This paper explores image modeling from the frequency space and introduces DCTdiff, an end-to-end diffusion generative paradigm that efficiently models images in the discrete cosine transform (DCT) space. We investigate the design space of DCTdiff and reveal the key design factors. Experiments on different frameworks (UViT, DiT), generation tasks, and various diffusion samplers demonstrate that DCTdiff outperforms pixel-based diffusion models regarding generative quality and training efficiency. Remarkably, DCTdiff can seamlessly scale up to 512$\times$512 resolution without using the latent diffusion paradigm and beats latent diffusion (using SD-VAE) with only 1/4 training cost. Finally, we illustrate several intriguing properties of DCT image modeling. For example, we provide a theoretical proof of why `image diffusion can be seen as spectral autoregression', bridging the gap between diffusion and autoregressive models. The effectiveness of DCTdiff and the introduced properties suggest a promising direction for image modeling in the frequency space. The code is at https://github.com/forever208/DCTdiff.

Scaling the effective context length is essential for advancing large language models (LLMs) toward artificial general intelligence (AGI). However, the quadratic increase in computational complexity inherent in traditional attention mechanisms presents a prohibitive overhead. Existing approaches either impose strongly biased structures, such as sink or window attention which are task-specific, or radically modify the attention mechanism into linear approximations, whose performance in complex reasoning tasks remains inadequately explored. In this work, we propose a solution that adheres to the ``less structure'' principle, allowing the model to determine where to attend autonomously, rather than introducing predefined biases. We introduce Mixture of Block Attention (MoBA), an innovative approach that applies the principles of Mixture of Experts (MoE) to the attention mechanism. This novel architecture demonstrates superior performance on long-context tasks while offering a key advantage: the ability to seamlessly transition between full and sparse attention, enhancing efficiency without the risk of compromising performance. MoBA has already been deployed to handle actual production workloads with long-context requirements, demonstrating significant advancements in efficient attention computation for LLMs. Our code is available at https://github.com/MoonshotAI/MoBA.

Diffusion models have demonstrated impressive generative capabilities, but their exposure bias problem, described as the input mismatch between training and sampling, lacks in-depth exploration. In this paper, we investigate the exposure bias problem in diffusion models by first analytically modelling the sampling distribution, based on which we then attribute the prediction error at each sampling step as the root cause of the exposure bias issue. Furthermore, we discuss potential solutions to this issue and propose an intuitive metric for it. Along with the elucidation of exposure bias, we propose a simple, yet effective, training-free method called Epsilon Scaling to alleviate the exposure bias. We show that Epsilon Scaling explicitly moves the sampling trajectory closer to the vector field learned in the training phase by scaling down the network output, mitigating the input mismatch between training and sampling. Experiments on various diffusion frameworks (ADM, DDIM, EDM, LDM, DiT, PFGM++) verify the effectiveness of our method. Remarkably, our ADM-ES, as a state-of-the-art stochastic sampler, obtains 2.17 FID on CIFAR-10 under 100-step unconditional generation. The code is at https://github.com/forever208/ADM-ES

Jianlin Su, Murtadha Ahmed, Bo Wen, Luo Ao, Mingren Zhu, Yunfeng Liu. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Learning the representation of sentences is fundamental work in the field of Natural Language Processing. Although BERT-like transformers have achieved new SOTAs for sentence embedding in many tasks, they have been proven difficult to capture semantic similarity without proper fine-tuning. A common idea to measure Semantic Textual Similarity (STS) is considering the distance between two text embeddings defined by the dot product or cosine function. However, the semantic embedding spaces induced by pretrained transformers are generally non-smooth and tend to deviate from a normal distribution, which makes traditional distance metrics imprecise. In this paper, we first empirically explain the failure of cosine similarity in semantic textual similarity measuring, and present CoSENT, a novelConsistentSENTence embedding framework. Concretely, a supervised objective function is designed to optimize the Siamese BERT network by exploiting ranked similarity labels of sample pairs. The loss function utilizes uniform cosine similarity-based optimization for both the training and prediction phases, improving the consistency of the learned semantic space. Additionally, the unified objective function can be adaptively applied to different datasets with various types of annotations and different comparison schemes of the STS tasks only by using sortable labels. Empirical evaluations on 14 common textual similarity benchmarks demonstrate that the proposed CoSENT excels in performance and reduces training time cost.

Graph neural networks (GNNs) are now the mainstream method for mining graph-structured data and learning low-dimensional node- and graph-level embeddings to serve downstream tasks. However, limited by the bottleneck of interpretability that deep neural networks present, existing GNNs have ignored the issue of estimating the appropriate number of dimensions for the embeddings. Hence, we propose a novel framework called Minimum Graph Entropy principle-guided Dimension Estimation, i.e. MGEDE, that learns the appropriate embedding dimensions for both node and graph representations. In terms of node-level estimation, a minimum entropy function that counts both structure and attribute entropy, appraises the appropriate number of dimensions. In terms of graph-level estimation, each graph is assigned a customized embedding dimension from a candidate set based on the number of dimensions estimated for the node-level embeddings. Comprehensive experiments with node and graph classification tasks and nine benchmark datasets verify the effectiveness and generalizability of MGEDE.

Abstract Semi-supervised text classification-based paradigms (SSTC) typically employ the spirit of self-training. The key idea is to train a deep classifier on limited labeled texts and then iteratively predict the unlabeled texts as their pseudo-labels for further training. However, the performance is largely affected by the accuracy of pseudo-labels, which may not be significant in real-world scenarios. This paper presents a Rank-aware Negative Training (RNT) framework to address SSTC in learning with noisy label settings. To alleviate the noisy information, we adapt a reasoning with uncertainty-based approach to rank the unlabeled texts based on the evidential support received from the labeled texts. Moreover, we propose the use of negative training to train RNT based on the concept that “the input instance does not belong to the complementary label”. A complementary label is randomly selected from all labels except the label on-target. Intuitively, the probability of a true label serving as a complementary label is low and thus provides less noisy information during the training, resulting in better performance on the test data. Finally, we evaluate the proposed solution on various text classification benchmark datasets. Our extensive experiments show that it consistently overcomes the state-of-the-art alternatives in most scenarios and achieves competitive performance in the others. The code of RNT is publicly available on GitHub.

Differentiable architecture search (DARTS) has been a popular one-shot paradigm for NAS due to its high efficiency. It introduces trainable architecture parameters to represent the importance of candidate operations and proposes first/second-order approximation to estimate their gradients, making it possible to solve NAS by gradient descent algorithm. However, our in-depth empirical results show that the approximation often distorts the loss landscape, leading to the biased objective to optimize and, in turn, inaccurate gradient estimation for architecture parameters. This work turns to zero-order optimization and proposes a novel NAS scheme, called ZARTS, to search without enforcing the above approximation. Specifically, three representative zero-order optimization methods are introduced: RS, MGS, and GLD, among which MGS performs best by balancing the accuracy and speed. Moreover, we explore the connections between RS/MGS and gradient descent algorithm and show that our ZARTS can be seen as a robust gradient-free counterpart to DARTS. Extensive experiments on multiple datasets and search spaces show the remarkable performance of our method. In particular, results on 12 benchmarks verify the outstanding robustness of ZARTS, where the performance of DARTS collapses due to its known instability issue. Also, we search on the search space of DARTS to compare with peer methods, and our discovered architecture achieves 97.54\% accuracy on CIFAR-10 and 75.7\% top-1 accuracy on ImageNet. Finally, we combine our ZARTS with three orthogonal variants of DARTS for faster search speed and better performance. Source code will be made publicly available at: \url{https://github.com/vicFigure/ZARTS}.

Graph-level anomaly detection aims to distinguish anomalous graphs in a graph dataset from normal graphs. Anomalous graphs represent a very few but essential patterns in the real world. The anomalous property of a graph may be referable to its anomalous attributes of particular nodes and anomalous substructures that refer to a subset of nodes and edges in the graph. In addition, due to the imbalance nature of anomaly problem, anomalous information will be diluted by normal graphs with overwhelming quantities. Various anomaly notions in the attributes and/or substructures and the imbalance nature together make detecting anomalous graphs a non-trivial task. In this paper, we propose a graph neural network for graph-level anomaly detection, namely iGAD. Specifically, an anomalous graph attribute-aware graph convolution and an anomalous graph substructure-aware deep Random Walk Kernel (deep RWK) are welded into a graph neural network to achieve the dual-discriminative ability on anomalous attributes and substructures. Deep RWK in iGAD makes up for the deficiency of graph convolution in distinguishing structural information caused by the simple neighborhood aggregation mechanism. Further, we propose a Point Mutual Information (PMI)-based loss function to target the problems caused by imbalance distributions. PMI-based loss function enables iGAD to capture essential correlation between input graphs and their anomalous/normal properties. We evaluate iGAD on four real-world graph datasets. Extensive experiments demonstrate the superiority of iGAD on the graph-level anomaly detection task.

The attention mechanism can refine the extracted feature maps and boost the classification performance of the deep network, which has become an essential technique in computer vision and natural language processing. However, the memory and computational costs of the dot-product attention mechanism increase quadratically with the spatiotemporal size of the input. Such growth hinders the usage of attention mechanisms considerably in application scenarios with large-scale inputs. In this letter, we propose a linear attention mechanism (LAM) to address this issue, which is approximately equivalent to dot-product attention with computational efficiency. Such a design makes the incorporation between attention mechanisms and deep networks much more flexible and versatile. Based on the proposed LAM, we refactor the skip connections in the raw U-Net and design a multistage attention ResU-Net (MAResU-Net) for semantic segmentation from fine-resolution remote sensing images. Experiments conducted on the Vaihingen data set demonstrated the effectiveness and efficiency of our MAResU-Net. Our code is available athttps://github.com/lironui/MAResU-Net.

Semantic segmentation of remote sensing images plays an important role in a wide range of applications, including land resource management, biosphere monitoring, and urban planning. Although the accuracy of semantic segmentation in remote sensing images has been increased significantly by deep convolutional neural networks, several limitations exist in standard models. First, for encoder–decoder architectures such as U-Net, the utilization of multiscale features causes the underuse of information, where low-level features and high-level features are concatenated directly without any refinement. Second, long-range dependencies of feature maps are insufficiently explored, resulting in suboptimal feature representations associated with each semantic class. Third, even though the dot-product attention mechanism has been introduced and utilized in semantic segmentation to model long-range dependencies, the large time and space demands of attention impede the actual usage of attention in application scenarios with large-scale input. This article proposed a multiattention network (MANet) to address these issues by extracting contextual dependencies through multiple efficient attention modules. A novel attention mechanism of kernel attention with linear complexity is proposed to alleviate the large computational demand in attention. Based on kernel attention and channel attention, we integrate local feature maps extracted by ResNet-50 with their corresponding global dependencies and reweight interdependent channel maps adaptively. Numerical experiments on two large-scale fine-resolution remote sensing datasets demonstrate the superior performance of the proposed MANet. Code is available athttps://github.com/lironui/Multi-Attention-Network.

Extracting relational triples from unstructured text is crucial for large-scale knowledge graph construction.However, few existing works excel in solving the overlapping triple problem where multiple relational triples in the same sentence share the same entities.In this work, we introduce a fresh perspective to revisit the relational triple extraction task and propose a novel cascade binary tagging framework (CASREL) derived from a principled problem formulation.Instead of treating relations as discrete labels as in previous works, our new framework models relations as functions that map subjects to objects in a sentence, which naturally handles the overlapping problem.Experiments show that the CAS-REL framework already outperforms state-ofthe-art methods even when its encoder module uses a randomly initialized BERT encoder, showing the power of the new tagging framework.It enjoys further performance boost when employing a pre-trained BERT encoder, outperforming the strongest baseline by 17.5 and 30.2 absolute gain in F1-score on two public datasets NYT and WebNLG, respectively.In-depth analysis on different scenarios of overlapping triples shows that the method delivers consistent performance gain across all these scenarios.The source code and data are released online 1 .