arXiv subjects

Xiaofeng Yang

Publications and source records attributed to Xiaofeng Yang.

At least 91 records · Page 5Linked to original sources

arXiv2025

DataLab: A Unified Platform for LLM-Powered Business Intelligence

Business intelligence (BI) transforms large volumes of data within modern organizations into actionable insights for informed decision-making. Recently, large language model (LLM)-based agents have streamlined the BI workflow by automatically performing task planning, reasoning, and actions in executable environments based on natural language (NL) queries. However, existing approaches primarily focus on individual BI tasks such as NL2SQL and NL2VIS. The fragmentation of tasks across different data roles and tools lead to inefficiencies and potential errors due to the iterative and collaborative nature of BI. In this paper, we introduce DataLab, a unified BI platform that integrates a one-stop LLM-based agent framework with an augmented computational notebook interface. DataLab supports various BI tasks for different data roles in data preparation, analysis, and visualization by seamlessly combining LLM assistance with user customization within a single environment. To achieve this unification, we design a domain knowledge incorporation module tailored for enterprise-specific BI tasks, an inter-agent communication mechanism to facilitate information sharing across the BI workflow, and a cell-based context management strategy to enhance context utilization efficiency in BI notebooks. Extensive experiments demonstrate that DataLab achieves state-of-the-art performance on various BI tasks across popular research benchmarks. Moreover, DataLab maintains high effectiveness and efficiency on real-world datasets from Tencent, achieving up to a 58.58% increase in accuracy and a 61.65% reduction in token cost on enterprise-specific BI tasks.

cs.DB↗

arXiv2025

RoMedFormer: A Rotary-Embedding Transformer Foundation Model for 3D Genito-Pelvic Structure Segmentation in MRI and CT

Deep learning-based segmentation of genito-pelvic structures in MRI and CT is crucial for applications such as radiation therapy, surgical planning, and disease diagnosis. However, existing segmentation models often struggle with generalizability across imaging modalities, and anatomical variations. In this work, we propose RoMedFormer, a rotary-embedding transformer-based foundation model designed for 3D female genito-pelvic structure segmentation in both MRI and CT. RoMedFormer leverages self-supervised learning and rotary positional embeddings to enhance spatial feature representation and capture long-range dependencies in 3D medical data. We pre-train our model using a diverse dataset of 3D MRI and CT scans and fine-tune it for downstream segmentation tasks. Experimental results demonstrate that RoMedFormer achieves superior performance segmenting genito-pelvic organs. Our findings highlight the potential of transformer-based architectures in medical image segmentation and pave the way for more transferable segmentation frameworks.

eess.IV↗

arXiv2025

Efficient Hardware Accelerator Based on Medium Granularity Dataflow for SpTRSV

Sparse triangular solve (SpTRSV) is widely used in various domains. Numerous studies have been conducted using CPUs, GPUs, and specific hardware accelerators, where dataflows can be categorized into coarse and fine granularity. Coarse dataflows offer good spatial locality but suffer from low parallelism, while fine dataflows provide high parallelism but disrupt the spatial structure, leading to increased nodes and poor data reuse. This paper proposes a novel hardware accelerator for SpTRSV or SpTRSV-like DAGs. The accelerator implements a medium granularity dataflow through hardware-software codesign and achieves both excellent spatial locality and high parallelism. Additionally, a partial sum caching mechanism is introduced to reduce the blocking frequency of processing elements (PEs), and a reordering algorithm of intra-node edges computation is developed to enhance data reuse. Experimental results on 245 benchmarks with node counts reaching up to 85,392 demonstrate that this work achieves average performance improvements of 7.0$\times$ (up to 27.8$\times$) over CPUs and 5.8$\times$ (up to 98.8$\times$) over GPUs. Compared to the state-of-the-art technique (DPU-v2), this work shows a 2.5$\times$ (up to 5.9$\times$) average performance improvement and 1.7$\times$ (up to 4.1$\times$) average energy efficiency enhancement.

cs.DC↗

arXiv2025

Towards Universal Text-driven CT Image Segmentation

Computed tomography (CT) is extensively used for accurate visualization and segmentation of organs and lesions. While deep learning models such as convolutional neural networks (CNNs) and vision transformers (ViTs) have significantly improved CT image analysis, their performance often declines when applied to diverse, real-world clinical data. Although foundation models offer a broader and more adaptable solution, their potential is limited due to the challenge of obtaining large-scale, voxel-level annotations for medical images. In response to these challenges, prompting-based models using visual or text prompts have emerged. Visual-prompting methods, such as the Segment Anything Model (SAM), still require significant manual input and can introduce ambiguity when applied to clinical scenarios. Instead, foundation models that use text prompts offer a more versatile and clinically relevant approach. Notably, current text-prompt models, such as the CLIP-Driven Universal Model, are limited to text prompts already encountered during training and struggle to process the complex and diverse scenarios of real-world clinical applications. Instead of fine-tuning models trained from natural imaging, we propose OpenVocabCT, a vision-language model pretrained on large-scale 3D CT images for universal text-driven segmentation. Using the large-scale CT-RATE dataset, we decompose the diagnostic reports into fine-grained, organ-level descriptions using large language models for multi-granular contrastive learning. We evaluate our OpenVocabCT on downstream segmentation tasks across nine public datasets for organ and tumor segmentation, demonstrating the superior performance of our model compared to existing methods. All code, datasets, and models will be publicly released at https://github.com/ricklisz/OpenVocabCT.

cs.CV↗

arXiv2025

Triad: Vision Foundation Model for 3D Magnetic Resonance Imaging

Vision foundation models (VFMs) are pre-trained on extensive image datasets to learn general representations for diverse types of data. These models can subsequently be fine-tuned for specific downstream tasks, significantly boosting performance across a broad range of applications. However, existing vision foundation models that claim to be applicable to various clinical tasks are mostly pre-trained on 3D computed tomography (CT), which benefits from the availability of extensive 3D CT databases. Significant differences between CT and magnetic resonance imaging (MRI) in imaging principles, signal characteristics, and data distribution may hinder their practical performance and versatility in MRI-specific applications. Here, we propose Triad, a vision foundation model for 3D MRI. Triad adopts a widely used autoencoder architecture to learn robust representations from 131,170 3D MRI volumes and uses organ-independent imaging descriptions to constrain the semantic distribution of the visual modality. The above pre-training dataset is called Triad-131K, which is currently the largest 3D MRI pre-training dataset. We evaluate Triad across three tasks, namely, organ/tumor segmentation, organ/cancer classification, and medical image registration, in two data modalities (within-domain and out-of-domain) settings using 25 downstream datasets. By initializing models with Triad's pre-trained weights, nnUNet-Triad improves segmentation performance by 2.51% compared to nnUNet-Scratch across 17 datasets. Swin-B-Triad achieves a 3.97% improvement over Swin-B-Scratch in classification tasks across five datasets. SwinUNETR-Triad improves by 4.00% compared to SwinUNETR-Scratch in registration tasks across two datasets. Our study demonstrates that pre-training can improve performance when the data modalities and organs of upstream and downstream tasks are consistent.

cs.CV↗

arXiv2025

Text-to-Image Rectified Flow as Plug-and-Play Priors

Large-scale diffusion models have achieved remarkable performance in generative tasks. Beyond their initial training applications, these models have proven their ability to function as versatile plug-and-play priors. For instance, 2D diffusion models can serve as loss functions to optimize 3D implicit models. Rectified flow, a novel class of generative models, enforces a linear progression from the source to the target distribution and has demonstrated superior performance across various domains. Compared to diffusion-based methods, rectified flow approaches surpass in terms of generation quality and efficiency, requiring fewer inference steps. In this work, we present theoretical and experimental evidence demonstrating that rectified flow based methods offer similar functionalities to diffusion models - they can also serve as effective priors. Besides the generative capabilities of diffusion priors, motivated by the unique time-symmetry properties of rectified flow models, a variant of our method can additionally perform image inversion. Experimentally, our rectified flow-based priors outperform their diffusion counterparts - the SDS and VSD losses - in text-to-3D generation. Our method also displays competitive performance in image inversion and editing.

cs.CV↗

arXiv2025

A Physics-Informed Deep Learning Model for MRI Brain Motion Correction

Background: MRI is crucial for brain imaging but is highly susceptible to motion artifacts due to long acquisition times. This study introduces PI-MoCoNet, a physics-informed motion correction network that integrates spatial and k-space information to remove motion artifacts without explicit motion parameter estimation, enhancing image fidelity and diagnostic reliability. Materials and Methods: PI-MoCoNet consists of a motion detection network (U-net with spatial averaging) to identify corrupted k-space lines and a motion correction network (U-net with Swin Transformer blocks) to reconstruct motion-free images. The correction is guided by three loss functions: reconstruction (L1), perceptual (LPIPS), and data consistency (Ldc). Motion artifacts were simulated via rigid phase encoding perturbations and evaluated on IXI and MR-ART datasets against Pix2Pix, CycleGAN, and U-net using PSNR, SSIM, and NMSE. Results: PI-MoCoNet significantly improved image quality. On IXI, for minor artifacts, PSNR increased from 34.15 dB to 45.95 dB, SSIM from 0.87 to 1.00, and NMSE reduced from 0.55% to 0.04%. For moderate artifacts, PSNR improved from 30.23 dB to 42.16 dB, SSIM from 0.80 to 0.99, and NMSE from 1.32% to 0.09%. For heavy artifacts, PSNR rose from 27.99 dB to 36.01 dB, SSIM from 0.75 to 0.97, and NMSE decreased from 2.21% to 0.36%. On MR-ART, PI-MoCoNet achieved PSNR gains of ~10 dB and SSIM improvements of up to 0.20, with NMSE reductions of ~6%. Ablation studies confirmed the importance of data consistency and perceptual losses, yielding a 1 dB PSNR gain and 0.17% NMSE reduction. Conclusions: PI-MoCoNet effectively mitigates motion artifacts in brain MRI, outperforming existing methods. Its ability to integrate spatial and k-space information makes it a promising tool for clinical use in motion-prone settings. Code: https://github.com/mosaf/PI-MoCoNet.git.

cs.CV↗

arXiv2025

A Comparative Dosimetric Study of Proton and Photon Therapy in Stereotactic Arrhythmia Radioablation for Ventricular Tachycardia

Purpose: VT is a life-threatening arrhythmia commonly treated with catheter ablation; however, some cases remain refractory to conventional treatment. STAR has emerged as a non-invasive option for such patients. While photon-based STAR has shown efficacy, proton therapy offers potential advantages due to its superior dose conformity and sparing of critical OARs, including the heart itself. This study aims to investigate and compare the dosimetry between proton and photon therapy for VT, focusing on target coverage and OAR sparing. Methods: We performed a retrospective study on a cohort of 34 VT patients who received photon STAR. Proton STAR plans were generated using robust optimization in RayStation to deliver the same prescription dose of 25 Gy in a single fraction while minimizing dose to OARs. Dosimetric metrics, including D99, D95, Dmean, and D0.03cc, were extracted for critical OARs and VAS. Shapiro-Wilk tests were used to assess normality, followed by paired t-tests or Wilcoxon signed-rank tests for statistical comparisons between modalities, with Bonferroni correction applied for multiple comparisons. Results: Proton and photon plans achieved comparable target coverage, with VAS D95 of 24.1 +/- 1.2 Gy vs. 24.7 +/- 1.0 Gy (p=0.294). Proton therapy significantly reduced OAR doses, including heart Dmean (3.6 +/- 1.5 Gy vs. 5.5 +/- 2.0 Gy, p<0.001), lungs Dmean (1.6 +/- 1.5 Gy vs. 2.1 +/- 1.4 Gy, p<0.001), and esophagus Dmean (0.3 +/- 0.6 Gy vs. 1.6 +/- 1.3 Gy, p<0.001), while maintaining optimal target coverage. Conclusion: Proton therapy for STAR demonstrates significant dosimetric advantages in sparing the heart and other critical OARs compared to photon therapy for VT, while maintaining equivalent target coverage. These findings highlight the potential of proton therapy to reduce treatment-related toxicity and improve outcomes for VT patients.

physics.med-ph↗

arXiv2025

An Adaptive Proton FLASH Therapy Using Modularized Pin Ridge Filter

In this paper, we proposed a method to optimize adaptive proton FLASH therapy (ADP FLASH) using modularized pin ridge filters (pRFs) by recycling module pins from the initial plan while reducing pRF adjustments in adaptive FLASH planning. Initially, single energy (250 MeV) FLASH pRF plans were created using pencil beam directions (PBDs) from initial IMPT plans on the planning CT (pCT). PBDs are classified as new/changed ($Δ$E > > 5 MeV) or unchanged by comparing spot maps for targets between pCT and re-CT. We used an iterative least square regression model to identify recyclable PBDs with minimal relative changes to spot MU weighting. Two PBDs with the least square error were retrieved per iteration and added to the background plan, and the remaining PBDs were reoptimized for the adaptive plan in subsequent iterations. The method was validated on three liver SBRT cases (50 Gy in 5 fractions) by comparing various dosimetric parameters across initial pRF plans on pCT, reCT and the ADP FLASH pRF plans on reCT. V100 for initial pRF plans on pCT, reCT, and ADP FLASH pRF plans for the three cases were as follows: (93.7%, 89.2%, 91.4%), (93.5%, 60.2%, 91.7%), (97.3%, 69.9%, 98.8%). We observe a decline in plan quality when applying the initial pRF to the reCT, whereas the ADP FLASH pRF approach restores quality comparable to the initial pRF on the pCT. FLASH effect of the initial pRF and ADP pRF plans were evaluated with a dose and dose rate threshold of 1Gy and 40Gy/s, respectively, using the FLASH effectiveness model. The proposed method recycled 91.2%, 71%, and 64.7% of PBDs from initial pRF plans for the three cases while maintaining all clinical goals and preserving FLASH effects across all cases.

physics.med-ph↗

arXiv2025

Ultrafast Proton Delivery with Pin Ridge Filters (pRFs): A Novel Approach for Motion Management in Proton Therapy

Active breath-hold techniques effectively mitigate respiratory motion but pose challenges for patients who are ineligible for the procedure. Conventional treatment planning relies on multiple energy layers, extending delivery time due to slow layer switching. We propose to use pin ridge filters (pRFs), initially developed for FLASH radiotherapy, to construct a single energy beam plan and minimize dose delivery time. The conventional ITV--based free--breathing treatment plan served as the reference. A GTV--based IMPT--DS plan with a downstream energy modulation strategy was developed based on a new beam model that was commissioned using the maximum energy of the IMPT plan. Consequently, a nested iterative pencil beam direction (PBD) spot reduction process removed low--weighted spots along each PBD, generating pRFs with coarser resolution. Finally, the IMPT--DS plan was then converted into an IMPT--pRF plan, using a monoenergetic beam with optimized spot positions and weights. This approach was validated on lung and liver SBRT cases (10 Gy RBE x 5). For the lung case, the mean lung--GTV dose decreased from 10.3 Gy to 6.9 Gy, with delivery time reduced from 188.79 to 36.16 seconds. The largest time reduction was at 150°, from 47.4 to 3.99 seconds. For the liver case, the mean liver--GTV dose decreased from 5.7 Gy to 3.8 Gy, with delivery time reduced from 111.13 to 30.54 seconds. The largest time reduction was at 180°, from 38.57 to 3.94 seconds. This method significantly reduces dose delivery time and organ at risk dose. Further analysis is needed to validate its clinical feasibility.

physics.med-ph↗

arXiv2025

IPVTON: Image-based 3D Virtual Try-on with Image Prompt Adapter

Given a pair of images depicting a person and a garment separately, image-based 3D virtual try-on methods aim to reconstruct a 3D human model that realistically portrays the person wearing the desired garment. In this paper, we present IPVTON, a novel image-based 3D virtual try-on framework. IPVTON employs score distillation sampling with image prompts to optimize a hybrid 3D human representation, integrating target garment features into diffusion priors through an image prompt adapter. To avoid interference with non-target areas, we leverage mask-guided image prompt embeddings to focus the image features on the try-on regions. Moreover, we impose geometric constraints on the 3D model with a pseudo silhouette generated by ControlNet, ensuring that the clothed 3D human model retains the shape of the source identity while accurately wearing the target garments. Extensive qualitative and quantitative experiments demonstrate that IPVTON outperforms previous methods in image-based 3D virtual try-on tasks, excelling in both geometry and texture.

cs.CV↗

arXiv2025

Magic-Boost: Boost 3D Generation with Multi-View Conditioned Diffusion

Benefiting from the rapid development of 2D diffusion models, 3D content generation has witnessed significant progress. One promising solution is to finetune the pre-trained 2D diffusion models to produce multi-view images and then reconstruct them into 3D assets via feed-forward sparse-view reconstruction models. However, limited by the 3D inconsistency in the generated multi-view images and the low reconstruction resolution of the feed-forward reconstruction models, the generated 3d assets are still limited to incorrect geometries and blurry textures. To address this problem, we present a multi-view based refine method, named Magic-Boost, to further refine the generation results. In detail, we first propose a novel multi-view conditioned diffusion model which extracts 3d prior from the synthesized multi-view images to synthesize high-fidelity novel view images and then introduce a novel iterative-update strategy to adopt it to provide precise guidance to refine the coarse generated results through a fast optimization process. Conditioned on the strong 3d priors extracted from the synthesized multi-view images, Magic-Boost is capable of providing precise optimization guidance that well aligns with the coarse generated 3D assets, enriching the local detail in both geometry and texture within a short time ($\sim15$min). Extensive experiments show Magic-Boost greatly enhances the coarse generated inputs, generates high-quality 3D assets with rich geometric and textural details. (Project Page: https://magic-research.github.io/magic-boost/)

cs.CV↗

arXiv2024

Optimization-Based Image Reconstruction Regularized with Inter-Spectral Structural Similarity for Limited-Angle Dual-Energy Cone-Beam CT

Background: Limited-angle (LA) dual-energy (DE) cone-beam CT (CBCT) is considered as a potential solution to achieve fast and low-dose DE imaging on current CBCT scanners without hardware modification. However, its clinical implementations are hindered by the challenging image reconstruction from LA projections. While optimization-based and deep learning-based methods have been proposed for image reconstruction, their utilization is limited by the requirement for X-ray spectra measurement or paired datasets for model training. Purpose: This work aims to facilitate the clinical applications of fast and low-dose DECBCT by developing a practical solution for image reconstruction in LA-DECBCT. Methods: An inter-spectral structural similarity-based regularization was integrated into the iterative image reconstruction in LA-DECBCT. By enforcing the similarity between the DE images, LA artifacts were efficiently reduced in the reconstructed DECBCT images. The proposed method was evaluated using four physical phantoms and three digital phantoms, demonstrating its efficacy in quantitative DECBCT imaging. Conclusions: The proposed method achieves accurate image reconstruction without the need for X-ray spectra measurement for optimization or paired datasets for model training, showing great practical value in clinical implementations of LA-DECBCT.

physics.med-ph↗

arXiv2024

Photon-Counting CT in Cancer Radiotherapy: Technological Advances and Clinical Benefits

Photon-counting computed tomography (PCCT) marks a significant advancement over conventional energy-integrating detector (EID) CT systems. This review highlights PCCT's superior spatial and contrast resolution, reduced radiation dose, and multi-energy imaging capabilities, which address key challenges in radiotherapy, such as accurate tumor delineation, precise dose calculation, and treatment response monitoring. PCCT's improved anatomical clarity enhances tumor targeting while minimizing damage to surrounding healthy tissues. Additionally, metal artifact reduction (MAR) and quantitative imaging capabilities optimize workflows, enabling adaptive radiotherapy and radiomics-driven personalized treatment. Emerging clinical applications in brachytherapy and radiopharmaceutical therapy (RPT) show promising outcomes, although challenges like high costs and limited software integration remain. With advancements in artificial intelligence (AI) and dedicated radiotherapy packages, PCCT is poised to transform precision, safety, and efficacy in cancer radiotherapy, marking it as a pivotal technology for future clinical practice.

physics.med-ph↗

arXiv2024

Customer Lifetime Value Prediction with Uncertainty Estimation Using Monte Carlo Dropout

Accurately predicting customer Lifetime Value (LTV) is crucial for companies to optimize their revenue strategies. Traditional deep learning models for LTV prediction are effective but typically provide only point estimates and fail to capture model uncertainty in modeling user behaviors. To address this limitation, we propose a novel approach that enhances the architecture of purely neural network models by incorporating the Monte Carlo Dropout (MCD) framework. We benchmarked the proposed method using data from one of the most downloaded mobile games in the world, and demonstrated a substantial improvement in predictive Top 5\% Mean Absolute Percentage Error compared to existing state-of-the-art methods. Additionally, our approach provides confidence metric as an extra dimension for performance evaluation across various neural network models, facilitating more informed business decisions.

cs.LG↗

arXiv2024

Topkima-Former: Low-energy, Low-Latency Inference for Transformers using top-k In-memory ADC

Transformer model has gained prominence as a popular deep neural network architecture for neural language processing (NLP) and computer vision (CV) applications. However, the extensive use of nonlinear operations, like softmax, poses a performance bottleneck during transformer inference and comprises up to 40% of the total latency. Hence, we propose innovations at the circuit, architecture, and algorithm levels to accelerate the transformer. At the circuit level, we propose topkima-combining top-k activation selection with in-memory ADC (IMA) to implement a low-energy and low-latency softmax without any sorting latency. Only the k largest activations are sent to the softmax calculation block, reducing the huge computational cost of softmax. Using a modified training scheme with top-k only in the forward pass, experimental results demonstrate only a 0.4% to 1.2% reduction in accuracy across ViT, distilBERT, and BERT-base models when evaluated on CIFAR-10, CIFAR-100, and SQuAD datasets with k=5. At the architecture level, an improved scale-free technique is introduced to reduce the computational cost of attention. The combined system, dubbed Topkima-Former, enhances 1.8x-84x speedup and 1.3x-35x energy efficiency (EE) over prior In-memory computing (IMC) accelerators. Compared to a conventional softmax macro and a digital top-k (Dtopk) softmax macro, our proposed tokima softmax macro achieves about 15x and 8x faster speed respectively.

cs.AR↗

arXiv2024

Self-Supervised Adversarial Diffusion Models for Fast MRI Reconstruction

Purpose: To propose a self-supervised deep learning-based compressed sensing MRI (DL-based CS-MRI) method named "Adaptive Self-Supervised Consistency Guided Diffusion Model (ASSCGD)" to accelerate data acquisition without requiring fully sampled datasets. Materials and Methods: We used the fastMRI multi-coil brain axial T2-weighted (T2-w) dataset from 1,376 cases and single-coil brain quantitative magnetization prepared 2 rapid acquisition gradient echoes (MP2RAGE) T1 maps from 318 cases to train and test our model. Robustness against domain shift was evaluated using two out-of-distribution (OOD) datasets: multi-coil brain axial postcontrast T1 -weighted (T1c) dataset from 50 cases and axial T1-weighted (T1-w) dataset from 50 patients. Data were retrospectively subsampled at acceleration rates R in {2x, 4x, 8x}. ASSCGD partitions a random sampling pattern into two disjoint sets, ensuring data consistency during training. We compared our method with ReconFormer Transformer and SS-MRI, assessing performance using normalized mean squared error (NMSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM). Statistical tests included one-way analysis of variance (ANOVA) and multi-comparison Tukey's Honesty Significant Difference (HSD) tests. Results: ASSCGD preserved fine structures and brain abnormalities visually better than comparative methods at R = 8x for both multi-coil and single-coil datasets. It achieved the lowest NMSE at R in {4x, 8x}, and the highest PSNR and SSIM values at all acceleration rates for the multi-coil dataset. Similar trends were observed for the single-coil dataset, though SSIM values were comparable to ReconFormer at R in {2x, 8x}. These results were further confirmed by the voxel-wise correlation scatter plots. OOD results showed significant (p << 10^-5 ) improvements in undersampled image quality after reconstruction.

eess.IV↗

arXiv2024

HELENE: Hessian Layer-wise Clipping and Gradient Annealing for Accelerating Fine-tuning LLM with Zeroth-order Optimization

Fine-tuning large language models (LLMs) poses significant memory challenges, as the back-propagation process demands extensive resources, especially with growing model sizes. Recent work, MeZO, addresses this issue using a zeroth-order (ZO) optimization method, which reduces memory consumption by matching the usage to the inference phase. However, MeZO experiences slow convergence due to varying curvatures across model parameters. To overcome this limitation, we introduce HELENE, a novel scalable and memory-efficient optimizer that integrates annealed A-GNB gradients with a diagonal Hessian estimation and layer-wise clipping, serving as a second-order pre-conditioner. This combination allows for faster and more stable convergence. Our theoretical analysis demonstrates that HELENE improves convergence rates, particularly for models with heterogeneous layer dimensions, by reducing the dependency on the total parameter space dimension. Instead, the method scales with the largest layer dimension, making it highly suitable for modern LLM architectures. Experimental results on RoBERTa-large and OPT-1.3B across multiple tasks show that HELENE achieves up to a 20x speedup compared to MeZO, with average accuracy improvements of 1.5%. Furthermore, HELENE remains compatible with both full parameter tuning and parameter-efficient fine-tuning (PEFT), outperforming several state-of-the-art optimizers. The codes will be released after reviewing.

cs.LG↗