Federated learning (FL) has garnered considerable interest for its capability to learn from decentralized data sources. Given the increasing application of FL in decision-making scenarios, addressing fairness issues across different sensitive groups (e.g., female, male) in FL is crucial. Current research typically focus on facilitating fairness at each client's data (local fairness) or within the entire dataset across all clients (global fairness). However, existing approaches that focus exclusively on either global or local fairness fail to address two key challenges: (CH1) Under statistical heterogeneity, global fairness does not imply local fairness, and vice versa. (CH2) Achieving fairness under model-agnostic setting. To tackle the aforementioned challenges, this paper proposes a novel post-processing framework for achieving both Local and Global Fairness in the FL context, namely LoGoFair. To address CH1, LoGoFair endeavors to seek the Bayes optimal classifier under local and global fairness constraints, which strikes the optimal accuracy-fairness balance in the probabilistic sense. To address CH2, LoGoFair employs a model-agnostic federated post-processing procedure that enables clients to collaboratively optimize global fairness while ensuring local fairness, thereby achieving the optimal fair classifier within FL. Experimental results on three real-world datasets further illustrate the effectiveness of the proposed LoGoFair framework.
论文检索
输入标题、作者或关键词，从 11,272 篇学术成果中精准定位
The Large Vision-Language Model (LVLM) integrates computer vision and natural language processing techniques, offering substantial application potential. However, these models demand extensive resources during inference. Adaptive attention techniques can dynamically reduce computational redundancy and thus improve efficiency. Although current adaptive attention methods significantly reduce the memory requirements of Transformer-based language models, they are not tailored for LVLMs. We observe that LVLMs generate responses from both remote image tokens and local text tokens, and different modalities have different attention patterns. This observation inspires us to manage the attention for each modality separately. Specifically, for visual input, we store the cache of potentially useful information but only compute the most critical parts. For language input, we care more about local information. Based on our observation and analysis of vision-language attention patterns, we develop A-VL, a plug-and-play adaptive attention tailored for LVLM inference. Extensive evaluations on three vision-language tasks and five datasets show the effectiveness of our designs. Our approach A-VL outperforms existing adaptive attention methods in reducing memory usage and computational load without compromising performance.
QiMLP: Quantum-inspired Multilayer Perceptron with Strong Correlation Mining and Parameter Compression
PDF ↗Multilayer Perceptron (MLP) is a simple practice of Neural Network (NN) and the cornerstone of research and development of deep learning. Each neuron is connected to all neurons in the previous layer and implements a non-linear mapping through activation functions. MLP can learn complex non-linear relationships among features through the superposition of multiple hidden layers, but it still cannot discover the inherent strong correlation among features. The reason is that each neuron uses a simple weighted summation method to organize all the neurons in the previous layer. Inspired by quantum theory, this paper builds a non-linear NN layer that can mine strong correlations among features based on multi-body quantum systems, and then constructs a multi-layer perceptron, called Quantum-inspired MLP (QiMLP). It is conceivable that QiMLP will have important inspirational significance in reshaping machine learning, deep learning and large language models. We theoretically analyzed the basis for QiMLP to mine strong correlations among features, and implemented experiments on multiple classic deep learning datasets. Experimental results verify that QiMLP not only learns strong correlations among features, but also significantly reduces the number of parameters with hundreds of times improvement.
Visual language models like Contrastive Language-Image Pretraining (CLIP) have shown impressive performance in analyzing natural images with language information. However, these models often encounter challenges when applied to specialized domains such as remote sensing due to the limited availability of image-text pairs for training. To tackle this issue, we introduce DiffCLIP, a novel framework that extends CLIP to effectively convey comprehensive language-driven semantic information for accurate classification of high-dimensional multimodal remote sensing images. DiffCLIP is a few-shot learning method that leverages unlabeled images for pretraining. It employs unsupervised mask diffusion learning to capture the distribution of diverse modalities without requiring labels. The modality-shared image encoder maps multimodal data into a unified subspace, extracting shared features with consistent parameters across modalities. A well-trained image encoder further enhances learning by aligning visual representations with class-label text information from CLIP. By integrating these approaches, DiffCLIP significantly boosts CLIP performance using a minimal number of image-text pairs. We evaluate DiffCLIP on widely used high-dimensional multimodal datasets, demonstrating its effectiveness in addressing few-shot annotated classification tasks. DiffCLIP achieves an overall accuracy improvement of 10.65% across three remote sensing datasets compared with CLIP, while utilizing only 2-shot image-text pairs.
Knowledge distillation (KD) is a model compression technique that transfers knowledge from a large teacher model to a smaller student model to enhance its performance. Existing methods often assume that the student model is inherently inferior to the teacher model. However, we identify that the fundamental issue affecting student performance is the bias transferred by the teacher. Current KD frameworks transmit both right and wrong knowledge, introducing bias that misleads the student model. To address this issue, we propose a novel strategy to rectify bias and greatly improve the student model's performance. Our strategy involves three steps: First, we differentiate knowledge and design a bias elimination method to filter out biases, retaining only the right knowledge for the student model to learn. Next, we propose a bias rectification method to rectify the teacher model's wrong predictions, fundamentally addressing bias interference. The student model learns from both the right knowledge and the rectified biases, greatly improving its prediction accuracy. Additionally, we introduce a dynamic learning approach with a loss function that updates weights dynamically, allowing the student model to quickly learn right knowledge-based easy tasks initially and tackle hard tasks corresponding to biases later, greatly enhancing the student model's learning efficiency. To the best of our knowledge, this is the first strategy enabling the student model to surpass the teacher model. Experiments demonstrate that our strategy, as a plug-and-play module, is versatile across various mainstream KD frameworks.
Towards Better Spherical Sliced-Wasserstein Distance Learning with Data-Adaptive Discriminative Projection Direction
PDF ↗Spherical Sliced-Wasserstein (SSW) has recently been proposed to measure the discrepancy between spherical data distributions in various fields, such as geology, medical domains, computer vision, and deep representation learning. However, in the original SSW, all projection directions are treated equally, which is too idealistic and cannot accurately reflect the importance of different projection directions for various data distributions. To address this issue, we propose a novel data-adaptive Discriminative Spherical Sliced-Wasserstein (DSSW) distance, which utilizes a projected energy function to determine the discriminative projection direction for SSW. In our new DSSW, we introduce two types of projected energy functions to generate the weights for projection directions with complete theoretical guarantees. The first type employs a non-parametric deterministic function that transforms the projected Wasserstein distance into its corresponding weight in each projection direction. This improves the performance of the original SSW distance with negligible additional computational overhead. The second type utilizes a neural network-induced function that learns the projection direction weight through a parameterized neural network based on data projections. This further enhances the performance of the original SSW distance with less extra computational overhead. Finally, we evaluate the performance of our proposed DSSW by comparing it with several state-of-the-art methods across a variety of machine learning tasks, including gradient flows, density estimation on real earth data, and self-supervised learning.
Classical neural ODEs trained with explicit methods are intrinsically limited by stability, crippling their efficiency and robustness for stiff learning problems that are common in graph learning and scientific machine learning. We present a semi-implicit neural ODE approach that exploits the partitionable structure of the underlying dynamics. Our technique leads to an implicit neural network with significant computational advantages over existing approaches because of enhanced stability and efficient linear solves during time integration. We show that our approach outperforms existing approaches on a variety of applications including graph classification and learning complex dynamical systems. We also demonstrate that our approach can train challenging neural ODEs where both explicit methods and fully implicit methods are intractable.
Leveraging First and Zeroth-Order Gradient to Address Imbalanced Black-Box Prompt Tuning via Minimax Optimization
PDF ↗Black-box prompt tuning has become a prevalent parameter-efficient paradigm that leverages the capabilities of large language models (LLMs) for customized applications in specific downstream tasks. In practical scenarios, downstream tasks frequently involve data distributions that are heavily imbalanced. Such imbalances tend to impair performance, causing severe performance collapse in minority classes. Conducting effective imbalanced black-box prompt tuning to mitigate the adverse effects of imbalanced data distribution on prompt performance remains a significant challenge. In this paper, we propose black-box prompt tuning with first and zeroth order gradient (BPT-FZG) for handling the imbalanced data. Specifically, BPT-FZG introduces AUC maximization as the objective for prompt tuning and equivalently formulates it as a nonconvex-concave saddle point problem to avoid the construction of sample pairs from opposite classes. Indeed, BPT-FZG optimizes the latent representation of the continuous prompt in the low-dimensional subspace with AUC loss and leverages the first and zeroth order gradients alternately to update the parameters. Furthermore, we establish the theoretical convergence guarantee for BPT-FZG under common assumptions, showing that our method can find a stationary point of the objective function. Our experiments on RoBERTa-large, GPT2-XL, and Llama3 show that BPT-FZG achieves improvement on various imbalanced datasets, emphasizing the effectiveness of our methods.
The great challenge of handwritten mathematical expression recognition (HMER) is the complex structures of the expressions, which are directly related to the symbol spatial positions. Existing HMER methods typically employ attention mechanisms in the decoder of their models to implicitly perceive the symbol positions, or employ symbol counting and tree-based strategies to model the symbol spatial relation. However, these methods still cannot effectively capture the structural information of formulas, thus negatively impacting the symbol decoding in HMER. To deal with this problem and enhance the HMER performance, this paper proposes a novel auxiliary task, namely predicting the symbol spatial distribution map of handwritten expression images. On such basis, this paper designs a symbol spatial-aware network (SSAN) for this task, which is jointly optimized with the HMER model. Specifically, considering the similarity of the symbol spatial positions between the handwritten mathematical expression images and their corresponding printed templates, we obtain the symbol spatial distribution map by first generating printed templates from LaTeX ground-truth for handwritten formula images and then replacing the connected components of printed templates with 2D Gaussian distribution maps of the same size. Meanwhile, due to the loose alignment of the symbol spatial positions between handwritten and printed formula images, and misclassification of similar symbols, we further propose a coarse-to-fine alignment strategy and an attention-guided symbol masking strategy in SSAN to tackle these issues. Extensive experiments demonstrate that SSAN significantly improves the recognition performance of the HMER models, and the proposed auxiliary tasks are more effective in enhancing HMER performance than existing auxiliary tasks.
Bagging-Expert Network for Multi-Task Learning: A Depolarization Solution in Multi-Gate Mixture-of-Experts
PDF ↗Multi-task learning (MTL) is widely utilized across a variety of real-world applications, including recommendation systems. For instance, in the field of e-commerce, MTL is commonly employed to simultaneously model click, conversion, and user dwelling time. Among a various of MTL models, the Multi-gate Mixture-of-Experts (MMoE) has gained significant popularity. However, MMoE suffers from the polarization issue during training, where the weights of certain experts tend to converge towards 0. To address this issue, we propose a novel method called Bagging-Expert network (BEnet) for multi-task learning. BEnet effectively mitigates the problem of polarization and achieves excellent performance in multi-task learning. It incorporates a bagging layer and an attention mechanism to encourage experts focusing on diverse knowledge domains. Simultaneously, polarization is avoided as different experts execute respective duties and specialize in distinct domains. Experimental results on real-world datasets demonstrate that BEnet has strong robustness and outperforms other state-of-the-art (SOTA) MTL methods.
As functional data assumes a central role in contemporary data analysis, the search for meaningful dimension reduction becomes critical due to its inherent infinite-dimensional structure. Traditional methods, such as Functional Principal Component Analysis (FPCA), adeptly explore the overarching structures within the functional data. However, these methods may not sufficiently identify low-dimensional representations that are specific or enriched in a foreground dataset (case or treatment group) relative to a background dataset (control group). This limitation becomes critical in scenarios where the foreground dataset, such as a specific treatment group in biomedical applications, contains unique patterns or trends that are not as pronounced in the background dataset. Addressing this gap, we propose Contrastive Functional Principal Component Analysis (CFPCA), a method designed to spotlight low-dimensional structures unique to or enriched in the foreground dataset relative to the background counterpart. We supplement our method with theoretical guarantees on CFPCA estimates supported by multiple simulations. Through a series of applications, CFPCA successfully identifies these foreground-specific structures, thereby revealing distinct patterns and trends that traditional FPCA overlooks.
Machine unlearning without access to real data distribution is challenging. The existing method based on data-free distillation achieved unlearning by filtering out synthetic samples containing forgetting information but struggled to distill the retaining-related knowledge efficiently. In this work, we analyze that such a problem is due to over-filtering, which reduces the synthesized retaining-related information. We propose a novel method, Inhibited Synthetic PostFilter (ISPF), to tackle this challenge from two perspectives: First, the Inhibited Synthetic, by reducing the synthesized forgetting information; Second, the PostFilter, by fully utilizing the retaining-related information in synthesized samples. Experimental results demonstrate that the proposed ISPF effectively tackles the challenge and outperforms existing methods.
Machine learning models are vulnerable to both security attacks (e.g., adversarial examples) and privacy attacks (e.g., private attribute inference). We take the first step to mitigate both the security and privacy attacks, and maintain task utility as well. Particularly, we propose an information-theoretic framework to achieve the goals through the lens of representation learning, i.e., learning representations that are robust to both adversarial examples and attribute inference adversaries. We also derive novel theoretical results under our framework, e.g., the inherent trade-off between adversarial robustness/utility and attribute privacy, and guaranteed attribute privacy leakage against attribute inference adversaries.
Text-Guided Nonverbal Enhancement Based on Modality-Invariant and -Specific Representations for Video Speaking Style Recognition
PDF ↗Video speaking style recognition (VSSR) aims to classify different types of conversations in videos, contributing significantly to understanding human interactions. A significant challenge in VSSR is the inherent similarity among conversation videos, which makes it difficult to distinguish between different speaking styles. Existing VSSR methods commit to providing available multimodal information to enhance the differentiation of conversation videos. Nevertheless, treating each modality equally leads to a suboptimal result for these methods due to text is inherently more aligned with conversation understanding compared to nonverbal modalities. To address this issue, we propose a text-guided nonverbal enhancement method, TNvE, which is composed of two core modules: 1) a text-guided nonverbal representation selection module employs cross-modal attention based on modality-invariant representations, picking out critical nonverbal information via textual guide; and 2) a modality-invariant and -specific representation decoupling module incorporates modality-specific representations and decouples them from modality-invariant representations, enabling a more comprehensive understanding of multimodal data. The former module encourages multimodal representations close to each other, while the latter module provides unique characteristics of each modality as a supplement. Extensive experiments are conducted on long-form video understanding datasets to demonstrate that TNvE is highly effective for VSSR, achieving a new state-of-the-art.
Existing class incremental learning is mainly designed for single-label classification task, which is ill-equipped for multi-label scenarios due to the inherent contradiction of learning objectives for samples with incomplete labels. We argue that the main challenge to overcome this contradiction in multi-label class-incremental learning (MLCIL) lies in the model's inability to clearly distinguish between known and unknown knowledge. This ambiguity hinders the model's ability to retain historical knowledge, master current classes, and prepare for future learning simultaneously. In this paper, we target at specifying what is known or not to accommodate Historical, Current, and Prospective knowledge for MLCIL and propose a novel framework termed as HCP. Specifically, (i) we clarify the known classes by dynamic feature purification and recall enhancement with distribution prior, enhancing the precision and retention of known information. (ii) We design prospective knowledge mining to probe the unknown, preparing the model for future learning. Extensive experiments validate that our method effectively alleviates catastrophic forgetting in MLCIL, surpassing the previous state-of-the-art by 3.3% on average accuracy for MS-COCO B0-C10 setting without replay buffers.
Blend the Separated: Mixture of Synergistic Experts for Data-Scarcity Drug-Target Interaction Prediction
PDF ↗Drug-target interaction prediction (DTI) is essential in various applications including drug discovery and clinical application. There are two perspectives of input data widely used in DTI prediction: Intrinsic data represents how drugs or targets are constructed, and extrinsic data represents how drugs or targets are related to other biological entities. However, any of the two perspectives of input data can be scarce for some drugs or targets, especially for those unpopular or newly discovered. Furthermore, ground-truth labels for specific interaction types can also be scarce. Therefore, we propose the first method to tackle DTI prediction under input data and/or label scarcity. To make our model functional when only one perspective of input data is available, we design two separate experts to process intrinsic and extrinsic data respectively and fuse them adaptively according to different samples. Furthermore, to make the two perspectives complement each other and remedy label scarcity, two experts synergize with each other in a mutually supervised way to exploit the enormous unlabeled data. Extensive experiments on 3 real-world datasets under different extents of input data scarcity and/or label scarcity demonstrate our model outperforms states of the art significantly and steadily, with a maximum improvement of 53.53%. We also test our model without any data scarcity and it still outperforms current methods.
Graph Neural Networks (GNNs) have recently gained widespread attention as a successful tool for analyzing graph-structured data. However, imperfect graph structure with noisy links lacks enough robustness and may damage graph representations, therefore limiting the GNNs' performance in practical tasks. Moreover, existing generative architectures fail to fit discriminative graph-related tasks. To tackle these issues, we introduce an unsupervised method based on a joint of generative training and discriminative training to learn graph structure and representation, aiming to improve the discriminative performance of generative models. We propose an Energy-based Contrastive Learning (ECL) guided Graph Structure Refinement (GSR) framework, denoted as ECL-GSR. To our knowledge, this is the first work to combine energy-based models with contrastive learning for GSR. Specifically, we leverage ECL to approximate the joint distribution of sample pairs, which increases the similarity between representations of positive pairs while reducing the similarity between negative ones. Refined structure is produced by augmenting and removing edges according to the similarity metrics among node representations. Extensive experiments demonstrate that ECL-GSR outperforms the state-of-the-art on eight benchmark datasets in node classification. ECL-GSR achieves faster training with fewer samples and memories against the leading baseline, highlighting its simplicity and efficiency in downstream tasks.
Large-scale pre-trained Vision Transformer (ViT) models have demonstrated remarkable performance on visual tasks but are computationally expensive to transfer to downstream tasks. Parameter-Efficient Fine-Tuning (PEFT) offers a promising transferring approach by updating only a subset of parameters. However, PEFT's effectiveness is hindered by discrepancies between pre-training and downstream tasks in terms of object scale and granularity. Downstream tasks often focus on finer-grained and more specialized recognition, requiring more detailed features. The diversity of feature scales of existing PEFT methods for ViT is limited. To address this, we propose a novel PEFT method named Wavelet-based multi-Scale Tuning (WST), which learns multi-scale features in a simple and efficient way. WST introduces a parallel fine-tuning patch embedding branch with a smaller patch size than the pre-trained model to capture finer-grained features. Furthermore, to handle the computational challenge from the resulting longer token sequence, WST designs wavelet fine-tuning blocks that balance both efficiency and performance. In the block, wavelet transform enables invertible and lossless down-sampling of the longer token sequence, aligning it with that of the backbone, and two lightweight linear mappings are employed to learn task-specific features. This design facilitates efficient multi-scale information exchange between the pre-trained backbone and fine-tuning branch. Extensive experiments on transfer learning demonstrate the promising performance and efficiency of our WST.
As data and computational resources continue to expand, incorporating a variety of knowledge during the pre-training phase enhances large models, providing them with strong zero-shot capabilities. Due to the alignment of modal features by visual language models, zero-shot image captioning no longer necessitates pre-training on paired image-text labeled data, enabling accurate text description generation for images not encountered before. While recent research focuses on methods utilizing entity retrieval as anchors to bridge the gap between different modalities, these approaches often fall short of thoroughly analyzing the impact of entity retrieval recall on the zero-shot generation capabilities. To address this issue, we propose MERCap, a zero-shot image captioning method employing Multi-type Entity representation Retrieval. More specifically, we first approximate image representation using the CLIP representation of text and Gaussian noise to address the modality gap. Then, we train a GPT-2 decoder to reconstruct text using entities as hard prompts and CLIP representations as soft prompts. Additionally, we construct a domain-specific entity set, assigning multiple representations to each entity and refining their representation vectors through contrastive learning. During inference, we retrieve entities and input them into the decoder to generate corresponding captions. Extensive experiments validate that our approach is efficient, achieving a new state-of-the-art level in cross-domain captioning and demonstrating strong competitiveness in in-domain captioning compared to existing methods.
Large Language Models (LLMs) have revolutionized natural language processing tasks. However, their practical application is constrained by substantial memory and computational demands. Post-training quantization (PTQ) is considered an effective method to accelerate LLM inference. Despite its growing popularity in LLM model compression, PTQ deployment faces two major challenges. First, low-bit quantization leads to performance degradation. Second, restricted by the limited integer computing unit type on GPUs, quantized matrix operations with different precisions cannot be effectively accelerated. To address these issues, we introduce a novel arbitrary-bit quantization algorithm and inference framework, ABQ-LLM. It achieves superior performance across various quantization settings and enables efficient arbitrary-precision quantized inference on the GPU. ABQ-LLM introduces several key innovations: (1) a distribution correction method for transformer blocks to mitigate distribution differences caused by full quantization of weights and activations, improving performance at low bit-widths. (2) the bit balance strategy to counteract performance degradation from asymmetric distribution issues at very low bit-widths (e.g., 2-bit). (3) an innovative quantization acceleration framework that reconstructs the quantization matrix multiplication of arbitrary precision combinations based on BTC (Binary TensorCore) equivalents, gets rid of the limitations of INT4/INT8 computing units. ABQ-LLM can convert each component bit width gain into actual acceleration gain, maximizing performance under mixed precision(e.g., W6A6, W2A8). Based on W2*A8 quantization configuration on LLaMA-7B model, it achieved a WikiText2 perplexity of 7.59 (2.17⬇ vs 9.76 in AffineQuant). Compared to SmoothQuant, we realized 1.6x acceleration improvement and 2.7x memory compression gain.