Spiking Neural Networks (SNNs) are emerging as a promising alternative to Artificial Neural Networks (ANNs) due to their inherent energy efficiency. Owing to the inherent sparsity in spike generation within SNNs, the in-depth analysis and optimization of intermediate output spikes are often neglected. This oversight significantly restricts the inherent energy efficiency of SNNs and diminishes their advantages in spatiotemporal feature extraction, resulting in a lack of accuracy and unnecessary energy expenditure. In this work, we analyze the inherent spiking characteristics of SNNs from both temporal and spatial perspectives. In terms of spatial analysis, we find that shallow layers tend to focus on learning vertical variations, while deeper layers gradually learn horizontal variations of features. Regarding temporal analysis, we observe that there is not a significant difference in feature learning across different time steps. This suggests that increasing the time steps has limited effect on feature learning. Based on the insights derived from these analyses, we propose a Frequency-based Spatial-Temporal Attention (FSTA) module to enhance feature learning in SNNs. This module aims to improve the feature learning capabilities by suppressing redundant spike features. The experimental results indicate that the introduction of the FSTA module significantly reduces the spike firing rate of SNNs, demonstrating superior performance compared to state-of-the-art baselines across multiple datasets.
论文检索
输入标题、作者或关键词，从 2,893 篇学术成果中精准定位
The Information Bottleneck (IB) principle has emerged as a promising approach for enhancing the generalization, robustness, and interpretability of deep neural networks, demonstrating efficacy across image segmentation, document clustering, and semantic communication. Among IB implementations, the IB Lagrangian method, employing Lagrangian multipliers, is widely adopted. While numerous methods for the optimizations of IB Lagrangian based on variational bounds and neural estimators are feasible, their performance is highly dependent on the quality of their design, which is inherently prone to errors. To address this limitation, we introduce Structured IB, a framework for investigating potential structured features. By incorporating auxiliary encoders to extract missing informative features, we generate more informative representations. Our experiments demonstrate superior prediction accuracy and task-relevant information preservation compared to the original IB Lagrangian method, even with reduced network size.
In real-world scenarios, complex data such as multispectral images and multi-frame videos inherently exhibit robust low-rank property. This property is vital for multi-dimensional inverse problems, such as tensor completion, spectral imaging reconstruction, and multispectral image denoising. Existing tensor singular value decomposition (t-SVD) definitions rely on hand-designed or pre-given transforms, which lack flexibility for defining tensor nuclear norm (TNN). The TNN-regularized optimization problem is solved by the singular value thresholding (SVT) operator, which leverages the t-SVD framework to obtain the low-rank tensor. However, it's quite complicated to introduce SVT into deep neural network due to the numerical instability problem in solving the derivatives of the eigenvectors. In this paper, we introduce a novel data-driven generative low-rank t-SVD model based on the learnable orthogonal transform, which can be naturally solved under its representation. Prompted by the linear algebra theorem of the Householder transformation, our learnable orthogonal transform is achieved by constructing an endogenously orthogonal matrix adaptable to neural networks, optimizing it as arbitrary orthogonal matrices. Additionally, we propose a low-rank solver as a generalization of SVT, which utilizes an efficient representation of generative networks to obtain low-rank structures. Extensive experiments highlight its significant restoration enhancements.
In noisy partial label learning, each training sample is associated with a set of candidate labels, and the ground-truth label may be contained within this set. With the emergence of powerful pre-trained vision-language models, e.g. CLIP, it is natural to consider using these models to automatically label training samples instead of relying on laborious manual annotation. In this paper, we investigate the pipeline of learning with CLIP annotated noisy partial labels and propose a novel collaborative consistency regularization method, in which we simultaneously train two neural networks, which collaboratively purify training labels for each other, called Co-Pseudo-Labeling, and perform consistency regularization between label and representation levels. For instance-dependent noise that embodies the underlying patterns of the pre-trained model, our method employs multiple mechanisms to avoid overfitting to noisy annotations, effectively mines information from potentially noisy sample set while iteratively optimizing both representations and pseudo-labels during the training process. Comparison experiments with various kinds of annotations and weakly supervised methods, as well as other pre-trained model application methods demonstrates the effectiveness of method and the feasibility of incorporating weakly supervised learning into the distillation of pre-trained models.
JAQ: Joint Efficient Architecture Design and Low-Bit Quantization with Hardware-Software Co-Exploration
PDF ↗The co-design of neural network architectures, quantization precisions, and hardware accelerators offers a promising approach to achieving an optimal balance between performance and efficiency, particularly for model deployment on resource-constrained edge devices. In this work, we propose the JAQ Framework, which jointly optimizes the three critical dimensions. However, effectively automating the design process across the vast search space of those three dimensions poses significant challenges, especially when pursuing extremely low-bit quantization. Specifical, the primary challenges include: (1) Memory overhead in software-side: Low-precision quantization-aware training can lead to significant memory usage due to storing large intermediate features and latent weights for backpropagation, potentially causing memory exhaustion. (2) Search time-consuming in hardware-side: The discrete nature of hardware parameters and the complex interplay between compiler optimizations and individual operators make the accelerator search time-consuming. To address these issues, JAQ mitigates the memory overhead through a channel-wise sparse quantization (CSQ) scheme, selectively applying quantization to the most sensitive components of the model during optimization. Additionally, JAQ designs BatchTile, which employs a hardware generation network to encode all possible tiling modes, thereby speeding up the search for the optimal compiler mapping strategy. Extensive experiments demonstrate the effectiveness of JAQ, achieving approximately 7% higher Top-1 accuracy on ImageNet compared to previous methods and reducing the hardware search time per iteration to 0.15 seconds.
Long-tailed (LT) data distribution is common in multi-label image classification (MLC) and can significantly impact the performance of classification models. One reason is the challenge of learning unbiased instance representations (i.e. features) for imbalanced datasets. Additionally, the co-occurrence of head/tail classes within the same instance, along with complex label dependencies, introduces further challenges. In this work, we delve into this problem through the lens of neural collapse (NC). NC refers to a phenomenon where the last-layer features and classifier of a deep neural network model exhibit a simplex Equiangular Tight Frame (ETF) structure during its terminal training phase. This structure creates an optimal linearly separable state. However, this phenomenon typically occurs in balanced datasets but rarely applies to the typical imbalanced problem. To induce NC properties under Long-tailed multi-label classification (LT-MLC) conditions, we propose an approach named MLC-NC, which aims to learn high-quality data representations and improve the model’s generalization ability. Specifically, MLC-NC accounts for the fact that different labels correspond to different feature parts located in images. MLC-NC extracts class-wise features from each instance through a cross-attention mechanism. To guide the features toward the ETF structure, we introduce visual-semantic feature alignment with a fixed ETF structured label embedding, which helps to learn evenly distributed class centers. To reduce within-class feature variation, we introduce collapse calibration within a lower-dimensional feature space. To mitigate classification bias, we concatenate features and feed them into a binarized fixed ETF classifier. As an orthogonal approach to existing methods, MLC-NC can be seamlessly integrated into various frameworks. Extensive experiments on widely-used benchmarks demonstrate the effectiveness of our method.
Hybrid Decentralized Optimization: Leveraging Both First- and Zeroth-Order Optimizers for Faster Convergence
PDF ↗Distributed optimization is the standard way of speeding up machine learning training, and most of the research in the area focuses on distributed first-order, gradient-based methods. Yet, there are settings where some computationally-bounded nodes may not be able to implement first-order, gradient-based optimization, while they could still contribute to joint optimization tasks. In this paper, we initiate the study of hybrid decentralized optimization, studying settings where nodes with zeroth-order and first-order optimization capabilities co-exist in a distributed system, and attempt to jointly solve an optimization task over some data distribution. We essentially show that, under reasonable parameter settings, such a system can not only withstand noisier zeroth-order agents but can even benefit from integrating such agents into the optimization process, rather than ignoring their information. At the core of our approach is a new analysis of distributed optimization with noisy and possibly-biased gradient estimators, which may be of independent interest. Our results hold for both convex and non-convex objectives. Experimental results on standard optimization tasks confirm our analysis, showing that hybrid first-zeroth order optimization can be practical, even when training deep neural networks.
For general users, training a neural network from scratch is usually challenging and labor-intensive. Fortunately, neural network zoos enable them to find a well-performing model for directly use or fine-tuning it in their local environments. Although current model retrieval solutions attempt to convert neural network models into vectors to avoid complex multiple inference processes required for model selection, it is still difficult to choose a suitable model due to inaccurate vectorization and biased correlation alignment between the query dataset and models. From the perspective of knowledge consistency, i.e., whether the knowledge possessed by the model can meet the needs of query tasks, we propose a model retrieval scheme, named Know2Vec, that acts as a black-box retrieval proxy for model zoo. Know2Vec first accesses to models via a black-box interface in advance, capturing vital decision knowledge from models while ensuring their privacy. Next, it employs an effective encoding technique to transform the knowledge into precise model vectors. Secondly, it maps the user's query task to a knowledge vector by probing the semantic relationships within query samples. Furthermore, the proxy ensures the knowledge-consistency between query vector and model vectors within their alignment space, which is optimized through the supervised learning with diverse loss functions, and finally it can identify the most suitable model for a given task during the inference stage. Extensive experiments show that our Know2Vec achieves superior retrieval accuracy against the state-of-the-art methods in diverse neural network retrieval tasks.
Recent advancements in graph neural networks (GNNs) have highlighted the critical need of calibrating model predictions, with neighborhood prediction similarity recognized as a pivotal component. Existing studies suggest that nodes with analogous neighborhood prediction similarity often exhibit similar calibration characteristics. Building on this insight, recent approaches incorporate neighborhood similarity into node-wise temperature scaling techniques. However, our analysis reveals that this assumption does not hold universally. Calibration errors can differ significantly even among nodes with comparable neighborhood similarity, depending on their confidence levels. This necessitates a re-evaluation of existing GNN calibration methods, as a single, unified approach may lead to sub-optimal calibration. In response, we introduce Simi-Mailbox, a novel approach that categorizes nodes by both neighborhood similarity and their own confidence, irrespective of proximity or connectivity. Our method allows fine-grained calibration by employing group-specific temperature scaling, with each temperature tailored to address the specific miscalibration level of affiliated nodes, rather than adhering to a uniform trend based on neighborhood similarity. Extensive experiments demonstrate the effectiveness of our Simi-Mailbox across diverse datasets on different GNN architectures, achieving up to 13.79% error reduction compared to uncalibrated GNN predictions.
Optical neural networks (ONNs) have attracted great attention due to their low power consumption and high-speed processing. When training an ONN implemented on a chip with possible fabrication variations, the well-known backpropagation algorithm cannot be executed accurately because the perfect information inside the chip cannot be observed. Instead, we employ a black-box optimization method such as zeroth-order (ZO) optimization. In this paper, we first discuss how ONN parameters should be perturbed to search for better values in a black-box manner. Conventionally, parameter perturbations are sampled from a normal distribution with an identity covariance matrix. This is plausible if the parameters are not interrelated in a module, like a linear module of an ordinary neural network. However, this is not the best way for ONN modules with layered parameters, which are interrelated by optical paths. We then propose to perturb the parameters by a normal distribution with a special covariance matrix computed by our novel method. The covariance matrix is designed so that the perturbations appearing at the module output caused by the parameter perturbations become as isotropic as possible to uniformly search for better values. Experimental results show that the proposed method using the special covariance matrix significantly outperformed conventional methods.
Graduated optimization is a global optimization technique that is used to minimize a multimodal nonconvex function by smoothing the objective function with noise and gradually refining the solution. This paper experimentally evaluates the performance of the explicit graduated optimization algorithm with an optimal noise scheduling derived from a previous study and discusses its limitations. The evaluation uses traditional benchmark functions and empirical loss functions for modern neural network architectures. In addition, this paper extends the implicit graduated optimization algorithm, which is based on the fact that stochastic noise in the optimization process of SGD implicitly smooths the objective function, to SGD with momentum, analyzes its convergence, and demonstrates its effectiveness through experiments on image classification tasks with ResNet architectures.
Graph Neural Networks (GNNs) have shown impressive performance in graph representation learning, but they face challenges in capturing long-range dependencies due to their limited expressive power. To address this, Graph Transformers (GTs) were introduced, utilizing self-attention mechanism to effectively model pairwise node relationships. Despite their advantages, GTs suffer from quadratic complexity w.r.t. the number of nodes in the graph, hindering their applicability to large graphs. In this work, we present Graph-Enhanced Contextual Operator (GECO), a scalable and effective alternative to GTs that leverages neighborhood propagation and global convolutions to effectively capture local and global dependencies in quasiliniear time. Our study on synthetic datasets reveals that GECO reaches 169x speedup on a graph with 2M nodes w.r.t. optimized attention. Further evaluations on diverse range of benchmarks showcase that it scales to large graphs where traditional GTs often face memory and time limitations. Notably, GECO consistently achieves comparable or superior quality compared to baselines, improving the SOTA up to 4.5%, and offering a scalable and effective solution for large-scale graph learning.
Multi-criteria decision making (MCDM) and preference learning (PL) are crucial subfields of intelligent decision-making, both aiming to aid decision-makers (DMs) in selecting, classifying, or ranking alternatives. While MCDM and PL can complement each other to some extent, existing approaches combining MCDM and PL often struggle with large data volumes and complex relational information. To address this, we propose a novel approach called ID-GMLM that integrates graph models and large language models (LLMs) for intelligent decision-making. It reformulates decision-making as a high-parallelism ranking function in the graph domain, using graph neural networks (GNNs) to learn and understand complex relationships between alternatives or criteria, and LLMs to parse and quantify the preferences of DMs. ID-GMLM features a multi-task learning framework that optimizes the primary task of predicting alternative rankings while modeling criterion interactions through the auxiliary task. Additionally, ID-GMLM incorporates a parameter tuning network based on criterion weights and an attention network, allowing the model to adaptively adjust to the context of the current task and the evolving preferences of DMs. Experiments on benchmark datasets demonstrate that ID-GMLM achieves significant performance improvements, inheriting the interpretability and intuitive appeal of MCDM while leveraging the computational efficiency and high accuracy of PL.
Deep neural networks have become foundational to advancements in multiple domains, including recommendation systems, natural language processing, and so on. Despite their successes, these models often contain incompatible parameters that can be underutilized or detrimental to model performance, particularly when faced with specific, varying data distributions. Existing research excels in removing such parameters or merging the outputs of multiple different pretrained models. However, the former focuses on efficiency rather than performance, while the latter requires several times more computing and storage resources to support inference. In this paper, we set the goal to explicitly improve these incompatible parameters by leveraging the complementary strengths of different models, thereby directly enhancing the models without any additional parameters. Specifically, we propose Compatibility-aware Knowledge Integration (CKI), which consists of Parameter Compatibility Assessment and Parameter Splicing, which are used to evaluate the knowledge content of multiple models and integrate the knowledge into one model, respectively. The integrated model can be used directly for inference or for further fine-tuning. Extensive experiments on various recommendation and language datasets show that CKI can effectively optimize incompatible parameters under multiple tasks and settings to break through the training limit of the original model without increasing the inference cost.
Previous graph neural networks (GNNs) usually assume that the graph data is with clean labels for representation learning, but it is not true in real applications. In this paper, we propose a new multi-teacher distillation method based on bi-level optimization (namely BO-NNC), to conduct noisy node classification on the graph data. Specifically, we first employ multiple self-supervised learning methods to train diverse teacher models, and then aggregate their predictions through a teacher weight matrix. Furthermore, we design a new bi-level optimization strategy to dynamically adjust the teacher weight matrix based on the training progress of the student model. Finally, we design a label improvement module to improve the label quality. Extensive experimental results on real datasets show that our method achieves the best results compared to state-of-the-art methods.
Integrating Co-Training with Edge Discrimination to Enhance Graph Neural Networks Under Heterophily
PDF ↗Graph Neural Networks (GNNs) have recently achieved significant success in several graph-related tasks. However, traditional GNNs and their variants are constantly limited by the implicit homophily, assuming neighboring nodes belong to the same class. This results in weak performance on heterophilic graphs where most nodes are linked to neighbors of different classes. Despite the numerous attempts to adequately deal with heterophily, most methods still use the uniform propagation aggregation mechanism. In this paper, we argue that identifying neighbors with different class labels and exploiting them individually is crucial for heterophilic GNNs. We then propose a simple and efficient novel co-training approach, EG-GCN, which uses group aggregation to handle homophilic and heterophilic neighbors separately. In EG-GCN, we first use an edge discriminator to classify edges and split the neighborhood of every node into two parts. We then apply group graph convolution to the divided neighborhoods to obtain node representations. During training, we continuously optimize the edge discriminator to improve neighborhood partition and use the node classification results to identify highly confident unlabeled nodes to expand the edge training set. This co-training strategy enables both components to enhance each other mutually. Extensive experiments demonstrate that EG-GCN significantly outperforms the state-of-the-art approaches.
Optimizing spectral graph neural networks (GNNs) remains a critical challenge in the field, yet the underlying processes are not well understood. In this paper, we investigate the inherent differences between graph convolution parameters and feature transformation parameters in spectral GNNs and their impact on the optimization landscape. Our analysis reveals that these differences contribute to a poorly conditioned problem, resulting in suboptimal performance. To address this issue, we introduce the concept of the block condition number of the Hessian matrix, which characterizes the difficulty of poorly conditioned problems in spectral GNN optimization. We then propose an asymmetric learning approach, dynamically preconditioning gradients during training to alleviate poorly conditioned problems. Theoretically, we demonstrate that asymmetric learning can reduce block condition numbers, facilitating easier optimization. Extensive experiments on eighteen benchmark datasets show that asymmetric learning consistently improves the performance of spectral GNNs for both heterophilic and homophilic graphs. This improvement is especially notable for heterophilic graphs, where the optimization process is generally more complex than for homophilic graphs.
End-to-end training with global optimization have popularized graph neural networks (GNNs) for node classification, yet inadvertently introduced vulnerabilities to adversarial edge-perturbing attacks. Adversaries can exploit the inherent opened interfaces of GNNs' input and output, perturbing critical edges and thus manipulating the classification results. Current defenses, due to their persistent utilization of global-optimization-based end-to-end training schemes, inherently encapsulate the vulnerabilities of GNNs. This is specifically evidenced in their inability to defend against targeted secondary attacks. In this paper, we propose the Graph Agent Network (GAgN) to address the aforementioned vulnerabilities of GNNs. GAgN is a graph-structured agent network in which each node is designed as an 1-hop-view agent. Through the decentralized interactions between agents, they can learn to infer global perceptions to perform tasks including inferring embeddings, degrees and neighbor relationships for given nodes. This empowers nodes to filtering adversarial edges while carrying out classification tasks. Furthermore, agents' limited view prevents malicious messages from propagating globally in GAgN, thereby resisting global-optimization-based secondary attacks. We prove that single-hidden-layer multilayer perceptrons (MLPs) are theoretically sufficient to achieve these functionalities. Experimental results show that GAgN effectively implements all its intended capabilities and, compared to state-of-the-art defenses, achieves optimal classification accuracy on the perturbed datasets.
The core challenge of high-dimensional and expensive black-box optimization (BBO) is how to obtain better performance faster with little function evaluation cost. The essence of the problem is how to design an efficient optimization strategy tailored to the target task. This paper designs a powerful optimization framework to automatically learn the optimization strategies from the target or cheap surrogate task without human intervention. However, current methods are weak for this due to poor representation of optimization strategy. To achieve this, 1) drawing on the mechanism of genetic algorithm, we propose a deep neural network framework called B2Opt, which has a stronger representation of optimization strategies based on survival of the fittest; 2) B2Opt can utilize the cheap surrogate functions of the target task to guide the design of the efficient optimization strategies. Compared to the state-of-the-art BBO baselines, B2Opt can achieve multiple orders of magnitude performance improvement with less function evaluation cost.
Efficient Event-Based Semantic Segmentation via Exploiting Frame-Event Fusion: A Hybrid Neural Network Approach
PDF ↗Event cameras have recently been introduced into image semantic segmentation, owing to their high temporal resolution and other advantageous properties. However, existing event-based semantic segmentation methods often fail to fully exploit the complementary information provided by frames and events, resulting in complex training strategies and increased computational costs. To address these challenges, we propose an efficient hybrid framework for image semantic segmentation, comprising a Spiking Neural Network branch for events and an Artificial Neural Network branch for frames. Specifically, we introduce three specialized modules to facilitate the interaction between these two branches: the Adaptive Temporal Weighting (ATW) Injector, the Event-Driven Sparse (EDS) Injector, and the Channel Selection Fusion (CSF) module. The ATW Injector dynamically integrates temporal features from event data into frame features, enhancing segmentation accuracy by leveraging critical dynamic temporal information. The EDS Injector effectively combines sparse event data with rich frame features, ensuring precise temporal and spatial information alignment. The CSF module selectively merges these features to optimize segmentation performance. Experimental results demonstrate that our framework not only achieves state-of-the-art accuracy across the DDD17-Seg, DSEC-Semantic, and M3ED-Semantic datasets but also significantly reduces energy consumption, achieving a 65% reduction on the DSEC-Semantic dataset.