arXiv subjects

Linze Li

Publications and source records attributed to Linze Li.

At least 19 recordsLinked to original sources

arXiv2026

EpaCache: Error-Propagation-Aware Caching for Accelerating Diffusion-Based Visual Generation

Diffusion-based visual generative models deliver strong image and video synthesis quality but incur high inference costs because sequential samplers repeatedly evaluate large networks. Caching-based methods reduce inference latency by reusing intermediate computations across adjacent timesteps. However, existing cache controllers rely primarily on local temporal variation and overlook the trajectory-level consequences of cache reuse. We introduce Error-Propagation-Aware Cache (EpaCache), a training-free caching policy that adaptively allocates the reuse budget on timesteps with lower downstream impact. Experiments on image and video synthesis models demonstrate that EpaCache consistently improves the latency--fidelity trade-off over existing caching methods. On FLUX.1-dev, EpaCache outperforms the prior state-of-the-art caching method in both latency and fidelity, reducing inference time from $11.7$ s to $11.3$ s while improving PSNR from $21.4$ to $22.8$. On HunyuanVideo, EpaCache achieves a $2.63\times$ speedup over uncached inference and improves SSIM from $0.891$ to $0.905$ over the prior state-of-the-art method at matched latency.

cs.AI↗

arXiv2026

SPARE: Structural Parameter-Free Affinity Regularization for Flow Matching

Denoising diffusion transformers achieve strong generation quality but converge slowly during training. Regularizing their internal representations has emerged as an effective accelerator, yet existing methods split into two families with complementary costs. Target-based methods strengthen representations by aligning them to external features, which requires an external encoder and a learnable projection head to bridge feature spaces. Target-free methods hold no reference at all, and can only repel the model's own features across samples or layers, discarding whatever structure the data contains. Prior work suggests that spatial structure, rather than global semantics, drives the gains of alignment. We therefore ask whether such structure can serve as a target directly, and whether it exists not only within an image but across images. Our key insight is that the clean data latent already carries this structure in the relations among its tokens, where a relation is the similarity between two tokens, a single scalar comparable across feature spaces without a projection head. We propose Structural Parameter-free Affinity Regularization (SPARE), a regularizer that matches the pairwise affinities of intermediate tokens to those of the clean latents. To exploit this structure fully, SPARE extends the matching to token pairs across images, precisely the pairs that prior target-free methods repel by default, and calibrates both relation types with a single learning objective. On ImageNet $256 \times 256$ with SiT backbones under matched 400K-iteration budgets, SPARE adds no encoder, head, or parameters and only 0.08 GB of training memory, yet attains the lowest FID among parameter-free regularizers in every tested setting, recovers 37 to 54\% of REPA's FID reduction, and improves over REPA when combined with it, reaching FID 1.90 under classifier-free guidance at 1M iterations.

cs.CV↗

arXiv2026

The Velocity Deficit: Initial Energy Injection for Flow Matching

While Flow Matching theoretically guarantees constant-velocity trajectories, we identify a critical breakdown in high-dimensional practice: the Velocity Deficit. We show that the MSE objective systematically underestimates velocity magnitude, causing generated samples to fail to reach the data manifold-a phenomenon we term Integration Lag. To rectify this, we propose Initial Energy Injection, instantiated via two complementary methods: the training-based Magnitude-Aware Flow Matching (MAFM) and the training-free Scale Schedule Corrector (SSC). Both are grounded in our discovery of a crucial asymmetry: velocity contraction causes harmful kinetic stagnation at the trajectory's start, yet acts as a beneficial denoising mechanism at its end. Empirically, SSC yields significant efficiency gains with zero retraining and just one line of code. On ImageNet-1k (256x256), it improves FID by 44.6% (from 13.68 to 7.58) and achieves a 5x speedup, enabling a 50-step generator (FID 7.58) to beat a 250-step baseline (FID 8.65). Furthermore, our methods generalize to Text-to-Image tasks and high-resolution generation, improving FID on MS-COCO by ~22%.

cs.CV↗

arXiv2026

NTIRE 2026 Challenge on Video Saliency Prediction: Methods and Results

This paper presents an overview of the NTIRE 2026 Challenge on Video Saliency Prediction. The goal of the challenge participants was to develop automatic saliency map prediction methods for the provided video sequences. The novel dataset of 2,000 diverse videos with an open license was prepared for this challenge. The fixations and corresponding saliency maps were collected using crowdsourced mouse tracking and contain viewing data from over 5,000 assessors. Evaluation was performed on a subset of 800 test videos using generally accepted quality metrics. The challenge attracted over 20 teams making submissions, and 7 teams passed the final phase with code review. All data used in this challenge is made publicly available - https://github.com/msu-video-group/NTIRE26_Saliency_Prediction.

cs.CV↗

arXiv2026

Stable Spike: Dual Consistency Optimization via Bitwise AND Operations for Spiking Neural Networks

Although the temporal spike dynamics of spiking neural networks (SNNs) enable low-power temporal pattern capture capabilities, they also incur inherent inconsistencies that severely compromise representation. In this paper, we perform dual consistency optimization via Stable Spike to mitigate this problem, thereby improving the recognition performance of SNNs. With the hardware-friendly ``AND" bit operation, we efficiently decouple the stable spike skeleton from the multi-timestep spike maps, thereby capturing critical semantics while reducing inconsistencies from variable noise spikes. Enforcing the unstable spike maps to converge to the stable spike skeleton significantly improves the inherent consistency across timesteps. Furthermore, we inject amplitude-aware spike noise into the stable spike skeleton to diversify the representations while preserving consistent semantics. The SNN is encouraged to produce perturbation-consistent predictions, thereby contributing to generalization. Extensive experiments across multiple architectures and datasets validate the effectiveness and versatility of our method. In particular, our method significantly advances neuromorphic object recognition under ultra-low latency, improving accuracy by up to 8.33\%. This will help unlock the full power consumption and speed potential of SNNs.

cs.NE↗

arXiv2025

All-optical control and multiplexed readout of multiple superconducting qubits

Superconducting quantum circuits operate at millikelvin temperatures, typically requiring independent microwave cables for each qubit for connecting room-temperature control and readout electronics. However, scaling to large-scale processors hosting hundreds of qubits faces a severe input/output (I/O) bottleneck, as the dense cable arrays impose prohibitive constraints on physical footprint, thermal load, wiring complexity, and cost. Here we demonstrate a complete optical I/O architecture for superconducting quantum circuits, in which all control and readout signals are transmitted exclusively via optical photons. Employing a broadband traveling-wave Brillouin microwave-to-optical transducer, we achieve simultaneous frequency-multiplexed optical readout of two qubits. Combined with fiber-integrated photodiode arrays for control signal delivery, this closed-loop optical I/O introduces no measurable degradation to qubit coherence times, with an optically driven single-qubit gate fidelity showing only a 0.19% reduction relative to standard microwave operation. These results establish optical interconnects as a viable path toward large-scale superconducting quantum processors, and open the possibility of networking multiple superconducting quantum computers housed in separate dilution refrigerators through a centralized room-temperature control infrastructure.

quant-ph↗

arXiv2025

Scalable Optical Links for Controlling Bosonic Quantum Processors

Superconducting quantum computing has the potential to revolutionize computational capabilities. However, scaling up large quantum processors is limited by the cumbersome and heat-conductive electronic cables that connect room-temperature control electronics to quantum processors, leading to significant signal attenuation. Optical fibers provide a promising solution, but their use has been restricted to controlling simple two-level quantum systems over short distances. Here, we demonstrate optical control of a bosonic quantum processor, achieving universal operations on the joint Hilbert space of a transmon qubit and a storage cavity. Using an array of cryogenic fiber-integrated uni-traveling-carrier photodiodes, we prepare Fock states containing up to ten photons. Additionally, remote control of bosonic modes over a transmission distance of 15 km has been achieved, with fidelities exceeding 95%. The combination of high-dimensional quantum control, multi-channel operation, and long-distance transmission addresses the key requirements for scaling superconducting quantum computers and enables architectures for distributed quantum data centers.

quant-ph↗

arXiv2025

LEDiT: Your Length-Extrapolatable Diffusion Transformer without Positional Encoding

Diffusion transformers (DiTs) struggle to generate images at resolutions higher than their training resolutions. The primary obstacle is that the explicit positional encodings(PE), such as RoPE, need extrapolating to unseen positions which degrades performance when the inference resolution differs from training. In this paper, We propose a Length-Extrapolatable Diffusion Transformer~(LEDiT) to overcome this limitation. LEDiT needs no explicit PEs, thereby avoiding PE extrapolation. The key innovation of LEDiT lies in the use of causal attention. We demonstrate that causal attention can implicitly encode global positional information and show that such information facilitates extrapolation. We further introduce a locality enhancement module, which captures fine-grained local information to complement the global coarse-grained position information encoded by causal attention. Experimental results on both conditional and text-to-image generation tasks demonstrate that LEDiT supports up to 4x resolution scaling (e.g., from 256x256 to 512x512), achieving better image quality compared to the state-of-the-art length extrapolation methods. We believe that LEDiT marks a departure from the standard RoPE-based methods and offers a promising insight into length extrapolation. Project page: https://shenzhang2145.github.io/ledit/

cs.CV↗

arXiv2025

A chip-based optoelectronic-oscillator frequency comb

Microresonator-based Kerr frequency combs ("Kerr microcombs") constitute chip-scale frequency combs of broad spectral bandwidth and repetition rate ranging from gigahertz to terahertz. An appealing application exploiting microcombs' coherence and large repetition rate is microwave and millimeter-wave generation. Latest endeavor applying two-point optical frequency division (OFD) on photonic-chip-based microcombs has created microwaves with exceptionally low phase noise. Nevertheless, microcomb-based OFD still requires extensive active locking, additional lasers, and external RF or microwave sources, as well as sophisticated initiation. Here we demonstrate a simple and entirely passive (no active locking) architecture, which incorporates an optoelectronic oscillator (OEO) and symphonizes a coherent microcomb and a low-noise microwave spontaneously. Our OEO microcomb leverages state-of-the-art integrated chip devices including a high-power DFB laser, a broadband silicon Mach-Zehnder modulator, an ultralow-loss silicon nitride microresonator, and a high-speed photodetector. Each can be manufactured in large volume with low cost and high yield using established CMOS and III-V foundries. Our system synergizes a microcomb of 10.7 GHz repetition rate and an X-band microwave with phase noise of $-$97/$-$126/$-$130 dBc/Hz at 1/10/100 kHz Fourier frequency offset, yet does not demand active locking, additional lasers, and external RF or microwave sources. With potential to be fully integrated, our OEO microcomb can become an invaluable technology and building block for microwave photonics, radio-over-fiber, and optical communication.

physics.optics↗

arXiv2025

Optimizing Knowledge Distillation in Transformers: Enabling Multi-Head Attention without Alignment Barriers

Knowledge distillation (KD) in transformers often faces challenges due to misalignment in the number of attention heads between teacher and student models. Existing methods either require identical head counts or introduce projectors to bridge dimensional gaps, limiting flexibility and efficiency. We propose Squeezing-Heads Distillation (SHD), a novel approach that enables seamless knowledge transfer between models with varying head counts by compressing multi-head attention maps via efficient linear approximation. Unlike prior work, SHD eliminates alignment barriers without additional parameters or architectural modifications. Our method dynamically approximates the combined effect of multiple teacher heads into fewer student heads, preserving fine-grained attention patterns while reducing redundancy. Experiments across language (LLaMA, GPT) and vision (DiT, MDT) generative and vision (DeiT) discriminative tasks demonstrate SHD's effectiveness: it outperforms logit-based and feature-alignment KD baselines, achieving state-of-the-art results in image classification, image generation language fine-tuning, and language pre-training. The key innovations of flexible head compression, projector-free design, and linear-time complexity make SHD a versatile and scalable solution for distilling modern transformers. This work bridges a critical gap in KD, enabling efficient deployment of compact models without compromising performance.

cs.CV↗

arXiv2025

Follow the Leader: Enhancing Systematic Trend-Following Using Network Momentum

We present a systematic, trend-following strategy, applied to commodity futures markets, that combines univariate trend indicators with cross-sectional trend indicators that capture so-called {\em momentum spillover}, which can occur when there is a lead-lag relationship between the trending behaviour of different markets. Our strategy utilises two methods for detecting lead-lag relationships, with a method for computing {\em network momentum}, to produce a novel trend-following indicator. We use our new trend indicator to construct a portfolio whose performance we compare to a baseline model which uses only univariate indicators, and demonstrate statistically significant improvements in Sharpe ratio, skewness of returns, and downside performance, using synthetic bootstrapped data samples taken from time-series of actual prices.

q-fin.TR↗

arXiv2025

Modified uni-travelling-carrier photodiodes with 206 GHz bandwidth and 0.81 A/W external responsivity

The accelerating demand for wireless communication necessitates wideband, energy-efficient photonic sub-terahertz (sub-THz) sources to enable ultra-fast data transfer. However, as critical components for THz photonic mixing, photodiodes (PDs) face a fundamental trade-off between quantum efficiency and bandwidth, presenting a major obstacle to achieving high-speed performance with high optoelectronic conversion efficiency. Here, we overcome this challenge by demonstrating an InP-based, waveguide-integrated modified uni-traveling carrier photodiode (MUTC-PD) with bandwidth exceeding 200 GHz and a bandwidth-efficiency product (BEP) surpassing 130 GHz. Incorporating a spot-size converter (SSC) together with optimized electric-field distribution, balanced carrier transport, and minimized parasitic capacitance, the device achieves a 3-dB bandwidth of 206 GHz and an external responsivity of 0.81 A/W, setting a new BEP benchmark. Packaged with WR-5.1 waveguide output, it delivers radio-frequency (RF) power exceeding -5 dBm across the 127-185 GHz frequency range. As a proof of concept, we achieved a wireless transmission of 54 meters with a single-line rate of up to 120 Gbps, leveraging photonics-aided technology without requiring a low-noise amplifier (LNA). This work establishes a pathway to significantly enhance optical power budgets and reduce energy consumption, presenting a transformative step toward high-bandwidth, high-efficiency sub-THz communication systems and next-generation wireless networks.

physics.app-ph↗

arXiv2024

MAT: Multi-Range Attention Transformer for Efficient Image Super-Resolution

Image super-resolution (SR) has significantly advanced through the adoption of Transformer architectures. However, conventional techniques aimed at enlarging the self-attention window to capture broader contexts come with inherent drawbacks, especially the significantly increased computational demands. Moreover, the feature perception within a fixed-size window of existing models restricts the effective receptive field (ERF) and the intermediate feature diversity. We demonstrate that a flexible integration of attention across diverse spatial extents can yield significant performance enhancements. In line with this insight, we introduce Multi-Range Attention Transformer (MAT) for SR tasks. MAT leverages the computational advantages inherent in dilation operation, in conjunction with self-attention mechanism, to facilitate both multi-range attention (MA) and sparse multi-range attention (SMA), enabling efficient capture of both regional and sparse global features. Combined with local feature extraction, MAT adeptly capture dependencies across various spatial ranges, improving the diversity and efficacy of its feature representations. We also introduce the MSConvStar module, which augments the model's ability for multi-range representation learning. Comprehensive experiments show that our MAT exhibits superior performance to existing state-of-the-art SR models with remarkable efficiency (~3.3 faster than SRFormer-light).

cs.CV↗

arXiv2024

MegActor-$\Sigma$: Unlocking Flexible Mixed-Modal Control in Portrait Animation with Diffusion Transformer

Diffusion models have demonstrated superior performance in the field of portrait animation. However, current approaches relied on either visual or audio modality to control character movements, failing to exploit the potential of mixed-modal control. This challenge arises from the difficulty in balancing the weak control strength of audio modality and the strong control strength of visual modality. To address this issue, we introduce MegActor-$\Sigma$: a mixed-modal conditional diffusion transformer (DiT), which can flexibly inject audio and visual modality control signals into portrait animation. Specifically, we make substantial advancements over its predecessor, MegActor, by leveraging the promising model structure of DiT and integrating audio and visual conditions through advanced modules within the DiT framework. To further achieve flexible combinations of mixed-modal control signals, we propose a ``Modality Decoupling Control" training strategy to balance the control strength between visual and audio modalities, along with the ``Amplitude Adjustment" inference strategy to freely regulate the motion amplitude of each modality. Finally, to facilitate extensive studies in this field, we design several dataset evaluation metrics to filter out public datasets and solely use this filtered dataset to train MegActor-$\Sigma$. Extensive experiments demonstrate the superiority of our approach in generating vivid portrait animations, outperforming previous methods trained on private dataset.

cs.CV↗

arXiv2024

Efficient Single Image Super-Resolution with Entropy Attention and Receptive Field Augmentation

Transformer-based deep models for single image super-resolution (SISR) have greatly improved the performance of lightweight SISR tasks in recent years. However, they often suffer from heavy computational burden and slow inference due to the complex calculation of multi-head self-attention (MSA), seriously hindering their practical application and deployment. In this work, we present an efficient SR model to mitigate the dilemma between model efficiency and SR performance, which is dubbed Entropy Attention and Receptive Field Augmentation network (EARFA), and composed of a novel entropy attention (EA) and a shifting large kernel attention (SLKA). From the perspective of information theory, EA increases the entropy of intermediate features conditioned on a Gaussian distribution, providing more informative input for subsequent reasoning. On the other hand, SLKA extends the receptive field of SR models with the assistance of channel shifting, which also favors to boost the diversity of hierarchical features. Since the implementation of EA and SLKA does not involve complex computations (such as extensive matrix multiplications), the proposed method can achieve faster nonlinear inference than Transformer-based SR models while maintaining better SR performance. Extensive experiments show that the proposed model can significantly reduce the delay of model inference while achieving the SR performance comparable with other advanced models.

eess.IV↗

arXiv2024

Large Kernel Distillation Network for Efficient Single Image Super-Resolution

Efficient and lightweight single-image super-resolution (SISR) has achieved remarkable performance in recent years. One effective approach is the use of large kernel designs, which have been shown to improve the performance of SISR models while reducing their computational requirements. However, current state-of-the-art (SOTA) models still face problems such as high computational costs. To address these issues, we propose the Large Kernel Distillation Network (LKDN) in this paper. Our approach simplifies the model structure and introduces more efficient attention modules to reduce computational costs while also improving performance. Specifically, we employ the reparameterization technique to enhance model performance without adding extra cost. We also introduce a new optimizer from other tasks to SISR, which improves training speed and performance. Our experimental results demonstrate that LKDN outperforms existing lightweight SR methods and achieves SOTA performance.

eess.IV↗

arXiv2024

C2C: Component-to-Composition Learning for Zero-Shot Compositional Action Recognition

Compositional actions consist of dynamic (verbs) and static (objects) concepts. Humans can easily recognize unseen compositions using the learned concepts. For machines, solving such a problem requires a model to recognize unseen actions composed of previously observed verbs and objects, thus requiring so-called compositional generalization ability. To facilitate this research, we propose a novel Zero-Shot Compositional Action Recognition (ZS-CAR) task. For evaluating the task, we construct a new benchmark, Something-composition (Sth-com), based on the widely used Something-Something V2 dataset. We also propose a novel Component-to-Composition (C2C) learning method to solve the new ZS-CAR task. C2C includes an independent component learning module and a composition inference module. Last, we devise an enhanced training strategy to address the challenges of component variations between seen and unseen compositions and to handle the subtle balance between learning seen and unseen actions. The experimental results demonstrate that the proposed framework significantly surpasses the existing compositional generalization methods and sets a new state-of-the-art. The new Sth-com benchmark and code are available at https://github.com/RongchangLi/ZSCAR_C2C.

cs.CV↗

arXiv2024

MegActor: Harness the Power of Raw Video for Vivid Portrait Animation

Despite raw driving videos contain richer information on facial expressions than intermediate representations such as landmarks in the field of portrait animation, they are seldom the subject of research. This is due to two challenges inherent in portrait animation driven with raw videos: 1) significant identity leakage; 2) Irrelevant background and facial details such as wrinkles degrade performance. To harnesses the power of the raw videos for vivid portrait animation, we proposed a pioneering conditional diffusion model named as MegActor. First, we introduced a synthetic data generation framework for creating videos with consistent motion and expressions but inconsistent IDs to mitigate the issue of ID leakage. Second, we segmented the foreground and background of the reference image and employed CLIP to encode the background details. This encoded information is then integrated into the network via a text embedding module, thereby ensuring the stability of the background. Finally, we further style transfer the appearance of the reference image to the driving video to eliminate the influence of facial details in the driving videos. Our final model was trained solely on public datasets, achieving results comparable to commercial models. We hope this will help the open-source community.The code is available at https://github.com/megvii-research/MegFaceAnimate.

cs.CV↗