arXiv subjects

Zexian Li

Publications and source records attributed to Zexian Li.

15 recordsLinked to original sources

arXiv2026

Beyond Point-Wise Matching: Structural Representation Alignment for Accelerating Diffusion Transformers

Recent advances in Diffusion Transformers (DiTs) demonstrate that aligning noisy latent states with well-trained semantic features-as pioneered by Representation Alignment (REPA)-can substantially accelerate training and improve generation fidelity. Subsequent analysis(e.g., iREPA) suggests that these gains arise primarily from transferring spatial structure contained in pre-trained vision representations. However, mostly existing alignment methods employ point-wise matching objectives or rely on implicit architectural tweaks, which fail to explicitly model the spatial relational geometry inherent in vision foundation models. We argue that such element-wise supervision is insufficient to capture the rich spatial topology of visual representations, and that effective alignment for generation should instead be formulated as an explicit structural constraint. To this end, we propose sREPA, a structural REPresentation Alignment framework to enforce consistency in the relational geometry of feature maps, rather than merely matching individual feature points. By encouraging the model to internalize holistic spatial layouts and structural correlations from pre-trained features, sREPA achieves faster and more stable convergence, along with improved sample quality, compared to state-of-the-art alignment strategies. Our code and models will be released.

cs.CV↗

arXiv2026

Edit-GRPO: A Locality-Preserving Policy Optimization Framework for Image Editing

A fundamental challenge in image editing lies in preserving spatial locality: edits should improve targeted content without inadvertently altering surrounding regions. However, most optimization-based editing approaches treat images as holistic entities, causing global policy updates that undermine locality and introduce undesired context changes. We observe that this issue stems from a mismatch between localized editing intent and globally applied optimization signals. Motivated by this insight, we propose Edit-GRPO, preserving Locality while optimizing image editing, a locality-preserving policy optimization framework that explicitly decouples editing and preservation objectives. By assigning region-specific optimization signals to edit and non-edit areas, Edit-GRPO aligns policy updates with the spatial structure of editing tasks, enabling localized improvements while maintaining global visual coherence. This design effectively suppresses common artifacts such as context distortion and boundary inconsistency. Extensive experiments across diverse image editing scenarios demonstrate that Edit-GRPO significantly improves locality preservation while maintaining strong editing performance compared to existing optimization-based methods, validating the generality and effectiveness of the proposed framework.

cs.CV↗

arXiv2026

AdvDMD: Adversarial Reward Meets DMD For High-Quality Few-Step Generation

Diffusion models offer superior generation quality at the expense of extensive sampling steps. Distillation methods, with Distribution Matching Distillation (DMD) as a popular example, can mitigate this issue, but performance degradation remains pronounced when sampling steps are limited. Reinforcement learning (RL) has been leveraged to improve the few-step generation quality during distillation, with the potential to even surpass the performance of the teacher model. However, existing approaches are combinatorial in nature, merely integrating an RL process with the distillation process, which introduces unnecessary complexities. To address this gap, we propose AdvDMD, a method that seamlessly unifies DMD distillation and RL. Specifically, AdvDMD employs the adversarially trained discriminator from DMD2 as the reward model, which assigns low scores to generated images and high scores to real ones. It is trained on both intermediate and final states of the denoising process and updated online with the distilled model, enabling a holistic supervision of the sampling trajectories and mitigating reward hacking. We adopt a unified SDE backward simulation and a different training schedule for DMD and RL to enable a more stable and efficient training. Experimental results demonstrate that the 4-step AdvDMD outperforms the original 40-step model for SD3.5 on DPG-Bench, while achieving significant performance gains for SD3 on the GenEval. On Qwen-Image, our 2-step AdvDMD achieves superior performance over TwinFlow.

cs.CV↗

arXiv2026

Reducing C-NOT Counts for State Preparation and Block Encoding via Diagonal Matrix Migration

Quantum state preparation and block encoding are versatile and practical input models for quantum algorithms in scientific computing. The circuit complexity of state preparation and block encoding frequently dominates the end-to-end gate complexity of quantum algorithms. We give algorithms with lower C-NOT counts for both the state preparation and block encoding. For a general $n$-qubit state, we improve the C-NOT count of the Plesch-Brukner algorithm (2011) from $(23/24)\times 2^n$ to $(11/12)\times2^n$. For block encoding, our single-ancilla protocol for $2^{n-1}\times 2^{n-1}$ matrices uses the spectral norm as subnormalization and achieves a C-NOT count leading term $(11/48)\times 4^n$. Further optimization is performed for low-rank matrices, which frequently arise in practical applications. Specifically, we achieve the C-NOT count leading term $(2^{\lceil\log_{2}K\rceil}+(11/12))\times 2^n$ for a rank-$K$ matrix. This is the first quantum algorithm that encodes matrices using the optimal normalization factor while also allowing the C-NOT count to be adjusted according to the matrix rank. Our approach builds upon the recursive Block-ZXZ decomposition from Krol et al. (2024) and introduces a diagonal matrix migration technique based on the commutativity of the diagonal matrix and the uniformly controlled rotation about the $z$-axis to minimize the use of C-NOT gates.

quant-ph↗

arXiv2025

Differentiable Solver Search for Fast Diffusion Sampling

Diffusion models have demonstrated remarkable generation quality but at the cost of numerous function evaluations. Recently, advanced ODE-based solvers have been developed to mitigate the substantial computational demands of reverse-diffusion solving under limited sampling steps. However, these solvers, heavily inspired by Adams-like multistep methods, rely solely on t-related Lagrange interpolation. We show that t-related Lagrange interpolation is suboptimal for diffusion model and reveal a compact search space comprised of time steps and solver coefficients. Building on our analysis, we propose a novel differentiable solver search algorithm to identify more optimal solver. Equipped with the searched solver, rectified-flow models, e.g., SiT-XL/2 and FlowDCN-XL/2, achieve FID scores of 2.40 and 2.35, respectively, on ImageNet256 with only 10 steps. Meanwhile, DDPM model, DiT-XL/2, reaches a FID score of 2.33 with only 10 steps. Notably, our searched solver outperforms traditional solvers by a significant margin. Moreover, our searched solver demonstrates generality across various model architectures, resolutions, and model sizes.

cs.CV↗

arXiv2025

Binary Tree Block Encoding of Classical Matrix

Block-encoding is a critical subroutine in quantum computing, enabling the transformation of classical data into a matrix representation within a quantum circuit. The resource trade-offs in simulating a block-encoding can be quantified by the circuit size, the normalization factor, and the time and space complexity of parameter computation. Previous studies have primarily focused either on the time and memory complexity of computing the parameters, or on the circuit size and normalization factor in isolation, often neglecting the balance between these trade-offs. In early fault-tolerant quantum computers, the number of qubits is limited. For a classical matrix of size $2^{n}\times 2^{n}$, our approach not only improves the time of decoupling unitary for block-encoding with time complexity $\mathcal{O}(n2^{2n})$ and memory complexity $\Theta(2^{2n})$ using only a few ancilla qubits, but also demonstrates superior resource trade-offs. Our proposed block-encoding protocol is named Binary Tree Block-encoding (\texttt{BITBLE}). Under the benchmark, \textit{size metric}, defined by the product of the number of gates and the normalization factor, numerical experiments demonstrate the improvement of both resource trade-off and classical computing time efficiency of the \texttt{BITBLE} protocol. The algorithms are all open-source.

quant-ph↗

arXiv2024

FlowDCN: Exploring DCN-like Architectures for Fast Image Generation with Arbitrary Resolution

Arbitrary-resolution image generation still remains a challenging task in AIGC, as it requires handling varying resolutions and aspect ratios while maintaining high visual quality. Existing transformer-based diffusion methods suffer from quadratic computation cost and limited resolution extrapolation capabilities, making them less effective for this task. In this paper, we propose FlowDCN, a purely convolution-based generative model with linear time and memory complexity, that can efficiently generate high-quality images at arbitrary resolutions. Equipped with a new design of learnable group-wise deformable convolution block, our FlowDCN yields higher flexibility and capability to handle different resolutions with a single model. FlowDCN achieves the state-of-the-art 4.30 sFID on $256\times256$ ImageNet Benchmark and comparable resolution extrapolation results, surpassing transformer-based counterparts in terms of convergence speed (only $\frac{1}{5}$ images), visual quality, parameters ($8\%$ reduction) and FLOPs ($20\%$ reduction). We believe FlowDCN offers a promising solution to scalable and flexible image synthesis.

cs.CV↗

arXiv2024

Dictionary-based Block Encoding of Sparse Matrices with Low Subnormalization and Circuit Depth

Block encoding severs as an important data input model in quantum algorithms, enabling quantum computers to simulate non-unitary operators effectively. In this paper, we propose an efficient block-encoding protocol for sparse matrices based on a novel data structure, called the dictionary data structure, which classifies all non-zero elements according to their values and indices. Non-zero elements with the same values, lacking common column and row indices, belong to the same classification in our block-encoding protocol's dictionary. When compiled into the \{\rm U(2), CNOT\} gate set, the protocol queries a $2^n \times 2^n$ sparse matrix with $s$ non-zero elements at a circuit depth of $\mathcal{O}(\log(ns))$, utilizing $\mathcal{O}(n^2s)$ ancillary qubits. This offers an exponential improvement in circuit depth relative to the number of system qubits, compared to existing methods~\cite{clader2022quantum,zhang2024circuit} with a circuit depth of $\mathcal{O}(n)$. Moreover, in our protocol, the subnormalization, a scaled factor that influences the measurement probability of ancillary qubits, is minimized to $\sum_{l=0}^{s_0}\vert A_l\vert$, where $s_0$ denotes the number of classifications in the dictionary and $A_l$ represents the value of the $l$-th classification. Furthermore, we show that our protocol connects to linear combinations of unitaries (LCU) and the sparse access input model (SAIM). To demonstrate the practical utility of our approach, we provide several applications, including Laplacian matrices in graph problems and discrete differential operators.

quant-ph↗

arXiv2023

G2P-DDM: Generating Sign Pose Sequence from Gloss Sequence with Discrete Diffusion Model

The Sign Language Production (SLP) project aims to automatically translate spoken languages into sign sequences. Our approach focuses on the transformation of sign gloss sequences into their corresponding sign pose sequences (G2P). In this paper, we present a novel solution for this task by converting the continuous pose space generation problem into a discrete sequence generation problem. We introduce the Pose-VQVAE framework, which combines Variational Autoencoders (VAEs) with vector quantization to produce a discrete latent representation for continuous pose sequences. Additionally, we propose the G2P-DDM model, a discrete denoising diffusion architecture for length-varied discrete sequence data, to model the latent prior. To further enhance the quality of pose sequence generation in the discrete space, we present the CodeUnet model to leverage spatial-temporal information. Lastly, we develop a heuristic sequential clustering method to predict variable lengths of pose sequences for corresponding gloss sequences. Our results show that our model outperforms state-of-the-art G2P models on the public SLP evaluation benchmark. For more generated results, please visit our project page: \textcolor{blue}{\url{https://slpdiffusier.github.io/g2p-ddm}}

cs.CV↗

arXiv2022

Group DETR v2: Strong Object Detector with Encoder-Decoder Pretraining

We present a strong object detector with encoder-decoder pretraining and finetuning. Our method, called Group DETR v2, is built upon a vision transformer encoder ViT-Huge~\cite{dosovitskiy2020image}, a DETR variant DINO~\cite{zhang2022dino}, and an efficient DETR training method Group DETR~\cite{chen2022group}. The training process consists of self-supervised pretraining and finetuning a ViT-Huge encoder on ImageNet-1K, pretraining the detector on Object365, and finally finetuning it on COCO. Group DETR v2 achieves $\textbf{64.5}$ mAP on COCO test-dev, and establishes a new SoTA on the COCO leaderboard https://paperswithcode.com/sota/object-detection-on-coco

cs.CV↗

arXiv2022

ViTKD: Practical Guidelines for ViT feature knowledge distillation

Knowledge Distillation (KD) for Convolutional Neural Network (CNN) is extensively studied as a way to boost the performance of a small model. Recently, Vision Transformer (ViT) has achieved great success on many computer vision tasks and KD for ViT is also desired. However, besides the output logit-based KD, other feature-based KD methods for CNNs cannot be directly applied to ViT due to the huge structure gap. In this paper, we explore the way of feature-based distillation for ViT. Based on the nature of feature maps in ViT, we design a series of controlled experiments and derive three practical guidelines for ViT's feature distillation. Some of our findings are even opposite to the practices in the CNN era. Based on the three guidelines, we propose our feature-based method ViTKD which brings consistent and considerable improvement to the student. On ImageNet-1k, we boost DeiT-Tiny from 74.42% to 76.06%, DeiT-Small from 80.55% to 81.95%, and DeiT-Base from 81.76% to 83.46%. Moreover, ViTKD and the logit-based KD method are complementary and can be applied together directly. This combination can further improve the performance of the student. Specifically, the student DeiT-Tiny, Small, and Base achieve 77.78%, 83.59%, and 85.41%, respectively. The code is available at https://github.com/yzd-v/cls_KD.

cs.CV↗

arXiv2021

ReconfigISP: Reconfigurable Camera Image Processing Pipeline

Image Signal Processor (ISP) is a crucial component in digital cameras that transforms sensor signals into images for us to perceive and understand. Existing ISP designs always adopt a fixed architecture, e.g., several sequential modules connected in a rigid order. Such a fixed ISP architecture may be suboptimal for real-world applications, where camera sensors, scenes and tasks are diverse. In this study, we propose a novel Reconfigurable ISP (ReconfigISP) whose architecture and parameters can be automatically tailored to specific data and tasks. In particular, we implement several ISP modules, and enable backpropagation for each module by training a differentiable proxy, hence allowing us to leverage the popular differentiable neural architecture search and effectively search for the optimal ISP architecture. A proxy tuning mechanism is adopted to maintain the accuracy of proxy networks in all cases. Extensive experiments conducted on image restoration and object detection, with different sensors, light conditions and efficiency constraints, validate the effectiveness of ReconfigISP. Only hundreds of parameters need tuning for every task.

eess.IV↗

arXiv2021

MvSR-NAT: Multi-view Subset Regularization for Non-Autoregressive Machine Translation

Conditional masked language models (CMLM) have shown impressive progress in non-autoregressive machine translation (NAT). They learn the conditional translation model by predicting the random masked subset in the target sentence. Based on the CMLM framework, we introduce Multi-view Subset Regularization (MvSR), a novel regularization method to improve the performance of the NAT model. Specifically, MvSR consists of two parts: (1) \textit{shared mask consistency}: we forward the same target with different mask strategies, and encourage the predictions of shared mask positions to be consistent with each other. (2) \textit{model consistency}, we maintain an exponential moving average of the model weights, and enforce the predictions to be consistent between the average model and the online model. Without changing the CMLM-based architecture, our approach achieves remarkable performance on three public benchmarks with 0.36-1.14 BLEU gains over previous NAT models. Moreover, compared with the stronger Transformer baseline, we reduce the gap to 0.01-0.44 BLEU scores on small datasets (WMT16 RO$\leftrightarrow$EN and IWSLT DE$\rightarrow$EN).

cs.CL↗

arXiv2018

5G Control Channel Design for Ultra-Reliable Low-Latency Communications

The fifth generation (5G) of wireless systems holds the promise of supporting a wide range of services with different communication requirements. Ultra-reliable low-latency communications (URLLC) is a generic service that enables mission-critical applications, such as industrial automation, augmented reality, and vehicular communications. URLLC has stringent requirements for reliability and latency of delivering both data and control information. In order to meet these requirements, the Third Generation Partnership Project (3GPP) has been introducing new features to the upcoming releases of the cellular system standards, namely releases 15 and beyond. This article reviews some of these features and introduces new enhancements for designing the control channels to efficiently support the URLLC. In particular, a flexible slot structure is presented as a solution to detect a failure in delivering the control information at an early stage, thereby allowing timely retransmission of the control information. Finally, some remaining challenges and envisioned research directions are discussed for shaping the 5G new radio (NR) as a unified wireless access technology for supporting different services.

eess.SP↗

arXiv2015

Spectrum Allocation for Multi-Operator Device-to-Device Communication

In order to harvest the business potential of device-to-device (D2D) communication, direct communication between devices subscribed to different mobile operators should be supported. This would also support meeting requirements resulting from D2D relevant scenarios, like vehicle-to-vehicle communication. In this paper, we propose to allocate the multi-operator D2D communication over dedicated cellular spectral resources contributed from both operators. Ideally, the operators should negotiate about the amount of spectrum to contribute, without revealing proprietary information to each other and/or to other parties. One possible way to do that is to use the sequence of operators' best responses, i.e., the operators make offers about the amount of spectrum to contribute using a sequential updating procedure until reaching consensus. Besides spectrum allocation, we need a mode selection scheme for the multi-operator D2D users. We use a stochastic geometry framework to capture the impact of mode selection on the distribution of D2D users and assess the performance of the best response iteration algorithm. With the performance metrics considered in the paper, we show that the best response iteration has a unique Nash equilibrium that can be reached from any initial strategy. In general, asymmetric operators would contribute unequal amounts of spectrum for multi-operator D2D communication. Provided that the multi-operator D2D density is not negligible, we show that both operators may experience significant performance gains as compared to the scheme without spectrum sharing.

cs.NI↗