Abstract Deep learning based keypoint detectors can localize specific object (or body) parts well, but still fall short of general keypoint detection. Instead, few-shot keypoint detection (FSKD) is an underexplored yet more general task of localizing either base or novel keypoints, depending on the prompted support samples. In FSKD, how to build robust keypoint representations is the key to success. To this end, we propose an FSKD approach that models relations between keypoints. As keypoints are located on objects, we exploit a class-agnostic visual prior, i.e ., the unsupervised saliency map or DINO attentiveness map to obtain the region of focus within which we perform relation learning between object patches. The class-agnostic visual prior also helps suppress the background noise largely irrelevant to keypoint locations. Then, we propose a novel Visual Prior guided Vision Transformer (VPViT). The visual prior maps are refined by a bespoke morphology learner to include relevant context of objects. The masked self-attention of VPViT takes the adapted prior map as a soft mask to constrain the self-attention to foregrounds. As robust FSKD must also deal with the low number of support samples and occlusions, based on VPViT, we further investigate i) transductive FSKD to enhance keypoint representations with unlabeled data and ii) FSKD with masking and alignment (MAA) to improve robustness. We show that our model performs well in seven public datasets, and also significantly improves the accuracy in transductive inference and under occlusions. Source codes are available at https://github.com/AlanLuSun/VPViT .

In this work, we present a novel theoretical framework for analyzing and modeling protein fitness landscapes using spectral graph theory. By representing the protein sequence space as a generalized Hamming graph and studying its spectral properties, we derive a set of powerful tools for quantifying the ruggedness, epistasis, and other key characteristics of the landscape. We prove strong approximation and sampling results, showing that the landscape can be efficiently learned and optimized from limited and noisy data. Building on this foundation, we introduce Propagational Convolutional Neural Networks (PCNNs), a new class of inductive surrogate oracle. We provide rigorous theoretical guarantees on the generalization and convergence properties of PCNNs, using techniques from the Neural Tangent Kernel framework. Extensive experiments on real-world protein engineering tasks demonstrate the superiority of PCNNs over state-of-the-art methods, achieving higher fitness and better generalization from limited data.

Federated learning (FL) has emerged as a promising paradigm for privacy-preserving collaborative machine learning. However, extending FL to class incremental learning settings introduces three key challenges: 1) spatial heterogeneity due to non-IID data distributions across clients, 2) temporal heterogeneity due to sequential arrival of tasks, and 3) resource heterogeneity due to diverse client capabilities. Existing approaches generally address these challenges in isolation, potentially leading to interference between updates, catastrophic forgetting or excessive communication overhead. In this paper, we propose personalized Federated class-incremental parameter efficient fine-tuning with Mixture of Frequency aggregation (pFedMxF), a novel framework that simultaneously addresses all three heterogeneity challenges through frequency domain decomposition. Our key insight is that assigning orthogonal frequency components to different clients and tasks enables interference-free learning to be achieved with minimal communication costs. We further design an Auto-task Agnostic Classifier that automatically routes samples to task-specific classifiers while adapting to heterogeneous class distributions. We conduct extensive experiments on three benchmark datasets, comparing our approach with eight state-of-the-art methods. The results demonstrate that pFedMxF achieves comparable test accuracy, while requiring less model parameters and incurring significantly lower communication costs than baseline methods.

Continual learning requires models to learn tasks sequentially while maintaining a delicate balance between stability (retaining knowledge of previous tasks) and plasticity (adapting to new tasks). A key challenge is preventing interference between tasks which degrades performance when learning new tasks over previously learned tasks. Recent approaches leverage parameter-efficient fine-tuning (PEFT) which adapts pre-trained models by injecting a small number of learnable parameters. However, existing PEFT-based continual learning methods such as InfLoRA face fundamental limitations, i.e., they rely on complex optimization procedures to learn orthogonal task-specific spaces which is increasingly difficult as tasks accumulate. Thus, we propose a novel bilinear reformulation that fundamentally reimagines the task separation through fixed orthogonal bases. Our key insight is that by expanding the parameter space quadratically through two fixed bases, we can achieve "almost orthogonal" task subspaces probabilistically, eliminating the need for explicit interference elimination procedures. We provide theoretical guarantees that this approach reduces the probability of task interference from ${\mathcal{O}}\left({{{(k/d)}^2}}\right)$ to ${\mathcal{O}}\left({{{\left({k/{d^2}}\right)}^2}}\right)$, ensuring reliable task separation without complex optimization. Through extensive experiments on ImageNet-R, CIFAR-100, and DomainNet, we validate our theoretical bounds and demonstrate state-of-the-art performance with reduced parameter count. The code is available at: https://github.com/yifeiacc/BiLoRA.

Vision-Language Models (VLMs) such as CLIP excel at zero-shot classification due to large-scale pre-training but are vulnerable to adversarial examples. Adversarial fine-tuning robustifies zero-shot models by aligning prediction scores of individual adversaries with their clean counterparts, which typically overlooks intermediate adversarial samples along the adversarial trajectory crossing the decision boundary. Such intermediate adversaries and their vicinity produce informative representations capturing the decision boundary in detail. They can be improved by sampling adversarial candidates from simplices formed by joining two consecutive vertices on the adversarial trajectory and their clean counterpart. However, sampling simplices for adversaries is very costly. To train robust VLM, we overcome these limitations by Taylor expansion and formulating an upper-bound of alignment loss that depends on the Jacobian/Hessian obtained at clean samples. As regions between clean and intermediate adversarial samples capture a larger decision landscape, we robustify VLM by plausible adversaries from simplices by our closed-form formulation equivalent to infinite uniform sampling of the simplex. We obtain state-of-the-art robustness across 15 datasets and diverse vision-language tasks.

As foundation Vision-Language Models (VLMs) unlock fine-tuning on smaller datasets while leveraging large-scale pre-training data, machine unlearning becomes critical in addressing privacy concerns and regulatory compliance. Task vector, representing the difference between parameters of models fine-tuned with and without specific data, is a popular retraining-free unlearning strategy. However, we observe that task vectors exhibit substantial sensitivity to various fine-tuning configurations, resulting in unstable unlearning effectiveness that correlates negatively with the prediction-level variance. While aggregating multiple functions (e.g., VLM with classifier) whose parameters are represented by different task vectors reduces function variance and improves unlearning, the computational cost of obtaining numerous task vectors and aggregating functions is computationally high. Thus, in order to capture the space of task vectors induced by diverse fine-tuning strategies, we propose modeling it within the convex hull of $(Q-1)$-simplex whose vertices represent $Q$ task vectors. Although a function ensemble can be formed by sampling numerous task vectors from such a simplex, we derive a closed-form ensemble of an infinite number of functions whose parameters are uniformly sampled from the simplex, enabling efficient function-level task vector ensembling with enhanced unlearning performance. Extensive experiments and analyses across diverse datasets and scenarios demonstrate the efficacy of our method.

Parameter-efficient fine-tuning (PEFT) is essential for adapting large foundation models without excessive storage cost. However, current approaches such as LoRA treat each layer’s adaptation independently, overlooking correlations across layers. This independence causes the number of trainable parameters to grow linearly with model depth. We provide theoretical and empirical evidence that skip connections in transformers create smooth gradient propagation across layers. This smoothness leads to weight adaptations that concentrate most of their energy in low-frequency spectral components, especially along the layer dimension. Empirical analysis confirms this effect, showing that most of adaptation energy lies in low frequencies. Building on this insight, we propose CrossSpectra, which parameterizes all attention-weight adaptations $(Q, K, V)$ across layers as a single 3D tensor and represents them with sparse spectral coefficients ($\kappa_1, \kappa_2$). Using $\kappa_{1}$ non-zero coefficients within each layer’s frequency space and truncating to $\kappa_{2}$ frequencies across layers, CrossSpectra requires $\mathcal{O}(\kappa_{1}\kappa_{2})$ parameters instead of LoRA’s $\mathcal{O}(Lrd)$, where $L$ is the number of layers and $r$ the rank. Across natural-language and vision benchmarks, \methodname{} matches or surpasses baseline performance while using fewer parameters than LoRA, achieving only $0.36\%$ of LoRA’s parameter count when fine-tuning LLaMA-7B on instruction-following tasks. These results show that exploiting the \textbf{architectural smoothness of transformers} through spectral analysis yields major efficiency gains in PEFT.

Traditional Graph Self-Supervised Learning (GSSL) struggles to capture complex structural properties well. This limitation stems from two main factors: (1) the inadequacy of conventional Graph Neural Networks (GNNs) in representing sophisticated topological features, and (2) the focus of self-supervised learning solely on final graph representations. To address these issues, we introduce GenHopNet, a GNN framework that integrates a k-hop message-passing scheme, enhancing its ability to capture local structural information without explicit substructure extraction. We theoretically demonstrate that GenHopNet surpasses the expressiveness of the classical Weisfeiler-Lehman (WL) test for graph isomorphism. Furthermore, we propose a structural- and positional-aware GSSL framework that incorporates topological information throughout the learning process. This approach enables the learning of representations that are both sensitive to graph topology and invariant to specific structural and feature augmentations. Comprehensive experiments on graph classification datasets, including those designed to test structural sensitivity, show that our method consistently outperforms the existing approaches and maintains computational efficiency. Our work significantly advances GSSL's capability in distinguishing graphs with similar local structures but different global topologies.

Contrastive learning, a form of Self-Supervised Learning (SSL), typically consists of an alignment term and a regularization term. The alignment term minimizes the distance between the embeddings of a positive pair, while the regularization term prevents trivial solutions and expresses prior beliefs about the embeddings. As a widely used regularization technique, soft decorrelation has been employed by several non-contrastive SSL methods to avoid trivial solutions. While the decorrelation term is designed to address the issue of dimensional collapse, we find that it fails to achieve this goal theoretically and experimentally. Based on such a finding, we extend the soft decorrelation regularization to minimize the distance between the covariance matrix and an identity matrix. We provide a new perspective on the geometric distance between positive definite matrices to investigate why the soft decorrelation cannot efficiently solve the dimensional collapse. Furthermore, we construct a family of loss functions utilizing the Bregman Matrix Divergence (BMD), with the soft decorrelation representing a specific instance within this family. We prove that a loss function (LogDet) in this family can solve the issue of dimensional collapse. Our novel loss functions based on BMD exhibit superior performance compared to the soft decorrelation and other baseline techniques, as demonstrated by experimental results on graph and image datasets.

Although augmentations (e.g., perturbation of graph edges, image crops) boost the efficiency of Contrastive Learning (CL), feature level augmentation is another plausible, complementary yet not well researched strategy. Thus, we present a novel spectral feature argumentation for contrastive learning on graphs (and images). To this end, for each data view, we estimate a low-rank approximation per feature map and subtract that approximation from the map to obtain its complement. This is achieved by the proposed herein incomplete power iteration, a non-standard power iteration regime which enjoys two valuable byproducts (under mere one or two iterations): (i) it partially balances spectrum of the feature map, and (ii) it injects the noise into rebalanced singular values of the feature map (spectral augmentation). For two views, we align these rebalanced feature maps as such an improved alignment step can focus more on less dominant singular values of matrices of both views, whereas the spectral augmentation does not affect the spectral angle alignment (singular vectors are not perturbed). We derive the analytical form for: (i) the incomplete power iteration to capture its spectrum-balancing effect, and (ii) the variance of singular values augmented implicitly by the noise. We also show that the spectral augmentation improves the generalization bound. Experiments on graph/image datasets show that our spectral feature augmentation outperforms baselines, and is complementary with other augmentation strategies and compatible with various contrastive losses.

Few-shot learning (FSL) is popular due to its ability to adapt to novel classes. Compared with inductive few-shot learning, transductive models typically perform better as they leverage all samples of the query set. The two existing classes of methods, prototype-based and graph-based, have the disadvantages of inaccurate prototype estimation and sub-optimal graph construction with kernel functions, respectively. In this paper, we propose a novel prototype-based label propagation to solve these issues. Specifically, our graph construction is based on the relation between prototypes and samples rather than between samples. As prototypes are being updated, the graph changes. We also estimate the label of each prototype instead of considering a prototype be the class centre. On mini-ImageNet, tiered-ImageNet, CIFAR-FS and CUB datasets, we show the proposed method outperforms other state-of-the-art methods in transductive FSL and semi-supervised FSL when some un-labeled data accompanies the novel few-shot task.

Graph-based Collaborative Filtering (GCF) is widely used in personalized recommendation systems. However, GCF suffers from a fundamental problem where features tend to occupy the embedding space inefficiently (by spanning only a low-dimensional subspace). Such an effect is characterized in GCF by the embedding space being dominated by a few of popular items with the user embeddings highly concentrated around them. This enhances the so-called Matthew effect of the popularity bias where popular items are highly recommend whereas remaining items are ignored. In this paper, we analyze the above effect in GCF and reveal that the simplified graph convolution operation (typically used in GCF) shrinks the singular space of the feature matrix. As typical approaches (i.e., optimizing the uniformity term) fail to prevent the embedding space degradation, we propose a decorrelation-enhanced GCF objective that promotes feature diversity by leveraging the so-called principle of redundancy reduction in embeddings. However, unlike conventional methods that use the Euclidean geometry to relax hard constraints for decorrelation, we exploit non-Euclidean geometry. Such a choice helps maintain the range space of the matrix and obtain small condition number, which prevents the embedding space degradation. Our method outperforms contrastive-based GCF models on several benchmark datasets and improves the performance for unpopular items.

Few-shot learning (FSL) has received a lot of attention due to its remarkable ability to adapt to novel classes. Although many techniques have been proposed for FSL, they mostly focus on improving FSL backbones. Some works also focus on learning on top of the features generated by these backbones to adapt them to novel classes. We present an unsupErvised discriminAnt Subspace lEarning (EASE) that improves transductive few-shot learning performance by learning a linear projection onto a subspace built from features of the support set and the unlabeled query set in the test time. Specifically, based on the support set and the unlabeled query set, we generate the similarity matrix and the dissimilarity matrix based on the structure prior for the proposed EASE method, which is efficiently solved with SVD. We also introduce conStraIned wAsserstein MEan Shift clustEring (SIAMESE) which extends Sinkhorn K-means by incorporating labeled support samples. SIAMESE works on the features obtained from EASE to estimate class centers and query predictions. On the miniImageNet, tiered-ImageNet, CIFAR-FS, CUB and OpenMIC benchmarks, both steps significantly boost the performance in transductive FSL and semi-supervised FSL.

Graph contrastive learning (GCL) improves graph representation learning, leading to SOTA on various downstream tasks. The graph augmentation step is a vital but scarcely studied step of GCL. In this paper, we show that the node embedding obtained via the graph augmentations is highly biased, somewhat limiting contrastive models from learning discriminative features for downstream tasks.Thus, instead of investigating graph augmentation in the input space, we alternatively propose to perform augmentations on the hidden features (feature augmentation). Inspired by so-called matrix sketching, we propose COSTA, a novel Covariance-preServing feaTure space Augmentation framework for GCL, which generates augmented features by maintaining a "good sketch" of original features. To highlight the superiority of feature augmentation with COSTA, we investigate a single-view setting (in addition to multi-view one) which conserves memory and computations. We show that the feature augmentation with COSTA achieves comparable/better results than graph augmentation based models.

Graph contrastive learning attracts/disperses node representations for similar/dissimilar node pairs under some notion of similarity. It may be combined with a low-dimensional embedding of nodes to preserve intrinsic and structural properties of a graph. COLES, a recent graph contrastive method combines traditional graph embedding and negative sampling into one framework. COLES in fact minimizes the trace difference between the within-class scatter matrix encapsulating the graph connectivity and the total scatter matrix encapsulating negative sampling. In this paper, we propose a more essential framework for graph embedding, called Generalized Laplacian EigeNmaps (GLEN), which learns a graph representation by maximizing the rank difference between the total scatter matrix and the within-class scatter matrix, resulting in the minimum class separation guarantee. However, the rank difference minimization is an NP-hard problem. Thus, we replace the trace difference that corresponds to the difference of nuclear norms by the difference of LogDet expressions, which we argue is a more accurate surrogate for the NP-hard rank difference than the trace difference. While enjoying a lesser computational cost, the difference of LogDet terms is lower-bounded by the Affine-invariant Riemannian metric (AIRM) and Jesen-Bregman the LogDet Divergence (JBLD), and upper-bounded by AIRM scaled by the factor of $\sqrt{m}$. We show that GLEN offers favourable accuracy/scalability compared to state-of-the-art baselines.

Graph Neural Networks (GNNs) have achieved remarkable success by extending traditional convolution to learning on non-Euclidean data. The key to the GNNs is adopting the neural message-passing paradigm with two stages: aggregation and update. The current design of GNNs considers the topology information in the aggregation stage. However, in the updating stage, all nodes share the same updating function. The identical updating function treats each node embedding as i.i.d. random variables and thus ignores the implicit relationships between neighborhoods, which limits the capacity of the GNNs. The updating function is usually implemented with a linear transformation followed by a non-linear activation function. To make the updating function topology-aware, we inject the topological information into the non-linear activation function and propose Graph-adaptive Rectified Linear Unit (GReLU), which is a new parametric activation function incorporating the neighborhood information in a novel and efficient way. The parameters of GReLU are obtained from a hyperfunction based on both node features and the corresponding adjacent matrix. To reduce the risk of overfitting and the computational cost, we decompose the hyperfunction as two independent components for nodes and features respectively. We conduct comprehensive experiments to show that our plug-and-play GReLU method is efficient and effective given different GNN backbones and various downstream tasks.

Network embedding approaches have recently attracted considerable interest as they learn low-dimensional vector representations of nodes. Embeddings based on the matrix factorization are effective but they are usually computationally expensive due to the eigen-decomposition step. In this paper, we propose a Random RangE FInder based Network Embedding (REFINE) algorithm, which can perform embedding on one million of nodes (YouTube) within 30 seconds in a single thread. REFINE is 10x faster than ProNE, which is 10-400x faster than other methods such as LINE, DeepWalk, Node2Vec, GraRep, and Hope. Firstly, we formulate our network embedding approach as a skip-gram model, but with an orthogonal constraint, and we reformulate it into the matrix factorization problem. Instead of using randomized tSVD (truncated SVD) as other methods, we employ the Randomized Blocked QR decomposition to obtain the node representation fast. Moreover, we design a simple but efficient spectral filter for network enhancement to obtain higher-order information for node representation. Experimental results prove that REFINE is very efficient on datasets of different sizes (from thousand to million of nodes and edges) for node classification, while enjoying a good performance.

Fraud detection is extremely critical for e-commerce business platforms. Utilizing graph structure data and identifying unexpected dense subgraphs as suspicious is a category of commonly used fraud detection methods. Among them, spectral methods solve the problem efficiently but hurt the performance due to the relaxed constraints. Heuristic methods cannot be accelerated with parallel computation and fail to control the scope of returned suspicious nodes. These drawbacks affect the real-world applications of existing graph-based methods. In this paper, we propose an Ensemble based Fraud DETection (ENSEMFDET) method to scale up fraud detection in bipartite graphs. By oversampling the graph and solving the subproblems, the ensemble approach further votes suspicious nodes without sacrificing the prediction accuracy. Extensive experiments have been done on real transaction data from JD.com and demonstrate the effectiveness, practicability, and scalability of ENSEMFDET.

Graph contrastive learning attracts/disperses node representations for similar/dissimilar node pairs under some notion of similarity. It may be combined with a low-dimensional embedding of nodes to preserve intrinsic and structural properties of a graph. In this paper, we extend the celebrated Laplacian Eigenmaps with contrastive learning, and call them COntrastive Laplacian EigenmapS (COLES). Starting from a GAN-inspired contrastive formulation, we show that the Jensen-Shannon divergence underlying many contrastive graph embedding models fails under disjoint positive and negative distributions, which may naturally emerge during sampling in the contrastive setting. In contrast, we demonstrate analytically that COLES essentially minimizes a surrogate of Wasserstein distance, which is known to cope well under disjoint distributions. Moreover, we show that the loss of COLES belongs to the family of so-called block-contrastive losses, previously shown to be superior compared to pair-wise losses typically used by contrastive methods. We show on popular benchmarks/backbones that COLES offers favourable accuracy/scalability compared to DeepWalk, GCN, Graph2Gauss, DGI and GRACE baselines.

Learning to hash via generative models has became a promising paradigm for fast similarity search in document retrieval. The binary hash codes are treated as Bernoulli latent variables when training a variational autoencoder (VAE). However, the prior of discrete distribution (i.e., Bernoulli distribution) is short of some structure regularization to generate more effi-cient binary codes. In this paper, we present an end-to-end Wasserstein Autoencoder (WAE) for text hashing to avoid in-differentiable operators in the reparameterization trick, where the latent variables can be imposed to any discrete priors we can sample by using adversarial learning. Moreover, we can generate more efficient discrete codes by imposing a structural constraint on priors like the bit balance constraint. Our experiments show that the proposed model is competitive to the state of the art methods on both unsupervised and supervised scenarios.