Denoising-based diffusion models have attained impressive image synthesis; however, their applications on videos can lead to unaffordable computational costs due to the per-frame denoising operations. In pursuit of efficient video generation, we present a Diffusion Reuse MOtion (Dr. Mo) network to accelerate the video-based denoising process. Our crucial observation is that the latent representations in early denoising steps between adjacent video frames exhibit high consistencies with motion clues. Inspired by the discovery, we propose to accelerate the video denoising process by incorporating lightweight, learnable motion features. Specifically, Dr. Mo will only compute all denoising steps for base frames. For a non-based frame, Dr. Mo will propagate the pre-computed based latents of a particular step with inter-frame motions to obtain a fast estimation of its coarse-grained latent representation, from which the denoising will continue to obtain more sensitive and fine-grained representations. On top of this, Dr. Mo employs a meta-network named Denoising Step Selector (DSS) to dynamically determine the step to perform motion-based propagations for each frame, ensuring the correct transformation of multi-granularity visual features. Extensive evaluations on video generation and editing tasks indicate that Dr. Mo delivers widely applicable acceleration for diffusion-based video generations while effectively retaining the visual quality and style. Video generation and visualization results can be found athttps://drmo-denoising-reuse.github.io.

Language models (LMs) only pretrained on a general and massive corpus usually cannot attain satisfying performance on domain-specific downstream tasks, and hence, applying domain-specific pretraining to LMs is a common and indispensable practice. However, domain-specific pretraining can be costly and time-consuming, hindering LMs' deployment in real-world applications. In this work, we consider the incapability to memorize domain-specific knowledge embedded in the general corpus with rare occurrences and long-tail distributions as the leading cause for pretrained LMs' inferior downstream performance. Analysis of Neural Tangent Kernels (NTKs) reveals that those long-tail data are commonly overlooked in the model's gradient updates and, consequently, are not effectively memorized, leading to poor domain-specific downstream performance. Based on the intuition that data with similar semantic meaning are closer in the embedding space, we devise a Cluster-guided Sparse Expert (CSE) layer to actively learn long-tail domain knowledge typically neglected in previous pretrained LMs. During pretraining, a CSE layer efficiently clusters domain knowledge together and assigns long-tail knowledge to designate extra experts. CSE is also a lightweight structure that only needs to be incorporated in several deep layers. With our training strategy, we found that during pretraining, data of long-tail knowledge gradually formulate isolated, outlier clusters in an LM's representation spaces, especially in deeper layers. Our experimental results show that only pretraining CSE-based LMs is enough to achieve superior performance than regularly pretrained-finetuned LMs on various downstream tasks, implying the prospects of domain-specific-pretraining-free language models.

Despite their prevalence in deep-learning communities, over-parameterized models convey high demands of computational costs for proper training. This work studies the fine-grained, modular-level learning dynamics of over-parameterized models to attain a more efficient and fruitful training strategy. Empirical evidence reveals that when scaling down into network modules, such as heads in self-attention models, we can observe varying learning patterns implicitly associated with each module's trainability. To describe such modular-level learning capabilities, we introduce a novel concept dubbed modular neural tangent kernel (mNTK), and we demonstrate that the quality of a module's learning is tightly associated with its mNTK's principal eigenvalue $\lambda_{\max}$. A large $\lambda_{\max}$ indicates that the module learns features with better convergence, while those miniature ones may impact generalization negatively. Inspired by the discovery, we propose a novel training strategy termed Modular Adaptive Training (MAT) to update those modules with their $\lambda_{\max}$ exceeding a dynamic threshold selectively, concentrating the model on learning common features and ignoring those inconsistent ones. Unlike most existing training schemes with a complete BP cycle across all network modules, MAT can significantly save computations by its partially-updating strategy and can further improve performance. Experiments show that MAT nearly halves the computational cost of model training and outperforms the accuracy of baselines.

Cross-modal learning is essential to enable accurate fake news detection due to the fast-growing multimodal contents in online social communities. A fundamental challenge of multimodal fake news detection lies in the inherent ambiguity across different content modalities, i.e., decisions made from unimodalities may disagree with each other, which may lead to inferior multimodal fake news detection. To address this issue, we formulate the cross-modal ambiguity learning problem from an information-theoretic perspective and propose CAFE — an ambiguity-aware multimodal fake news detection method. CAFE consists of 1) a cross-modal alignment module to transform the heterogeneous unimodality features into a shared semantic space, 2) a cross-modal ambiguity learning module to estimate the ambiguity between different modalities, and 3) a cross-modal fusion module to capture the cross-modal correlations. CAFE improves fake news detection accuracy by judiciously and adaptively aggregating unimodal features and cross-modal correlations, i.e., relying on unimodal features when cross-modal ambiguity is weak and referring to cross-modal correlations when cross-modal ambiguity is strong. Experimental studies on two widely used datasets (Twitter and Weibo) demonstrate that CAFE outperforms state-of-the-art fake news detection methods by 2.2-18.9% and 1.7-11.4% on accuracy, respectively.

Wide dissemination of unverified claims has negative influence on social lives. Rumors are easy to emerge and spread in the crowds especially in Online Social Network (OSN), due to its openness and extensive amount of users. Therefore, rumor detection in OSN is a very challenging and urgent issue. In this paper, we propose an Attention-Residual network combined with CNN (ARC), which is based on the content features for rumor detection. First, we build a data encoding model based on word-level data for contextual feature representation. Second, we propose a residual framework based on fine-tuned attention mechanism to capture long-range dependency. Third, we apply convolution neural network with varying window size to select important components and local features. Experiments on two twitter datasets demonstrate that the proposed model has better performance than other content-based methods both in rumor detection and early rumor verification. To the best of our knowledge, we are the first work that utilize attention model in conjunction with residual network on rumor detection.