arXiv subjects

Hao Zhang

Publications and source records attributed to Hao Zhang.

At least 559 records · Page 31Linked to original sources

arXiv2025

Long$^2$RAG: Evaluating Long-Context & Long-Form Retrieval-Augmented Generation with Key Point Recall

Retrieval-augmented generation (RAG) is a promising approach to address the limitations of fixed knowledge in large language models (LLMs). However, current benchmarks for evaluating RAG systems suffer from two key deficiencies: (1) they fail to adequately measure LLMs' capability in handling long-context retrieval due to a lack of datasets that reflect the characteristics of retrieved documents, and (2) they lack a comprehensive evaluation method for assessing LLMs' ability to generate long-form responses that effectively exploits retrieved information. To address these shortcomings, we introduce the Long$^2$RAG benchmark and the Key Point Recall (KPR) metric. Long$^2$RAG comprises 280 questions spanning 10 domains and across 8 question categories, each associated with 5 retrieved documents with an average length of 2,444 words. KPR evaluates the extent to which LLMs incorporate key points extracted from the retrieved documents into their generated responses, providing a more nuanced assessment of their ability to exploit retrieved information.

cs.CL↗

arXiv2025

Sunny.jl: A Julia Package for Spin Dynamics

Sunny is a Julia package designed to serve the needs of the quantum magnetism community. It supports the specification of a very broad class of spin models and a diverse suite of numerical solvers. These include powerful methods for simulating spin dynamics both in and out of equilibrium. Uniquely, it features a broad generalization of classical and semiclassical approaches to SU(N) coherent states, which is useful for studying systems exhibiting strong spin-orbit coupling or local entanglement effects. Sunny also offers a well-developed framework for calculating the dynamical spin structure factor, enabling direct comparison with scattering experiments. Ease of use is a priority, with tools for symmetry-guided modeling and interactive visualization.

quant-ph↗

arXiv2025

Iterative Proximal-Minimization for Computing Saddle Points with Fixed Index

Computing saddle points with a prescribed Morse index on potential energy surfaces is crucial for characterizing transition states for nosie-induced rare transition events in physics and chemistry. Many numerical algorithms for this type of saddle points are based on the eigenvector-following idea and can be cast as an iterative minimization formulation (SINUM. Vol. 53, p.1786, 2015), but they may struggle with convergence issues and require good initial guesses. To address this challenge, we discuss the differential game interpretation of this iterative minimization formulation and investigate the relationship between this game's Nash equilibrium and saddle points on the potential energy surface. Our main contribution is that adding a proximal term, which grows faster than quadratic, to the game's cost function can enhance the stability and robustness. This approach produces a robust Iterative Proximal Minimization (IPM) algorithm for saddle point computing. We show that the IPM algorithm surpasses the preceding methods in robustness without compromising the convergence rate or increasing computational expense. The algorithm's efficacy and robustness are showcased through a two-dimensional test problem, and the Allen-Cahn, Cahn-Hilliard equation, underscoring its numerical robustness.

math.OC↗

arXiv2025

SMRU: Split-and-Merge Recurrent-based UNet for Acoustic Echo Cancellation and Noise Suppression

The proliferation of deep neural networks has spawned the rapid development of acoustic echo cancellation and noise suppression, and plenty of prior arts have been proposed, which yield promising performance. Nevertheless, they rarely consider the deployment generality in different processing scenarios, such as edge devices, and cloud processing. To this end, this paper proposes a general model, termed SMRU, to cover different application scenarios. The novelty lies in two-fold. First, a multi-scale band split layer and band merge layer are proposed to effectively fuse local frequency bands for lower complexity modeling. Besides, by simulating the multi-resolution feature modeling characteristic of the classical UNet structure, a novel recurrent-dominated UNet is devised. It consists of multiple variable frame rate blocks, each of which involves the causal time down-/up-sampling layer with varying compression ratios and the dual-path structure for inter- and intra-band modeling. The model is configured from 50 M/s to 6.8 G/s in terms of MACs, and the experimental results show that the proposed approach yields competitive or even better performance over existing baselines, and has the full potential to adapt to more general scenarios with varying complexity requirements.

cs.SD↗

arXiv2025

Predicting Compact Phrasal Rewrites with Large Language Models for ASR Post Editing

Large Language Models (LLMs) excel at rewriting tasks such as text style transfer and grammatical error correction. While there is considerable overlap between the inputs and outputs in these tasks, the decoding cost still increases with output length, regardless of the amount of overlap. By leveraging the overlap between the input and the output, Kaneko and Okazaki (2023) proposed model-agnostic edit span representations to compress the rewrites to save computation. They reported an output length reduction rate of nearly 80% with minimal accuracy impact in four rewriting tasks. In this paper, we propose alternative edit phrase representations inspired by phrase-based statistical machine translation. We systematically compare our phrasal representations with their span representations. We apply the LLM rewriting model to the task of Automatic Speech Recognition (ASR) post editing and show that our target-phrase-only edit representation has the best efficiency-accuracy trade-off. On the LibriSpeech test set, our method closes 50-60% of the WER gap between the edit span model and the full rewrite model while losing only 10-20% of the length reduction rate of the edit span model.

cs.CL↗

arXiv2025

Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models

Recently, promising progress has been made by open-source vision-language models (VLMs) in bringing their capabilities closer to those of proprietary frontier models. However, most open-source models only publish their final model weights, leaving the critical details of data strategies and implementation largely opaque. In this work, we address VLM post-training from a data-centric perspective, showing the key role of data strategy in developing frontier VLMs. By studying and building our post-training data strategy from scratch, we share detailed insights into the development processes, aiming to benefit the development of competitive models for the open-source community. Our introduced data strategy, together with training recipes and model design, leads to a family of performant VLMs named Eagle2. Specifically, Eagle2-9B achieves state-of-the-art results across various multimodal benchmarks, matching certain competitive models with up to 70B parameters.

cs.CV↗

arXiv2025

OpenMLDB: A Real-Time Relational Data Feature Computation System for Online ML

Efficient and consistent feature computation is crucial for a wide range of online ML applications. Typically, feature computation is divided into two distinct phases, i.e., offline stage for model training and online stage for model serving. These phases often rely on execution engines with different interface languages and function implementations, causing significant inconsistencies. Moreover, many online ML features involve complex time-series computations (e.g., functions over varied-length table windows) that differ from standard streaming and analytical queries. Existing data processing systems (e.g., Spark, Flink, DuckDB) often incur multi-second latencies for these computations, making them unsuitable for real-time online ML applications that demand timely feature updates. This paper presents OpenMLDB, a feature computation system deployed in 4Paradigm's SageOne platform and over 100 real scenarios. Technically, OpenMLDB first employs a unified query plan generator for consistent computation results across the offline and online stages, significantly reducing feature deployment overhead. Second, OpenMLDB provides an online execution engine that resolves performance bottlenecks caused by long window computations (via pre-aggregation) and multi-table window unions (via data self-adjusting). It also provides a high-performance offline execution engine with window parallel optimization and time-aware data skew resolving. Third, OpenMLDB features a compact data format and stream-focused indexing to maximize memory usage and accelerate data access. Evaluations in testing and real workloads reveal significant performance improvements and resource savings compared to the baseline systems. The open community of OpenMLDB now has over 150 contributors and gained 1.6k stars on GitHub.

cs.DB↗

arXiv2025

Overcoming the surface paradox: Buried perovskite quantum dots in wide-bandgap perovskite thin films

Colloidal perovskite quantum dots (PQDs) are an exciting platform for on-demand quantum, and classical optoelectronic and photonic devices. However, their potential success is limited by the extreme sensitivity and low stability arising from their weak intrinsic lattice bond energy and complex surface chemistry. Here we report a novel platform of buried perovskite quantum dots (b-PQDs) in a three-dimensional perovskite thin-film, fabricated using one-step, flash annealing, which overcomes surface related instabilities in colloidal perovskite dots. The b-PQDs demonstrate ultrabright and stable single-dot emission, with resolution-limited linewidths below 130 μeV, photon-antibunching (g^2(0)=0.1), no blinking, suppressed spectral diffusion, and high photon count rates of 10^4/s, consistent with unity quantum yield. The ultrasharp linewidth resolves exciton fine-structures (dark and triplet excitons) and their dynamics under a magnetic field. Additionally, b-PQDs can be electrically driven to emit single photons with 1 meV linewidth and photon-antibunching (g^2(0)=0.4). These results pave the way for on-chip, low-cost single-photon sources for next generation quantum optical communication and sensing.

physics.optics↗

arXiv2025

PalmBench: A Comprehensive Benchmark of Compressed Large Language Models on Mobile Platforms

Deploying large language models (LLMs) locally on mobile devices is advantageous in scenarios where transmitting data to remote cloud servers is either undesirable due to privacy concerns or impractical due to network connection. Recent advancements (MLC, 2023a; Gerganov, 2023) have facilitated the local deployment of LLMs. However, local deployment also presents challenges, particularly in balancing quality (generative performance), latency, and throughput within the hardware constraints of mobile devices. In this paper, we introduce our lightweight, all-in-one automated benchmarking framework that allows users to evaluate LLMs on mobile devices. We provide a comprehensive benchmark of various popular LLMs with different quantization configurations (both weights and activations) across multiple mobile platforms with varying hardware capabilities. Unlike traditional benchmarks that assess full-scale models on high-end GPU clusters, we focus on evaluating resource efficiency (memory and power consumption) and harmful output for compressed models on mobile devices. Our key observations include i) differences in energy efficiency and throughput across mobile platforms; ii) the impact of quantization on memory usage, GPU execution time, and power consumption; and iii) accuracy and performance degradation of quantized models compared to their non-quantized counterparts; and iv) the frequency of hallucinations and toxic content generated by compressed LLMs on mobile devices.

cs.LG↗

arXiv2025

Deep Unfolding Network with Spatial Alignment for multi-modal MRI reconstruction

Multi-modal Magnetic Resonance Imaging (MRI) offers complementary diagnostic information, but some modalities are limited by the long scanning time. To accelerate the whole acquisition process, MRI reconstruction of one modality from highly undersampled k-space data with another fully-sampled reference modality is an efficient solution. However, the misalignment between modalities, which is common in clinic practice, can negatively affect reconstruction quality. Existing deep learning-based methods that account for inter-modality misalignment perform better, but still share two main common limitations: (1) The spatial alignment task is not adaptively integrated with the reconstruction process, resulting in insufficient complementarity between the two tasks; (2) the entire framework has weak interpretability. In this paper, we construct a novel Deep Unfolding Network with Spatial Alignment, termed DUN-SA, to appropriately embed the spatial alignment task into the reconstruction process. Concretely, we derive a novel joint alignment-reconstruction model with a specially designed cross-modal spatial alignment term. By relaxing the model into cross-modal spatial alignment and multi-modal reconstruction tasks, we propose an effective algorithm to solve this model alternatively. Then, we unfold the iterative steps of the proposed algorithm and design corresponding network modules to build DUN-SA with interpretability. Through end-to-end training, we effectively compensate for spatial misalignment using only reconstruction loss, and utilize the progressively aligned reference modality to provide inter-modality prior to improve the reconstruction of the target modality. Comprehensive experiments on three real datasets demonstrate that our method exhibits superior reconstruction performance compared to state-of-the-art methods.

eess.IV↗

arXiv2025

Reducing disorder in PbTe nanowires for Majorana research

Material challenges are the key issue in Majorana research where surface disorder constrains device performance. Here, we tackle this challenge by embedding PbTe nanowires within a lattice-constant-matched crystal. The wire edges are shaped by self-organized growth instead of lithography, resulting in nearly-atomic-flat facets along both cross-sectional and longitudinal directions. Quantized conductance is observed at zero magnetic field with channel lengths maximally reaching 1.7 $μ$m, significantly surpassing the state-of-the-art of III-V nanowires (an order-of-magnitude improvement compared to InSb). Coupling PbTe to a Pb film unveils a flat interface spanning microns and a large superconducting gap of 1 meV. Our result not only represents a stride toward meeting the stringent low-disorder requirement for Majoranas, but may also open the door to various hybrid quantum devices requiring a low level of disorder.

cond-mat.mes-hall↗

arXiv2025

Anisotropy of PbTe nanowires with and without a superconductor

We investigate the anisotropic behaviors in PbTe and PbTe-Pb hybrid nanowires. In previous studies on PbTe, wire-to-wire variations in anisotropy indicate poor device control, posing a serious challenge for applications. Here, we achieve reproducible anisotropy in PbTe nanowires through a substantial reduction of disorder. We then couple PbTe to a superconductor Pb, and observe a pronounced deviation in the anisotropy behavior compared to bare PbTe nanowires. This deviation is gate-tunable and attributed to spin-orbit interaction and orbital effect, controlled by charge transfer between Pb and PbTe. These results provide a guidance for the controlled engineering of exotic quantum states in this hybrid material platform.

cond-mat.mes-hall↗

arXiv2025

MotionBridge: Dynamic Video Inbetweening with Flexible Controls

By generating plausible and smooth transitions between two image frames, video inbetweening is an essential tool for video editing and long video synthesis. Traditional works lack the capability to generate complex large motions. While recent video generation techniques are powerful in creating high-quality results, they often lack fine control over the details of intermediate frames, which can lead to results that do not align with the creative mind. We introduce MotionBridge, a unified video inbetweening framework that allows flexible controls, including trajectory strokes, keyframes, masks, guide pixels, and text. However, learning such multi-modal controls in a unified framework is a challenging task. We thus design two generators to extract the control signal faithfully and encode feature through dual-branch embedders to resolve ambiguities. We further introduce a curriculum training strategy to smoothly learn various controls. Extensive qualitative and quantitative experiments have demonstrated that such multi-modal controls enable a more dynamic, customizable, and contextually accurate visual narrative.

cs.CV↗

arXiv2025

Spatiotemporal Gaussian Optimization for 4D Cone Beam CT Reconstruction from Sparse Projections

In image-guided radiotherapy (IGRT), four-dimensional cone-beam computed tomography (4D-CBCT) is critical for assessing tumor motion during a patients breathing cycle prior to beam delivery. However, generating 4D-CBCT images with sufficient quality requires significantly more projection images than a standard 3D-CBCT scan, leading to extended scanning times and increased imaging dose to the patient. To address these limitations, there is a strong demand for methods capable of reconstructing high-quality 4D-CBCT images from a 1-minute 3D-CBCT acquisition. The challenge lies in the sparse sampling of projections, which introduces severe streaking artifacts and compromises image quality. This paper introduces a novel framework leveraging spatiotemporal Gaussian representation for 4D-CBCT reconstruction from sparse projections, achieving a balance between streak artifact reduction, dynamic motion preservation, and fine detail restoration. Each Gaussian is characterized by its 3D position, covariance, rotation, and density. Two-dimensional X-ray projection images can be rendered from the Gaussian point cloud representation via X-ray rasterization. The properties of each Gaussian were optimized by minimizing the discrepancy between the measured projections and the rendered X-ray projections. A Gaussian deformation network is jointly optimized to deform these Gaussian properties to obtain a 4D Gaussian representation for dynamic CBCT scene modeling. The final 4D-CBCT images are reconstructed by voxelizing the 4D Gaussians, achieving a high-quality representation that preserves both motion dynamics and spatial detail. The code and reconstruction results can be found at https://github.com/fuyabo/4DGS_for_4DCBCT/tree/main

physics.med-ph↗

arXiv2024

Loss-Aware Curriculum Learning for Chinese Grammatical Error Correction

Chinese grammatical error correction (CGEC) aims to detect and correct errors in the input Chinese sentences. Recently, Pre-trained Language Models (PLMS) have been employed to improve the performance. However, current approaches ignore that correction difficulty varies across different instances and treat these samples equally, enhancing the challenge of model learning. To address this problem, we propose a multi-granularity Curriculum Learning (CL) framework. Specifically, we first calculate the correction difficulty of these samples and feed them into the model from easy to hard batch by batch. Then Instance-Level CL is employed to help the model optimize in the appropriate direction automatically by regulating the loss function. Extensive experimental results and comprehensive analyses of various datasets prove the effectiveness of our method.

cs.CL↗

arXiv2024

Molar: Multimodal LLMs with Collaborative Filtering Alignment for Enhanced Sequential Recommendation

Sequential recommendation (SR) systems have evolved significantly over the past decade, transitioning from traditional collaborative filtering to deep learning approaches and, more recently, to large language models (LLMs). While the adoption of LLMs has driven substantial advancements, these models inherently lack collaborative filtering information, relying primarily on textual content data neglecting other modalities and thus failing to achieve optimal recommendation performance. To address this limitation, we propose Molar, a Multimodal large language sequential recommendation framework that integrates multiple content modalities with ID information to capture collaborative signals effectively. Molar employs an MLLM to generate unified item representations from both textual and non-textual data, facilitating comprehensive multimodal modeling and enriching item embeddings. Additionally, it incorporates collaborative filtering signals through a post-alignment mechanism, which aligns user representations from content-based and ID-based models, ensuring precise personalization and robust performance. By seamlessly combining multimodal content with collaborative filtering insights, Molar captures both user interests and contextual semantics, leading to superior recommendation accuracy. Extensive experiments validate that Molar significantly outperforms traditional and LLM-based baselines, highlighting its strength in utilizing multimodal data and collaborative signals for sequential recommendation tasks. The source code is available at https://anonymous.4open.science/r/Molar-8B06/.

cs.IR↗

arXiv2024

Exploiting Hybrid Policy in Reinforcement Learning for Interpretable Temporal Logic Manipulation

Reinforcement Learning (RL) based methods have been increasingly explored for robot learning. However, RL based methods often suffer from low sampling efficiency in the exploration phase, especially for long-horizon manipulation tasks, and generally neglect the semantic information from the task level, resulted in a delayed convergence or even tasks failure. To tackle these challenges, we propose a Temporal-Logic-guided Hybrid policy framework (HyTL) which leverages three-level decision layers to improve the agent's performance. Specifically, the task specifications are encoded via linear temporal logic (LTL) to improve performance and offer interpretability. And a waypoints planning module is designed with the feedback from the LTL-encoded task level as a high-level policy to improve the exploration efficiency. The middle-level policy selects which behavior primitives to execute, and the low-level policy specifies the corresponding parameters to interact with the environment. We evaluate HyTL on four challenging manipulation tasks, which demonstrate its effectiveness and interpretability. Our project is available at: https://sites.google.com/view/hytl-0257/.

cs.RO↗

arXiv2024

Cyclic Quantum Annealing: Searching for Deep Low-Energy States in 5000-Qubit Spin Glass

Quantum computers promise a qualitative speedup in solving a broad spectrum of practical optimization problems. The latter can be mapped onto the task of finding low-energy states of spin glasses, which is known to be exceedingly difficult. Using D-Wave's 5000-qubit quantum processor, we demonstrate that a recently proposed iterative cyclic quantum annealing algorithm can find deep low-energy states in record time. We also find intricate structures in a low-energy landscape of spin glasses, such as a power-law distribution of connected clusters with a small surface energy. These observations offer guidance for further improvement of the optimization algorithms.

cond-mat.dis-nn↗