Warning: This paper contains offensive content that may disturb some readers. Vision-language models (VLMs) demonstrate strong multimodal capabilities but have been found to be more susceptible to generating harmful content compared to their backbone large language models (LLMs). Our investigation reveals that the integration of images significantly shifts the model's internal activations during the forward pass, diverging from those triggered by textual input. Moreover, the safety alignments of LLMs embedded within VLMs are not sufficiently robust to handle the activations discrepancies, making the models vulnerable to even the simplest jailbreaking attacks. To address this issue, we propose an internal activation revision approach that efficiently revises activations during generation, steering the model toward safer outputs. Our framework incorporates revisions at both the layer and head levels, offering control over the model's generation at varying levels of granularity. In addition, we explore three strategies for constructing positive and negative samples and two approaches for extracting revision vectors, resulting in different variants of our method. Comprehensive experiments demonstrate that the internal activation revision method significantly improves the safety of widely used VLMs, reducing attack success rates by an average of 48.94%, 34.34%, 43.92%, and 52.98% on SafeBench, Safe-Unsafe, Unsafe, and MM-SafetyBench, respectively, while minimally impacting model helpfulness.

In this paper, we explain the mechanism of bilinear pooling as a module of hard sample generation, and find that bilinear pooling significantly expands variances of the first-order vectors when it produces discriminative bilinear features. In conjunction with the extremely high dimensionality of the obtained bilinear features, those variances lead to overfitting in subsequent learning models. To solve this issue, we construct a bi-level optimization problem, where the high-level problem is the supervised classification loss, and the low-level problem is the principal component analysis (PCA). Then, we find that PCA on bilinear features is equivalent to spectral clustering, which allows us to mathematically prove that the first $\log _{2}(C)$log2(C) principal components can support the discriminant information of $C$C classes. By removing the rest principal components, the dimensionality and variances are simultaneously reduced. To the best of our knowledge, this is the first work providing a lower bound for dimension reduction for bilinear pooling. However, the PCA projection matrix $\mathbf{L}$L is prone to overfitting due to having many parameters. To address this issue, we propose a rank-$k$k general bilinear projection (RK-GBP) that decomposes $\mathbf{L}$L into two small matrices $\mathbf{U}$U and $\mathbf{V}$V, whose learnable parameters are smaller. Different from traditional bilinear projections used in factorized bilinear pooling (FBiP), our RK-GBP can preserve the orthogonality of columns in $\mathbf{L}$L by constraining the orthogonality of columns in $\mathbf{U}$U and $\mathbf{V}$V. For computational efficiency, we relax the PCA in the low-level task into a dictionary learning problem, obtaining the rank-$k$k orthogonal factorization bilinear pooling (RK-OFBP). The RK-OFBP can be considered as a general form of current factorization bilinear pooling methods (e.g., Hadamard product-based ones). Finally, we evaluate our approach on fine-grained images and large-scale datasets, demonstrating that our proposed method not only produces extremely low-dimensional features but also outperforms other methods in classification tasks. For example, our RK-OFBP can employ 32-dimensional vectors to achieve comparable results to B-CNN (Lin, 2015) (dimension: 512*512) for the 200-class classification task.

In this paper, we find that the training of Normalizing Flows (NFs) are easily affected by the outliers and a small number (or high dimensionality) of training samples. To solve this problem, we propose a Kullback–Leibler (KL) divergence regularization on the Jacobian matrix of NFs. We prove that such regularization is equivalent to adding a set of samples whose covariance matrix is the identity matrix to the training set. Thus, it reduces the negative influence of the outliers and the small sample number on the estimation of the covariance matrix, simultaneously. Therefore, our regularization makes the training of NFs robust. Ultimately, we evaluate the performance of NFs on out-of-distribution (OoD) detection tasks. The excellent results obtained demonstrate the effectiveness of the proposed regularization term. For example, with the help of the proposed regularization, the OoD detection score increases at most 30% compared with the one without the regularization.

In this paper, we reveal that metric learning would suffer from serious inseparable problem if without informative sample mining. Since the inseparable samples are often mixed with hard samples, current informative sample mining strategies used to deal with inseparable problem may bring up some side-effects, such as instability of objective function, etc. To alleviate this problem, we propose a novel distance metric learning algorithm, named adaptive neighborhood metric learning (ANML). In ANML, we design two thresholds to adaptively identify the inseparable similar and dissimilar samples in the training procedure, thus inseparable sample removing and metric parameter learning are implemented in the same procedure. Due to the non-continuity of the proposed ANML, we develop an ingenious function, named log-exp mean function to construct a continuous formulation to surrogate it, which can be efficiently solved by the gradient descent method. Similar to Triplet loss, ANML can be used to learn both the linear and deep embeddings. By analyzing the proposed method, we find it has some interesting properties. For example, when ANML is used to learn the linear embedding, current famous metric learning algorithms such as the large margin nearest neighbor (LMNN) and neighbourhood components analysis (NCA) are the special cases of the proposed ANML by setting the parameters different values. When it is used to learn deep features, the state-of-the-art deep metric learning algorithms such as Triplet loss, Lifted structure loss, and Multi-similarity loss become the special cases of ANML. Furthermore, the log-exp mean function proposed in our method gives a new perspective to review the deep metric learning methods such as Prox-NCA and N-pairs loss. At last, promising experimental results demonstrate the effectiveness of the proposed method.

The amount of matrix data has increased rapidly nowadays. How to classify matrix data efficiently is an important issue. In this paper, by discovering the shortages of 2-D linear discriminant analysis and 2-D logistic regression, a novel 2-D framework named rank- 2-D multinomial logistic regression (2DMLR-RK) is proposed. The 2DMLR-RK is designed for a multiclass matrix classification problem. In the proposed framework, each category is modeled by a left projection matrix and a right projection matrix with rank . The left projection matrices capture the row information of matrix data, and the right projection matrices acquire the column information. We identify the parameter plays the role of balancing the capacity of learning and generalization of the 2DMLR-RK. In addition, we develop an effective framework for solving the proposed nonconvex optimization problem. The convergence, initialization, and computational complexity are discussed. Extensive experiments on various types of data sets are conducted. Comparing with 1-D methods, 2DMLR-RK not only achieves a better classification accuracy, but also costs less computation time. Comparing with other state-of-the-art 2-D methods, the 2DMLR-RK achieves a better performance for matrix data classification.

With the development of the information technology, the amount of data, e.g. text, image and video, has been increased rapidly. Efficiently clustering those large scale data sets is a challenge. To address this problem, this paper proposes a novel co-clustering method named bilateral k-means algorithm (BKM) for fast co-clustering. Different from traditional k-means algorithms, the proposed method has two indicator matrices P and Q and a diagonal matrix S to be solved, which represent the cluster memberships of samples and features, and the co-cluster centres, respectively. Therefore, it could implement different clustering tasks on the samples and features simultaneously. We also introduce an effective approach to solve the proposed method, which involves less multiplication. The computational complexity is analyzed. Extensive experiments on various types of data sets are conducted. Compared with the state-of-the-art clustering methods, the proposed BKM not only has faster computational speed, but also achieves promising clustering results.

We introduce a novel supervised metric learning algorithm named parameter free large margin nearest neighbor (PFLMNN) which can be seen as an improvement of the classical large margin nearest neighbor (LMNN) algorithm. The contributions of our work consist of two aspects. First, our method discards the costterm which shrinks the distances between inquiry input and its k target neighbors (the k nearest neighbors with same labels as inquiry input) in LMNN, and only focuses on improving the action to push the imposters (the samples with different labels form the inquiry input) apart out of the neighborhood of inquiry. As a result, our method does not have the parameter needed to tune on the validating set, which makes it more convenient to use. Second, by leveraging the geometry information of the imposters, we construct a novel cost function to penalize the smalldistances between each inquiry and its imposters. Different from LMNN considering every imposter located in the neighborhood of each inquiry, our method only takes care of the nearest imposters. Because when the nearest imposter is pushed out of the neighborhood of its inquiry, other imposters would be all out. In this way, the constraints in our model are much less than that of LMNN, which makes our method much easier to find the optimal distance metric. Consequently, our method not only learns a better distance metric than LMNN, but also runs faster than LMNN. Extensive experiments on different data sets with various sizes and difficulties are conducted, and the results have shown that, compared with LMNN, PFLMNN achieves better classification results.

Matrices are common forms of data that are encountered in a wide range of real applications. How to classify this kind of data is an important research topic. In this paper, we propose a novel distance metric learning method named two dimensional large margin nearest neighbor (2DLMNNN), for improving the performance of k nearest neighbor (KNN) classifier in matrix classification. In the proposed method, left and right projection matrices are employed to define the matrix-based Mahalanobis distance, which is used to construct the objective aimed at separating points in different classes by a large margin. The parameters in those two projection matrices are much less than that in its vector-based counterpart, thus our method reduces the risks of overfitting. We also introduce a framework for solving the proposed 2DLMNN. The convergence behavior, initialization, and parameter determination are also analyzed. Compared with vector-based methods, 2DLMNN performs better for matrix data classification. Promising experimental results on several data sets are provided to demonstrate the effectiveness of our method.