This work proposes MeanFlow-Transfer, which maps heterogeneous source outputs into a shared velocity representation, uses it to initialize an MF generator from the source weights, and optimizes an MF objective on the target domain, and introduces Continuous Adversarial MeanFlow, a post-training stage that extends continuous adversarial flow models from instantaneous velocities to MF's finite-interval average velocities.
Abstract
Training fast generators on new domains with limited data remains challenging for two reasons. First, adapting a pretrained diffusion or flow model to a new domain leaves its costly multi-step sampling unaddressed, and existing acceleration methods are tied to the source parameterization--$\epsilon$, $x$, $v$, or $u$--leaving heterogeneous pretrained models with no common acceleration target. Second, while adversarial refinement is proven effective for few-step quality, it is formulated only for instantaneous-velocity flows, not for the finite-interval average velocities that MeanFlow (MF) models predict. We address both problems. We propose MeanFlow-Transfer, which maps heterogeneous source outputs into a shared velocity representation, uses it to initialize an MF generator from the source weights, and optimizes an MF objective on the target domain. This unifies adaptation and acceleration in a single training loop across a broad range of pretrained models. We then introduce Continuous Adversarial MeanFlow, a post-training stage that extends continuous adversarial flow models from instantaneous velocities to MF's finite-interval average velocities. CAMF contrasts changes in a learned potential between real and predicted interval endpoints, recovering fine detail that MF regression averages away, and reduces to the instantaneous criterion in the vanishing-interval limit. Adapting four ImageNet-based source models--DiT ($\epsilon$), SiT ($v$), JiT ($x$), iMF ($u$)--to five target domains, MF-T with CAMF matches or exceeds the fine-tuned teacher in FID and FDD at up to $125\times$ fewer Neural Function Evaluations (NFEs), while CAMF improves MF-T's few-step FID by $29\%$ on average.
While deep learning has significantly advanced automatic modulation recognition in complex environments, its performance is often limited by domain shifts caused by factors like channel fading and frequency offset. Domain adaptation has emerged as the primary paradigm to address this challenge. However, existing methods face a critical trade-off, as global alignment strategies tend to disrupt class-specific structures, while local alignment methods are overly sensitive to the quality of pseudo-labels. To address this trade-off, this paper proposes a joint adversarial and subdomain adaptation network (JASA-Net) centered on a dual domain adaptation (DDA) strategy. This strategy employs a “global-first, then-local” alignment ap-proach, where an initial global adversarial alignment establishes a strong foundation for generating high-quality pseudo-labels that subsequently guide a local alignment via the local maximum mean discrepancy (LMMD) metric. To extract robust features, we design a patch adaptive multimodal transformer (PAMT) encoder. Furthermore, an efficient “source domain warm-up + aggressive scheduling” training strategy is developed to enhance performance. Extensive experiments on custom-simulated datasets demonstrate that the proposed framework significantly outperforms representative baselines across various cross-domain tasks. Notably, it exhibits remarkable robustness in few-shot scenarios. This work also provides a critical analysis of the task-dependent nature of entropy weighting, offering valuable insights for future research in the field.
Ming Cheng, Wen Deng, Jintian Xiong et al.· IEEE Transactions on Cogniti...· 0 citations
This work concerns adversarial training against the small-norm adversarial examples that arise from the inherent input instability of a trained deep neural network. Examples in this class are small as measured in the relative $\ell^2$-norm, and therefore lie in the neighborhood of the input on which the model acts approximately linearly, the regime in which the perturbation remains imperceptible. We first show that such examples can be computed directly from the trained network parameters, without input gradient iterations, by means of a linearization called the low-rank Householder expansion (LRHE). The expansion describes the composed affine map rather than any individual layer, and the directions it identifies are read from the activation pattern already available in the forward pass. We then propose a simple adversarial training scheme built on this construction. No differentiation with respect to the input is performed at any point: training requires only additional forward evaluations, with weight parameters updated by the standard backward pass, and the inner maximization of the usual min-max formulation is eliminated entirely. That such a regularizer exists is our main finding: the methods that dispense with the inner search all obtain their local geometry by differentiating with respect to the input, and we show this is not necessary. The regularizer costs the equivalent of $2.8$ PGD steps per epoch, an $8.7\times$ reduction relative to 40-step adversarial training on MNIST and below the cost of 3-step training. The resulting models match three-step PGD adversarial training for relative $\ell^2$ budgets $\varepsilon \le 0.02$ and 40-step training for $\varepsilon \le 0.012$, falling away beyond, consistent with the locality of the expansion.
Specific emitter identification (SEI) is challenged by both limited labeled data and domain shifts caused by variations in acquisition conditions, devices, and propagation environments. To address these issues, this paper proposes TransPileSiam, a domain-aware self-supervised pretraining framework for few-shot SEI. Built upon SimSiam, the proposed method constructs multiple augmented views from each raw IQ sample and performs multi-view consistency learning to improve representation robustness under complex perturbations. To further enhance cross-domain transferability, a gradient-reversal-based domain-adversarial regularization is introduced to suppress domain-specific information in the learned features. The pretrained encoder is then adapted to downstream SEI tasks through few-shot fine-tuning. Experimental results show that TransPileSiam consistently improves downstream recognition under limited-label conditions. In particular, it improves the test accuracy of a ResNet34-based supervised model by 3.32% and yields an average gain of about 1% in few-shot evaluation. These results demonstrate that TransPileSiam effectively improves robustness, label efficiency, and cross-domain generalization for practical SEI tasks.
Junwei Peng, Siyang Xu, Jiao Wang et al.· International Conference on...· 0 citations
An in-depth and up- to-date overview of the GANs environment, principally highlighting the progress made over 2020 and beyond and proposing the idea of hybrid generative systems in the future while emphasizing the oppositional approach's extraordinary and enduring features.
Zahraa Salah Dhaif, Hind Jumaa Serteep· International Journal of Adv...· 0 citations
FDT-PC (Frequency Domain Transformation with Perceptual Constraints), a novel method that enhances adversarial transferability across different model architectures, is proposed, which achieves superior black-box attack performance on both CNNs and Vision Transformers, outperforming existing state-of-the-art input transformation methods.
Bo Li, Li Tang, Xin Jin et al.· ACM Transactions on Multimed...· 0 citations
This work proposes BMAT (Bilevel-Minimax Adversarial Transfer), an integrated bottom-up solver that combines a Soft Weight Modulator and an Implicit Gradient Approximator to enable ternary coupling among initialization, surrogate adaptation, and perturbation optimization.