Plug-and-Play Enhancement

Generalized Category Discovery (GCD) aims to recognize unlabeled images from known and novel classes by distinguishing novel classes from known ones, while also transferring knowledge from another set of labeled images with known classes. Existing GCD methods rely on self-supervised vision transformers such as DINO for representation learning. However, focusing solely on the global representation of the DINO CLS token introduces an inherent trade-off between discriminability and generalization. In this paper, we introduce an adaptive part discovery and learning method, called APL, which generates consistent object parts and their correspondences across different similar images using a set of shared learnable part queries and DINO part priors, without requiring any additional annotations. More importantly, we propose a novel all-min contrastive loss to learn discriminative yet generalizable part representation, which adaptively highlights discriminative object parts to distinguish similar categories for enhanced discriminability while simultaneously sharing other parts to facilitate knowledge transfer for improved generalization. Our APL can easily be incorporated into different GCD frameworks by replacing their CLS token feature with our part representations, showing significant enhancements on fine-grained datasets.

cs.CV↗

arXiv2025

Spotlighting Partially Visible Cinematic Language for Video-to-Audio Generation via Self-distillation

Video-to-Audio (V2A) Generation achieves significant progress and plays a crucial role in film and video post-production. However, current methods overlook the cinematic language, a critical component of artistic expression in filmmaking. As a result, their performance deteriorates in scenarios where Foley targets are only partially visible. To address this challenge, we propose a simple self-distillation approach to extend V2A models to cinematic language scenarios. By simulating the cinematic language variations, the student model learns to align the video features of training pairs with the same audio-visual correspondences, enabling it to effectively capture the associations between sounds and partial visual information. Our method not only achieves impressive improvements under partial visibility across all evaluation metrics, but also enhances performance on the large-scale V2A dataset, VGGSound.

cs.CV↗

arXiv2025

Synergy of Rashba and Topological Effects for High-Performance Bismuth-Based Thermoelectrics

Band convergence is a key strategy for enhancing thermoelectric (TE) performance. Herein, we demonstrate a promising approach to enhance band convergence through inducing Rashba splitting in topological insulators. Theoretically designed Janus $β$-Bi$_2$Se$_2$Te and $β$-Bi$_2$Te$_2$Se exhibited inherent topological band inversion and Rashba splitting due to the strong spin-orbit coupling (SOC) with broken inversion symmetry. These characteristics synergistically improve band convergence, leading to a substantially enhanced power factor. Meanwhile, the Janus structural asymmetry suppresses lattice thermal conductivity. Consequently, Janus structures achieve boosted TE performance, especially for $β$-Bi$_2$Se$_2$Te, peaking figure of merit ($zT$) of 2.82. This work establishes a new framework for designing Janus topological compounds with high TE performance by the synergistic effect of Rashba splitting and band inversion.

cond-mat.mtrl-sci↗

arXiv2025

Shaping freeform nanophotonic devices with geometric neural parameterization

Nanophotonic freeform design has the potential to push the performance of optical components to new limits, but there remains a challenge to effectively perform optimization while reliably enforcing design and manufacturing constraints. We present Neuroshaper, a framework for freeform geometric parameterization in which nanophotonic device layouts are defined using an analytic neural network representation. Neuroshaper serves as a qualitatively new way to perform shape optimization by capturing multi-scalar, freeform geometries in an overparameterized representation scheme, enabling effective optimization in a smoothened, high dimensional geometric design space. We show that Neuroshaper can enforce constraints and topology manipulation in a manner where local constraints lead to global changes in device morphology. We further show numerically and experimentally that Neuroshaper can apply to a diversity of nanophotonic devices. The versatility and capabilities of Neuroshaper reflect the ability of neural representation to augment concepts in topological design.

physics.optics↗

arXiv2025

CodeI/O: Condensing Reasoning Patterns via Code Input-Output Prediction

Reasoning is a fundamental capability of Large Language Models. While prior research predominantly focuses on enhancing narrow skills like math or code generation, improving performance on many other reasoning tasks remains challenging due to sparse and fragmented training data. To address this issue, we propose CodeI/O, a novel approach that systematically condenses diverse reasoning patterns inherently embedded in contextually-grounded codes, through transforming the original code into a code input-output prediction format. By training models to predict inputs/outputs given code and test cases entirely in natural language as Chain-of-Thought (CoT) rationales, we expose them to universal reasoning primitives -- like logic flow planning, state-space searching, decision tree traversal, and modular decomposition -- while decoupling structured reasoning from code-specific syntax and preserving procedural rigor. Experimental results demonstrate CodeI/O leads to consistent improvements across symbolic, scientific, logic, math & numerical, and commonsense reasoning tasks. By matching the existing ground-truth outputs or re-executing the code with predicted inputs, we can verify each prediction and further enhance the CoTs through multi-turn revision, resulting in CodeI/O++ and achieving higher performance. Our data and models are available at https://github.com/hkust-nlp/CodeIO.

cs.CL↗

arXiv2025

Learning to Help in Multi-Class Settings

Deploying complex machine learning models on resource-constrained devices is challenging due to limited computational power, memory, and model retrainability. To address these limitations, a hybrid system can be established by augmenting the local model with a server-side model, where samples are selectively deferred by a rejector and then sent to the server for processing. The hybrid system enables efficient use of computational resources while minimizing the overhead associated with server usage. The recently proposed Learning to Help (L2H) model trains a server model given a fixed local (client) model, differing from the Learning to Defer (L2D) framework, which trains the client for a fixed (expert) server. In both L2D and L2H, the training includes learning a rejector at the client to determine when to query the server. In this work, we extend the L2H model from binary to multi-class classification problems and demonstrate its applicability in a number of different scenarios of practical interest in which access to the server may be limited by cost, availability, or policy. We derive a stage-switching surrogate loss function that is differentiable, convex, and consistent with the Bayes rule corresponding to the 0-1 loss for the L2H model. Experiments show that our proposed methods offer an efficient and practical solution for multi-class classification in resource-constrained environments.

cs.LG↗