arXiv subjects

Zhifeng Li

Publications and source records attributed to Zhifeng Li.

At least 19 recordsLinked to original sources

arXiv2026

Train What You Deploy: Closing the MLP Reachability Gap in Low-Rank Clone Distillation

A compressed student has two shapes that need not agree: the weight it deploys at inference and the weight family its training can reach. We show that a state-of-the-art weight-inheritance distiller, Low-Rank Clone (LRC), deploys a full-width student MLP but ties training to a teacher-induced slice, leaving 62.5-81.4% of each deployed matrix's independent linear degrees of freedom unreachable-paid for at inference, never trainable. Our principle is one line: train what you deploy. From the identical LRC warm start, we make the training object the entire deployed matrix, with no change in deployed shape, deployed parameter count, or inference FLOPs, via two mergeable realizations (Dense-LRC and CORE-LRC) that both collapse to one deployed weight. This recovers stranded capacity: taking the stronger realization per teacher, +2.36/+2.71/+10.45 Avg9 over matched-budget plain-LRC baselines across three teachers (Llama3.2-3B, Llama3.1-8B, Qwen2.5-3B), with the largest gain on the widest teacher (Qwen), where it reaches the original recipe's approx. 20B-token accuracy at 10B tokens (2x token efficiency); there the strictly same-lineage arm still recovers +6.39, the fully controlled figure. Controls strongly support attributing the gain to the enlarged reachable set, rather than to added parameters or the recipe. From approx. 10B distillation tokens plus a short SFT, a half-parameter 1.5B student matches its approx. 9T-token teacher's 9-task macro-average, within evaluation noise and with a residual MMLU deficit, and a 2.7B student beats Meta's own official compression of Llama3.1-8B at ~900x fewer compression tokens (a token count under unmatched recipes, not a compute claim). All results are from single-seed runs on the LRC backbone.

cs.LG↗

arXiv2026

OmniCustom: Sync Audio-Video Customization Via Joint Audio-Video Generation Model

Existing mainstream video customization methods focus on generating identity-consistent videos based on given reference images and textual prompts. Benefiting from the rapid advancement of joint audio-video generation, this paper proposes a more compelling new task: sync audio-video customization, which aims to synchronously customize both video identity and audio timbre. Specifically, given a reference image $I^{r}$ and a reference audio $A^{r}$, this novel task requires generating videos that maintain the identity of the reference image while imitating the timbre of the reference audio, with spoken content freely specifiable through user-provided textual prompts. To this end, we propose OmniCustom, a powerful DiT-based audio-video customization framework that can synthesize a video following reference image identity, audio timbre, and text prompts all at once in a zero-shot manner. Our framework is built on three key contributions. First, identity and audio timbre control are achieved through separate reference identity and audio LoRA modules that operate through self-attention layers within the base audio-video generation model. Second, we introduce a contrastive learning objective alongside the standard flow matching objective. It uses predicted flows conditioned on reference inputs as positive examples and those without reference conditions as negative examples, thereby enhancing the model ability to preserve identity and timbre. Third, we train OmniCustom on our constructed large-scale, high-quality audio-visual human dataset. Extensive experiments demonstrate that OmniCustom outperforms existing methods in generating audio-video content with consistent identity and timbre fidelity. Project page: https://omnicustom-project.github.io/page/.

cs.SD↗

arXiv2026

Controllable Video Generation: A Survey

With the rapid development of AI-generated content (AIGC), video generation has emerged as one of its most dynamic and impactful subfields. In particular, the advancement of video generation foundation models has led to growing demand for controllable video generation methods that can more accurately reflect user intent. Most existing foundation models are designed for text-to-video generation, where text prompts alone are often insufficient to express complex, multi-modal, and fine-grained user requirements. This limitation makes it challenging for users to generate videos with precise control using current models. To address this issue, recent research has explored the integration of additional non-textual conditions, such as camera motion, depth maps, and human pose, to extend pretrained video generation models and enable more controllable video synthesis. These approaches aim to enhance the flexibility and practical applicability of AIGC-driven video generation systems. In this survey, we provide a systematic review of controllable video generation, covering both theoretical foundations and recent advances in the field. We begin by introducing the key concepts and commonly used open-source video generation models. We then focus on control mechanisms in video diffusion models, analyzing how different types of conditions can be incorporated into the denoising process to guide generation. Finally, we categorize existing methods based on the types of control signals they leverage, including single-condition generation, multi-condition generation, and universal controllable generation. For a complete list of the literature on controllable video generation reviewed, please visit our curated repository at https://github.com/mayuelala/Awesome-Controllable-Video-Generation.

cs.GR↗

arXiv2025

Towards Dataset Copyright Evasion Attack against Personalized Text-to-Image Diffusion Models

Text-to-image (T2I) diffusion models enable high-quality image generation conditioned on textual prompts. However, fine-tuning these pre-trained models for personalization raises concerns about unauthorized dataset usage. To address this issue, dataset ownership verification (DOV) has recently been proposed, which embeds watermarks into fine-tuning datasets via backdoor techniques. These watermarks remain dormant on benign samples but produce owner-specified outputs when triggered. Despite its promise, the robustness of DOV against copyright evasion attacks (CEA) remains unexplored. In this paper, we investigate how adversaries can circumvent these mechanisms, enabling models trained on watermarked datasets to bypass ownership verification. We begin by analyzing the limitations of potential attacks achieved by backdoor removal, including TPD and T2IShield. In practice, TPD suffers from inconsistent effectiveness due to randomness, while T2IShield fails when watermarks are embedded as local image patches. To this end, we introduce CEAT2I, the first CEA specifically targeting DOV in T2I diffusion models. CEAT2I consists of three stages: (1) motivated by the observation that T2I models converge faster on watermarked samples with respect to intermediate features rather than training loss, we reliably detect watermarked samples; (2) we iteratively ablate tokens from the prompts of detected samples and monitor feature shifts to identify trigger tokens; and (3) we apply a closed-form concept erasure method to remove the injected watermarks. Extensive experiments demonstrate that CEAT2I effectively evades state-of-the-art DOV mechanisms while preserving model performance. The code is available at https://github.com/csyufei/CEAT2I.

cs.CV↗

arXiv2025

Vox-Evaluator: Enhancing Stability and Fidelity for Zero-shot TTS with A Multi-Level Evaluator

Recent advances in zero-shot text-to-speech (TTS), driven by language models, diffusion models and masked generation, have achieved impressive naturalness in speech synthesis. Nevertheless, stability and fidelity remain key challenges, manifesting as mispronunciations, audible noise, and quality degradation. To address these issues, we introduce Vox-Evaluator, a multi-level evaluator designed to guide the correction of erroneous speech segments and preference alignment for TTS systems. It is capable of identifying the temporal boundaries of erroneous segments and providing a holistic quality assessment of the generated speech. Specifically, to refine erroneous segments and enhance the robustness of the zero-shot TTS model, we propose to automatically identify acoustic errors with the evaluator, mask the erroneous segments, and finally regenerate speech conditioning on the correct portions. In addition, the fine-gained information obtained from Vox-Evaluator can guide the preference alignment for TTS model, thereby reducing the bad cases in speech synthesis. Due to the lack of suitable training datasets for the Vox-Evaluator, we also constructed a synthesized text-speech dataset annotated with fine-grained pronunciation errors or audio quality issues. The experimental results demonstrate the effectiveness of the proposed Vox-Evaluator in enhancing the stability and fidelity of TTS systems through the speech correction mechanism and preference optimization. The demos are shown.

cs.SD↗

arXiv2025

Follow-Your-Emoji-Faster: Towards Efficient, Fine-Controllable, and Expressive Freestyle Portrait Animation

We present Follow-Your-Emoji-Faster, an efficient diffusion-based framework for freestyle portrait animation driven by facial landmarks. The main challenges in this task are preserving the identity of the reference portrait, accurately transferring target expressions, and maintaining long-term temporal consistency while ensuring generation efficiency. To address identity preservation and accurate expression retargeting, we enhance Stable Diffusion with two key components: a expression-aware landmarks as explicit motion signals, which improve motion alignment, support exaggerated expressions, and reduce identity leakage; and a fine-grained facial loss that leverages both expression and facial masks to better capture subtle expressions and faithfully preserve the reference appearance. With these components, our model supports controllable and expressive animation across diverse portrait types, including real faces, cartoons, sculptures, and animals. However, diffusion-based frameworks typically struggle to efficiently generate long-term stable animation results, which remains a core challenge in this task. To address this, we propose a progressive generation strategy for stable long-term animation, and introduce a Taylor-interpolated cache, achieving a 2.6X lossless acceleration. These two strategies ensure that our method produces high-quality results efficiently, making it user-friendly and accessible. Finally, we introduce EmojiBench++, a more comprehensive benchmark comprising diverse portraits, driving videos, and landmark sequences. Extensive evaluations on EmojiBench++ demonstrate that Follow-Your-Emoji-Faster achieves superior performance in both animation quality and controllability. The code, training dataset and benchmark will be found in https://follow-your-emoji.github.io/.

cs.CV↗

arXiv2025

USM-VC: Mitigating Timbre Leakage with Universal Semantic Mapping Residual Block for Voice Conversion

Voice conversion (VC) transforms source speech into a target voice by preserving the content. However, timbre information from the source speaker is inherently embedded in the content representations, causing significant timbre leakage and reducing similarity to the target speaker. To address this, we introduce a Universal Semantic Matching (USM) residual block to a content extractor. The residual block consists of two weighted branches: 1) universal semantic dictionary based Content Feature Re-expression (CFR) module, supplying timbre-free content representation. 2) skip connection to the original content layer, providing complementary fine-grained information. In the CFR module, each dictionary entry in the universal semantic dictionary represents a phoneme class, computed statistically using speech from multiple speakers, creating a stable, speaker-independent semantic set. We introduce a CFR method to obtain timbre-free content representations by expressing each content frame as a weighted linear combination of dictionary entries using corresponding phoneme posteriors as weights. Extensive experiments across various VC frameworks demonstrate that our approach effectively mitigates timbre leakage and significantly improves similarity to the target speaker.

eess.AS↗

arXiv2025

Follow-Your-Motion: Video Motion Transfer via Efficient Spatial-Temporal Decoupled Finetuning

Recently, breakthroughs in the video diffusion transformer have shown remarkable capabilities in diverse motion generations. As for the motion-transfer task, current methods mainly use two-stage Low-Rank Adaptations (LoRAs) finetuning to obtain better performance. However, existing adaptation-based motion transfer still suffers from motion inconsistency and tuning inefficiency when applied to large video diffusion transformers. Naive two-stage LoRA tuning struggles to maintain motion consistency between generated and input videos due to the inherent spatial-temporal coupling in the 3D attention operator. Additionally, they require time-consuming fine-tuning processes in both stages. To tackle these issues, we propose Follow-Your-Motion, an efficient two-stage video motion transfer framework that finetunes a powerful video diffusion transformer to synthesize complex motion. Specifically, we propose a spatial-temporal decoupled LoRA to decouple the attention architecture for spatial appearance and temporal motion processing. During the second training stage, we design the sparse motion sampling and adaptive RoPE to accelerate the tuning speed. To address the lack of a benchmark for this field, we introduce MotionBench, a comprehensive benchmark comprising diverse motion, including creative camera motion, single object motion, multiple object motion, and complex human motion. We show extensive evaluations on MotionBench to verify the superiority of Follow-Your-Motion.

cs.CV↗

arXiv2025

Temporal Entailment Pretraining for Clinical Language Models over EHR Data

Clinical language models have achieved strong performance on downstream tasks by pretraining on domain specific corpora such as discharge summaries and medical notes. However, most approaches treat the electronic health record as a static document, neglecting the temporally-evolving and causally entwined nature of patient trajectories. In this paper, we introduce a novel temporal entailment pretraining objective for language models in the clinical domain. Our method formulates EHR segments as temporally ordered sentence pairs and trains the model to determine whether a later state is entailed by, contradictory to, or neutral with respect to an earlier state. Through this temporally structured pretraining task, models learn to perform latent clinical reasoning over time, improving their ability to generalize across forecasting and diagnosis tasks. We pretrain on a large corpus derived from MIMIC IV and demonstrate state of the art results on temporal clinical QA, early warning prediction, and disease progression modeling.

cs.CL↗

arXiv2025

The Holography of the 2D inhomogeneously deformed CFT

We systematically study inhomogeneous Hamiltonians in two-dimensional conformal field theories within the framework of the AdS/CFT correspondence by relating them to two-dimensional curved backgrounds. We propose a classification of inhomogeneous Hamiltonians based on the Virasoro coadjoint orbit. The corresponding bulk dual geometries are described by the generalized Ba$\tilde{\text{n}}$ados solutions, for which we introduce a generalized Roberts mapping to facilitate their study. Our classification provides previously underexplored classes of deformations, offering fresh insights into their holographic properties. Revisiting the well-known example of the M$\ddot{\text{o}}$bius Hamiltonian, we establish a connection to the 3D C-metric, which describes three-dimensional accelerating solutions. Furthermore, we extend our analysis to KdV-type asymptotic boundary conditions, revealing a broader class of solvable inhomogeneous Hamiltonians that are not linear combinations of Virasoro charges but instead involve KdV charges.

hep-th↗

arXiv2025

Embedding Self-Correction as an Inherent Ability in Large Language Models for Enhanced Mathematical Reasoning

Accurate mathematical reasoning with Large Language Models (LLMs) is crucial in revolutionizing domains that heavily rely on such reasoning. However, LLMs often encounter difficulties in certain aspects of mathematical reasoning, leading to flawed reasoning and erroneous results. To mitigate these issues, we introduce a novel mechanism, the Chain of Self-Correction (CoSC), specifically designed to embed self-correction as an inherent ability in LLMs, enabling them to validate and rectify their own results. The CoSC mechanism operates through a sequence of self-correction stages. In each stage, the LLMs generate a program to address a given problem, execute this program using program-based tools to obtain an output, subsequently verify this output. Based on the verification, the LLMs either proceed to the next correction stage or finalize the answer. This iterative self-correction process allows the LLMs to refine its reasoning steps and improve the accuracy of its mathematical reasoning. We implement CoSC using a two-phase fine-tuning approach. First, LLMs are trained with a relatively small volume of seeding data generated from GPT-4. Then, we enhance CoSC by training with a larger volume of self-generated data, without relying on GPT-4. Experiments show that CoSC significantly boosts performance on standard mathematical datasets compared to existing open-source LLMs. Notably, our CoSC-Code-34B model achieved a 53.5% score on the challenging MATH dataset, outperforming models like ChatGPT, GPT-4, and multi-modal LLMs such as GPT-4V and Gemini-1.0. Importantly, CoSC operates in a zero-shot manner without requiring demonstrations.

cs.AI↗

arXiv2024

System-2 Mathematical Reasoning via Enriched Instruction Tuning

Solving complex mathematical problems via system-2 reasoning is a natural human skill, yet it remains a significant challenge for current large language models (LLMs). We identify the scarcity of deliberate multi-step reasoning data as a primary limiting factor. To this end, we introduce Enriched Instruction Tuning (EIT), a method that enriches existing human-annotated mathematical datasets by synergizing human and AI feedback to create fine-grained reasoning trajectories. These datasets are then used to fine-tune open-source LLMs, enhancing their mathematical reasoning abilities without reliance on any symbolic verification program. Concretely, EIT is composed of two critical steps: Enriching with Reasoning Plan (ERP) and Enriching with Reasoning Step (ERS). The former generates a high-level plan that breaks down complex instructions into a sequence of simpler objectives, while ERS fills in reasoning contexts often overlooked by human annotators, creating a smoother reasoning trajectory for LLM fine-tuning. Unlike existing CoT prompting methods that generate reasoning chains only depending on LLM's internal knowledge, our method leverages human-annotated initial answers as ``meta-knowledge'' to help LLMs generate more detailed and precise reasoning processes, leading to a more trustworthy LLM expert for complex mathematical problems. In experiments, EIT achieves an accuracy of 84.1% on GSM8K and 32.5% on MATH, surpassing state-of-the-art fine-tuning and prompting methods, and even matching the performance of tool-augmented methods.

cs.AI↗

arXiv2024

Backdoor Attack with Sparse and Invisible Trigger

Deep neural networks (DNNs) are vulnerable to backdoor attacks, where the adversary manipulates a small portion of training data such that the victim model predicts normally on the benign samples but classifies the triggered samples as the target class. The backdoor attack is an emerging yet threatening training-phase threat, leading to serious risks in DNN-based applications. In this paper, we revisit the trigger patterns of existing backdoor attacks. We reveal that they are either visible or not sparse and therefore are not stealthy enough. More importantly, it is not feasible to simply combine existing methods to design an effective sparse and invisible backdoor attack. To address this problem, we formulate the trigger generation as a bi-level optimization problem with sparsity and invisibility constraints and propose an effective method to solve it. The proposed method is dubbed sparse and invisible backdoor attack (SIBA). We conduct extensive experiments on benchmark datasets under different settings, which verify the effectiveness of our attack and its resistance to existing backdoor defenses. The codes for reproducing main experiments are available at \url{https://github.com/YinghuaGao/SIBA}.

cs.CV↗

arXiv2024

Nearest is Not Dearest: Towards Practical Defense against Quantization-conditioned Backdoor Attacks

Model quantization is widely used to compress and accelerate deep neural networks. However, recent studies have revealed the feasibility of weaponizing model quantization via implanting quantization-conditioned backdoors (QCBs). These special backdoors stay dormant on released full-precision models but will come into effect after standard quantization. Due to the peculiarity of QCBs, existing defenses have minor effects on reducing their threats or are even infeasible. In this paper, we conduct the first in-depth analysis of QCBs. We reveal that the activation of existing QCBs primarily stems from the nearest rounding operation and is closely related to the norms of neuron-wise truncation errors (i.e., the difference between the continuous full-precision weights and its quantized version). Motivated by these insights, we propose Error-guided Flipped Rounding with Activation Preservation (EFRAP), an effective and practical defense against QCBs. Specifically, EFRAP learns a non-nearest rounding strategy with neuron-wise error norm and layer-wise activation preservation guidance, flipping the rounding strategies of neurons crucial for backdoor effects but with minimal impact on clean accuracy. Extensive evaluations on benchmark datasets demonstrate that our EFRAP can defeat state-of-the-art QCB attacks under various settings. Code is available at https://github.com/AntigoneRandy/QuantBackdoor_EFRAP.

cs.CR↗

arXiv2024

Energy-Latency Manipulation of Multi-modal Large Language Models via Verbose Samples

Despite the exceptional performance of multi-modal large language models (MLLMs), their deployment requires substantial computational resources. Once malicious users induce high energy consumption and latency time (energy-latency cost), it will exhaust computational resources and harm availability of service. In this paper, we investigate this vulnerability for MLLMs, particularly image-based and video-based ones, and aim to induce high energy-latency cost during inference by crafting an imperceptible perturbation. We find that high energy-latency cost can be manipulated by maximizing the length of generated sequences, which motivates us to propose verbose samples, including verbose images and videos. Concretely, two modality non-specific losses are proposed, including a loss to delay end-of-sequence (EOS) token and an uncertainty loss to increase the uncertainty over each generated token. In addition, improving diversity is important to encourage longer responses by increasing the complexity, which inspires the following modality specific loss. For verbose images, a token diversity loss is proposed to promote diverse hidden states. For verbose videos, a frame feature diversity loss is proposed to increase the feature diversity among frames. To balance these losses, we propose a temporal weight adjustment algorithm. Experiments demonstrate that our verbose samples can largely extend the length of generated sequences.

cs.CV↗

arXiv2024

DGL-GAN: Discriminator Guided Learning for GAN Compression

Generative Adversarial Networks (GANs) with high computation costs, e.g., BigGAN and StyleGAN2, have achieved remarkable results in synthesizing high-resolution images from random noise. Reducing the computation cost of GANs while keeping generating photo-realistic images is a challenging field. In this work, we propose a novel yet simple {\bf D}iscriminator {\bf G}uided {\bf L}earning approach for compressing vanilla {\bf GAN}, dubbed {\bf DGL-GAN}. Motivated by the phenomenon that the teacher discriminator may contain some meaningful information about both real images and fake images, we merely transfer the knowledge from the teacher discriminator via the adversarial interaction between the teacher discriminator and the student generator. We apply DGL-GAN to compress the two most representative large-scale vanilla GANs, i.e., StyleGAN2 and BigGAN. Experiments show that DGL-GAN achieves state-of-the-art (SOTA) results on both StyleGAN2 and BigGAN. Moreover, DGL-GAN is also effective in boosting the performance of original uncompressed GANs. Original uncompressed StyleGAN2 boosted with DGL-GAN achieves FID 2.65 on FFHQ, which achieves a new state-of-the-art performance. Code and models are available at \url{https://github.com/yuesongtian/DGL-GAN}

cs.CV↗

arXiv2024

BadCLIP: Trigger-Aware Prompt Learning for Backdoor Attacks on CLIP

Contrastive Vision-Language Pre-training, known as CLIP, has shown promising effectiveness in addressing downstream image recognition tasks. However, recent works revealed that the CLIP model can be implanted with a downstream-oriented backdoor. On downstream tasks, one victim model performs well on clean samples but predicts a specific target class whenever a specific trigger is present. For injecting a backdoor, existing attacks depend on a large amount of additional data to maliciously fine-tune the entire pre-trained CLIP model, which makes them inapplicable to data-limited scenarios. In this work, motivated by the recent success of learnable prompts, we address this problem by injecting a backdoor into the CLIP model in the prompt learning stage. Our method named BadCLIP is built on a novel and effective mechanism in backdoor attacks on CLIP, i.e., influencing both the image and text encoders with the trigger. It consists of a learnable trigger applied to images and a trigger-aware context generator, such that the trigger can change text features via trigger-aware prompts, resulting in a powerful and generalizable attack. Extensive experiments conducted on 11 datasets verify that the clean accuracy of BadCLIP is similar to those of advanced prompt learning methods and the attack success rate is higher than 99% in most cases. BadCLIP is also generalizable to unseen classes, and shows a strong generalization capability under cross-dataset and cross-domain settings.

cs.CV↗

arXiv2024

Inducing High Energy-Latency of Large Vision-Language Models with Verbose Images

Large vision-language models (VLMs) such as GPT-4 have achieved exceptional performance across various multi-modal tasks. However, the deployment of VLMs necessitates substantial energy consumption and computational resources. Once attackers maliciously induce high energy consumption and latency time (energy-latency cost) during inference of VLMs, it will exhaust computational resources. In this paper, we explore this attack surface about availability of VLMs and aim to induce high energy-latency cost during inference of VLMs. We find that high energy-latency cost during inference of VLMs can be manipulated by maximizing the length of generated sequences. To this end, we propose verbose images, with the goal of crafting an imperceptible perturbation to induce VLMs to generate long sentences during inference. Concretely, we design three loss objectives. First, a loss is proposed to delay the occurrence of end-of-sequence (EOS) token, where EOS token is a signal for VLMs to stop generating further tokens. Moreover, an uncertainty loss and a token diversity loss are proposed to increase the uncertainty over each generated token and the diversity among all tokens of the whole generated sequence, respectively, which can break output dependency at token-level and sequence-level. Furthermore, a temporal weight adjustment algorithm is proposed, which can effectively balance these losses. Extensive experiments demonstrate that our verbose images can increase the length of generated sequences by 7.87 times and 8.56 times compared to original images on MS-COCO and ImageNet datasets, which presents potential challenges for various applications. Our code is available at https://github.com/KuofengGao/Verbose_Images.

cs.CV↗