arXiv subjects

Lei Zhao

Publications and source records attributed to Lei Zhao.

At least 91 records · Page 5Linked to original sources

arXiv2025

DPO-Shift: Shifting the Distribution of Direct Preference Optimization

Direct Preference Optimization (DPO) and its variants have become increasingly popular for aligning language models with human preferences. These methods aim to teach models to better distinguish between chosen (or preferred) and rejected (or dispreferred) responses. However, prior research has identified that the probability of chosen responses often decreases during training, and this phenomenon is known as likelihood displacement. To tackle this challenge, in this work we introduce DPO-Shift to controllably shift the distribution of the chosen probability. Then, we show that DPO-Shift exhibits a fundamental trade-off between improving the chosen probability and sacrificing the reward margin, as supported by both theoretical analysis and experimental validation. Furthermore, we demonstrate the superiority of DPO-Shift over DPO on downstream tasks such as MT-Bench and a designed win rate experiment. We believe this study shows that the likelihood displacement issue of DPO can be effectively mitigated with a simple, theoretically grounded solution. Our code is available at https://github.com/Meaquadddd/DPO-Shift.

cs.CL↗

arXiv2025

DualSpec: Text-to-spatial-audio Generation via Dual-Spectrogram Guided Diffusion Model

Text-to-audio (TTA), which generates audio signals from textual descriptions, has received huge attention in recent years. However, recent works focused on text to monaural audio only. As we know, spatial audio provides more immersive auditory experience than monaural audio, e.g. in virtual reality. To address this issue, we propose a text-to-spatial-audio (TTSA) generation framework named DualSpec. Specifically, it first trains variational autoencoders (VAEs) for extracting the latent acoustic representations from sound event audio. Then, given text that describes sound events and event directions, the proposed method uses the encoder of a pretrained large language model to transform the text into text features. Finally, it trains a diffusion model from the latent acoustic representations and text features for the spatial audio generation. In the inference stage, only the text description is needed to generate spatial audio. Particularly, to improve the synthesis quality and azimuth accuracy of the spatial sound events simultaneously, we propose to use two kinds of acoustic features. One is the Mel spectrograms which is good for improving the synthesis quality, and the other is the short-time Fourier transform spectrograms which is good at improving the azimuth accuracy. We provide a pipeline of constructing spatial audio dataset with text prompts, for the training of the VAEs and diffusion model. We also introduce new spatial-aware evaluation metrics to quantify the azimuth errors of the generated spatial audio recordings. Experimental results demonstrate that the proposed method can generate spatial audio with high directional and event consistency.

cs.SD↗

arXiv2025

Rapid Bone Scintigraphy Enhancement via Semantic Prior Distillation from Segment Anything Model

Rapid bone scintigraphy is crucial for diagnosing skeletal disorders and detecting tumor metastases in children, as it shortens scan duration and reduces discomfort. However, accelerated acquisition often degrades image quality, impairing the visibility of fine anatomical details and potentially compromising diagnosis. To overcome this limitation, we introduce the first application of SAM-based semantic priors for medical image restoration, utilizing the Segment Anything Model (SAM) to enhance pediatric rapid bone scintigraphy. Our approach employs two cascaded networks, $f^{IR1}$ and $f^{IR2}$, supported by three specialized modules: a Semantic Prior Integration (SPI) module, a Semantic Knowledge Distillation (SKD) module, and a Semantic Consistency Module (SCM). The SPI and SKD modules inject domain-specific semantic cues from a fine-tuned SAM, while the SCM preserves coherent semantic feature representations across both cascaded stages. Moreover, we present RBS, a novel Rapid Bone Scintigraphy dataset comprising paired standard (20 cm/min) and rapid (40 cm/min) scans from 137 pediatric patients aged 0.5 - 16 years, making it the first dataset tailored for pediatric rapid bone scintigraphy restoration. Extensive experiments on both a public endoscopic dataset and our RBS dataset demonstrate that our method consistently surpasses existing techniques in PSNR, SSIM, FID, and LPIPS metrics.

eess.IV↗

arXiv2025

StreamBP: Memory-Efficient Exact Backpropagation for Long Sequence Training of LLMs

Training language models on long sequence data is a demanding requirement for enhancing the model's capability on complex tasks, e.g., long-chain reasoning. However, as the sequence length scales up, the memory cost for storing activation values becomes huge during the Backpropagation (BP) process, even with the application of gradient checkpointing technique. To tackle this challenge, we propose a memory-efficient and exact BP method called StreamBP, which performs a linear decomposition of the chain rule along the sequence dimension in a layer-wise manner, significantly reducing the memory cost of activation values and logits. The proposed method is applicable to common objectives such as SFT, GRPO, and DPO. From an implementation perspective, StreamBP achieves less computational FLOPs and faster BP speed by leveraging the causal structure of the language model. Compared to gradient checkpointing, StreamBP scales up the maximum sequence length of BP by 2.8-5.5 times larger, while using comparable or even less BP time. Note that StreamBP's sequence length scaling ability can be directly transferred to batch size scaling for accelerating training. We further develop a communication-efficient distributed StreamBP to effectively support multi-GPU training and broaden its applicability. Our code can be easily integrated into the training pipeline of any transformer models and is available at https://github.com/Ledzy/StreamBP.

cs.LG↗

arXiv2025

A Low Power Monolithic Active Pixel Sensor Prototype for the STCF Inner Tracker

The Super Tau-Charm Facility (STCF) is a proposed $e^+e^-$ collider with a peak luminosity 100 times higher than that of the present tau-charm factory. The inner tracker (ITK) of STCF should feature a low material budget and high readout speed. Under these requirements, the monolithic active pixel sensor (MAPS) is considered as a promising candidate for the ITK. To minimize the power consumption of MAPS (for low material budget), larger-size sensors are proposed to reduce the scale of the readout circuitry while preserving the required position resolution. Multiple sensors with varying dimensions and structures were designed and integrated in several prototype chips for performance comparison, fabricated in a 180~nm CIS process. The in-pixel readout circuit can also provide time of arrival (ToA) and time-over-threshold (ToT) of the hit signal, with a least significant bit (LSB) of 50 ns. The peripheral readout circuit performs operations including timestamp correction, data aggregation, caching, framing, 8b/10b encoding, and serialization. According to simulation, the power consumption for a full-scale chip is about 55.7 mW/cm2. Preliminary measurements have been conducted on the prototype chips.

physics.ins-det↗

arXiv2025

Enhancing Keyphrase Extraction from Academic Articles Using Section Structure Information

The exponential increase in academic papers has significantly increased the time required for researchers to access relevant literature. Keyphrase Extraction (KPE) offers a solution to this situation by enabling researchers to efficiently retrieve relevant literature. The current study on KPE from academic articles aims to improve the performance of extraction models through innovative approaches using Title and Abstract as input corpora. However, the semantic richness of keywords is significantly constrained by the length of the abstract. While full-text-based KPE can address this issue, it simultaneously introduces noise, which significantly diminishes KPE performance. To address this issue, this paper utilized the structural features and section texts obtained from the section structure information of academic articles to extract keyphrase from academic papers. The approach consists of two main parts: (1) exploring the effect of seven structural features on KPE models, and (2) integrating the extraction results from all section texts used as input corpora for KPE models via a keyphrase integration algorithm to obtain the keyphrase integration result. Furthermore, this paper also examined the effect of the classification quality of section structure on the KPE performance. The results show that incorporating structural features improves KPE performance, though different features have varying effects on model efficacy. The keyphrase integration approach yields the best performance, and the classification quality of section structure can affect KPE performance. These findings indicate that using the section structure information of academic articles contributes to effective KPE from academic articles. The code and dataset supporting this study are available at https://github.com/yan-xinyi/SSB_KPE.

cs.CL↗

arXiv2025

SPAST: Arbitrary Style Transfer with Style Priors via Pre-trained Large-scale Model

Given an arbitrary content and style image, arbitrary style transfer aims to render a new stylized image which preserves the content image's structure and possesses the style image's style. Existing arbitrary style transfer methods are based on either small models or pre-trained large-scale models. The small model-based methods fail to generate high-quality stylized images, bringing artifacts and disharmonious patterns. The pre-trained large-scale model-based methods can generate high-quality stylized images but struggle to preserve the content structure and cost long inference time. To this end, we propose a new framework, called SPAST, to generate high-quality stylized images with less inference time. Specifically, we design a novel Local-global Window Size Stylization Module (LGWSSM)tofuse style features into content features. Besides, we introduce a novel style prior loss, which can dig out the style priors from a pre-trained large-scale model into the SPAST and motivate the SPAST to generate high-quality stylized images with short inference time.We conduct abundant experiments to verify that our proposed method can generate high-quality stylized images and less inference time compared with the SOTA arbitrary style transfer methods.

cs.CV↗

arXiv2025

Test-Time Domain Generalization via Universe Learning: A Multi-Graph Matching Approach for Medical Image Segmentation

Despite domain generalization (DG) has significantly addressed the performance degradation of pre-trained models caused by domain shifts, it often falls short in real-world deployment. Test-time adaptation (TTA), which adjusts a learned model using unlabeled test data, presents a promising solution. However, most existing TTA methods struggle to deliver strong performance in medical image segmentation, primarily because they overlook the crucial prior knowledge inherent to medical images. To address this challenge, we incorporate morphological information and propose a framework based on multi-graph matching. Specifically, we introduce learnable universe embeddings that integrate morphological priors during multi-source training, along with novel unsupervised test-time paradigms for domain adaptation. This approach guarantees cycle-consistency in multi-matching while enabling the model to more effectively capture the invariant priors of unseen data, significantly mitigating the effects of domain shifts. Extensive experiments demonstrate that our method outperforms other state-of-the-art approaches on two medical image segmentation benchmarks for both multi-source and single-source domain generalization tasks. The source code is available at https://github.com/Yore0/TTDG-MGM.

cs.CV↗

arXiv2025

DyArtbank: Diverse Artistic Style Transfer via Pre-trained Stable Diffusion and Dynamic Style Prompt Artbank

Artistic style transfer aims to transfer the learned style onto an arbitrary content image. However, most existing style transfer methods can only render consistent artistic stylized images, making it difficult for users to get enough stylized images to enjoy. To solve this issue, we propose a novel artistic style transfer framework called DyArtbank, which can generate diverse and highly realistic artistic stylized images. Specifically, we introduce a Dynamic Style Prompt ArtBank (DSPA), a set of learnable parameters. It can learn and store the style information from the collection of artworks, dynamically guiding pre-trained stable diffusion to generate diverse and highly realistic artistic stylized images. DSPA can also generate random artistic image samples with the learned style information, providing a new idea for data augmentation. Besides, a Key Content Feature Prompt (KCFP) module is proposed to provide sufficient content prompts for pre-trained stable diffusion to preserve the detailed structure of the input content image. Extensive qualitative and quantitative experiments verify the effectiveness of our proposed method. Code is available: https://github.com/Jamie-Cheung/DyArtbank

cs.CV↗

arXiv2025

Excitation of non-modal perturbations in hypersonic boundary layers by freestream forcing: shock-fitting harmonic linearised Navier-Stokes approach

In this paper, we study the receptivity of non-modal perturbations in hypersonic boundary layers over a blunt wedge subject to freestream vortical, entropy and acoustic perturbations. Due to the absence of the Mack-mode instability and the rather weak entropy-layer instability, the non-modal perturbation is considered as the dominant factor triggering transition. This is a highly intricate problem, given the complexities arising from the presence of the bow shock, entropy layer, and their interactions with oncoming disturbances. To tackle this challenge, we develop a novel numerical tool, the shock-fitting harmonic linearized Navier-Stokes (SF-HLNS) approach, which offers a comprehensive investigation on the dependence of the receptivity efficiency on the nose bluntness and properties of the freestream forcing. The numerical findings suggest that the non-modal perturbations are more susceptible to freestream acoustic and entropy perturbations compared to the vortical perturbations, with the optimal spanwise length scale being comparable with the downstream boundary-layer thickness. Notably, as the nose bluntness increases, the receptivity to the acoustic and entropy perturbations intensifies, reflecting the transition reversal phenomenon observed experimentally in configurations with relatively large bluntness. In contrast, the receptivity to freestream vortical perturbations weakens with increasing bluntness. Additionally, through the SF-HLNS calculations, we examine the credibility of the optimal growth theory (OGT) on describing the evolution of non-modal perturbations. Its accuracy in predicting the early-stage evolution and the energy amplification proves to be unreliable. Given its high-efficiency and high-accuracy nature, the SF-HLNS approach shows great potential as a valuable tool for conducting future research on hypersonic blunt-body boundary-layer transition.

physics.flu-dyn↗

arXiv2025

Gaussian highpass guided image filtering

Guided image filtering (GIF) is a popular smoothing technique, in which an additional image is used as a structure guidance for noise removal with edge preservation. The original GIF and some of its subsequent improvements are derived from a two-parameter local affine model (LAM), where the filtering output is a local affine transformation of the guidance image, but the input image is not taken into account in the LAM formulation. In this paper, we first introduce a single-parameter Prior Model based on Gaussian (highpass/lowpass) Filtering (PM-GF), in which the filtering output is the sum of a weighted portion of Gaussian highpass filtering of the guidance image and Gaussian smoothing of the input image. In the PM-GF, the guidance structure determined by Gaussian highpass filtering is obviously transferred to the filtering output, thereby better revealing the structure transfer mechanism of guided filtering. Then we propose several Gaussian highpass GIFs (GH-GIFs) based on the PM-GF by emulating the original GIF and some improvements, i.e., using PM-GF instead of LAM in these GIFs. Experimental results illustrate that the proposed GIFs outperform their counterparts in several image processing applications.

cs.CV↗

arXiv2025

Robust and Efficient Deep Hedging via Linearized Objective Neural Network

Deep hedging represents a cutting-edge approach to risk management for financial derivatives by leveraging the power of deep learning. However, existing methods often face challenges related to computational inefficiency, sensitivity to noisy data, and optimization complexity, limiting their practical applicability in dynamic and volatile markets. To address these limitations, we propose Deep Hedging with Linearized-objective Neural Network (DHLNN), a robust and generalizable framework that enhances the training procedure of deep learning models. By integrating a periodic fixed-gradient optimization method with linearized training dynamics, DHLNN stabilizes the training process, accelerates convergence, and improves robustness to noisy financial data. The framework incorporates trajectory-wide optimization and Black-Scholes Delta anchoring, ensuring alignment with established financial theory while maintaining flexibility to adapt to real-world market conditions. Extensive experiments on synthetic and real market data validate the effectiveness of DHLNN, demonstrating its ability to achieve faster convergence, improved stability, and superior hedging performance across diverse market scenarios.

cs.LG↗

arXiv2025

Adaptive Nesterov Accelerated Distributional Deep Hedging for Efficient Volatility Risk Management

In the field of financial derivatives trading, managing volatility risk is crucial for protecting investment portfolios from market changes. Traditional Vega hedging strategies, which often rely on basic and rule-based models, are hard to adapt well to rapidly changing market conditions. We introduce a new framework for dynamic Vega hedging, the Adaptive Nesterov Accelerated Distributional Deep Hedging (ANADDH), which combines distributional reinforcement learning with a tailored design based on adaptive Nesterov acceleration. This approach improves the learning process in complex financial environments by modeling the hedging efficiency distribution, providing a more accurate and responsive hedging strategy. The design of adaptive Nesterov acceleration refines gradient momentum adjustments, significantly enhancing the stability and speed of convergence of the model. Through empirical analysis and comparisons, our method demonstrates substantial performance gains over existing hedging techniques. Our results confirm that this innovative combination of distributional reinforcement learning with the proposed optimization techniques improves financial risk management and highlights the practical benefits of implementing advanced neural network architectures in the finance sector.

cs.LG↗

arXiv2025

Robust Federated Learning with Global Sensitivity Estimation for Financial Risk Management

In decentralized financial systems, robust and efficient Federated Learning (FL) is promising to handle diverse client environments and ensure resilience to systemic risks. We propose Federated Risk-Aware Learning with Central Sensitivity Estimation (FRAL-CSE), an innovative FL framework designed to enhance scalability, stability, and robustness in collaborative financial decision-making. The framework's core innovation lies in a central acceleration mechanism, guided by a quadratic sensitivity-based approximation of global model dynamics. By leveraging local sensitivity information derived from robust risk measurements, FRAL-CSE performs a curvature-informed global update that efficiently incorporates second-order information without requiring repeated local re-evaluations, thereby enhancing training efficiency and improving optimization stability. Additionally, distortion risk measures are embedded into the training objectives to capture tail risks and ensure robustness against extreme scenarios. Extensive experiments validate the effectiveness of FRAL-CSE in accelerating convergence and improving resilience across heterogeneous datasets compared to state-of-the-art baselines.

cs.LG↗

arXiv2025

AudioSpa: Spatializing Sound Events with Text

Text-to-audio (TTA) systems have recently demonstrated strong performance in synthesizing monaural audio from text. However, the task of generating binaural spatial audio from text, which provides a more immersive auditory experience by incorporating the sense of spatiality, have not been explored yet. In this work, we introduce text-guided binaural audio generation. As an early effort, we focus on the scenario where a monaural reference audio is given additionally. The core problem is to associate specific sound events with their directions, thereby creating binaural spatial audio. The challenge lies in the complexity of textual descriptions and the limited availability of single-source sound event datasets. To address this, we propose AudioSpa, an end-to-end model that applies large language models to process both acoustic and textual information. We employ fusion multi-head attention (FMHA) to integrate text tokens, which enhances the generation capability of the multimodal learning. Additionally, we propose a binaural source localization model to assess the quality of the generated audio. Finally, we design a data augmentation strategy to generate diverse datasets, which enables the model to spatialize sound events across various spatial positions. Experimental results demonstrate that our model is able to put sounds at the specified locations accurately. It achieves competitive performance in both localization accuracy and signal distortion. Our demonstrations are available at https://linfeng-feng.github.io/AudioSpa-demo.

eess.AS↗

arXiv2025

X-TIME: An in-memory engine for accelerating machine learning on tabular data with CAMs

Structured, or tabular, data is the most common format in data science. While deep learning models have proven formidable in learning from unstructured data such as images or speech, they are less accurate than simpler approaches when learning from tabular data. In contrast, modern tree-based Machine Learning (ML) models shine in extracting relevant information from structured data. An essential requirement in data science is to reduce model inference latency in cases where, for example, models are used in a closed loop with simulation to accelerate scientific discovery. However, the hardware acceleration community has mostly focused on deep neural networks and largely ignored other forms of machine learning. Previous work has described the use of an analog content addressable memory (CAM) component for efficiently mapping random forests. In this work, we develop an analog-digital architecture that implements a novel increased precision analog CAM and a programmable chip for inference of state-of-the-art tree-based ML models, such as XGBoost, CatBoost, and others. Thanks to hardware-aware training, X-TIME reaches state-of-the-art accuracy and 119x higher throughput at 9740x lower latency with >150x improved energy efficiency compared with a state-of-the-art GPU for models with up to 4096 trees and depth of 8, with a 19W peak power consumption.

cs.LG↗

arXiv2025

MC-VTON: Minimal Control Virtual Try-On Diffusion Transformer

Virtual try-on methods based on diffusion models achieve realistic try-on effects. They use an extra reference network or an additional image encoder to process multiple conditional image inputs, which adds complexity pre-processing and additional computational costs. Besides, they require more than 25 inference steps, bringing longer inference time. In this work, with the development of diffusion transformer (DiT), we rethink the necessity of additional reference network or image encoder and introduce MC-VTON, which leverages DiT's intrinsic backbone to seamlessly integrate minimal conditional try-on inputs. Compared to existing methods, the superiority of MC-VTON is demonstrated in four aspects: (1) Superior detail fidelity. Our DiT-based MC-VTON exhibits superior fidelity in preserving fine-grained details. (2) Simplified network and inputs. We remove any extra reference network or image encoder. We also remove unnecessary conditions like the long prompt, pose estimation, human parsing, and depth map. We require only the masked person image and the garment image. (3) Parameter-efficient training. To process the try-on task, we fine-tune the FLUX.1-dev with only 39.7M additional parameters (0.33% of the backbone parameters). (4) Less inference steps. We apply distillation diffusion on MC-VTON and only need 8 steps to generate a realistic try-on image, with only 86.8M additional parameters (0.72% of the backbone parameters). Experiments show that MC-VTON achieves superior qualitative and quantitative results with fewer condition inputs, trainable parameters, and inference steps than baseline methods.

cs.CV↗

arXiv2024

Better Knowledge Enhancement for Privacy-Preserving Cross-Project Defect Prediction

Cross-Project Defect Prediction (CPDP) poses a non-trivial challenge to construct a reliable defect predictor by leveraging data from other projects, particularly when data owners are concerned about data privacy. In recent years, Federated Learning (FL) has become an emerging paradigm to guarantee privacy information by collaborative training a global model among multiple parties without sharing raw data. While the direct application of FL to the CPDP task offers a promising solution to address privacy concerns, the data heterogeneity arising from proprietary projects across different companies or organizations will bring troubles for model training. In this paper, we study the privacy-preserving cross-project defect prediction with data heterogeneity under the federated learning framework. To address this problem, we propose a novel knowledge enhancement approach named FedDP with two simple but effective solutions: 1. Local Heterogeneity Awareness and 2. Global Knowledge Distillation. Specifically, we employ open-source project data as the distillation dataset and optimize the global model with the heterogeneity-aware local model ensemble via knowledge distillation. Experimental results on 19 projects from two datasets demonstrate that our method significantly outperforms baselines.

cs.LG↗