We present a small sphere and large margin approach for novelty detection problems, where the majority of training data are normal examples. In addition, also contain number abnormal examples or outliers. The basic idea is to construct hypersphere that contains most examples, such volume this as possible, while at same time between surface outlier possible. This can result in closed tight boundary around data. To build sphere, we only need solve convex optimization problem be efficiently...

Principal component analysis (PCA) has been extensively applied in data mining, pattern recognition and information retrieval for unsupervised dimensionality reduction. When labels of are available, e.g., a classification or regression task, PCA is however not able to use this information. The problem more interesting if only part the input labeled, i.e., semi-supervised setting. In paper we propose supervised model called SPPCA S2PPCA, both which extensions probabilistic model. proposed...

Current Image Captioning (IC) methods predict textual words sequentially based on the input visual information from feature extractor and partially generated sentence information. However, for most cases, may dominate target word prediction due to insufficiency of information, making descriptions irrelevant content given image. In this paper, we propose a Dual Information Flow Network (DIFNet <sup xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Most existing Human-Object Interaction (HOI) Detection methods rely heavily on full annotations with predefined HOI categories, which is limited in diversity and costly to scale further. We aim at advancing zero-shot detection detect both seen unseen HOIs simultaneously. The fundamental challenges are discover potential human-object pairs identify novel categories. To overcome the above challenges, we propose a End-to-end (EoID) framework via vision-language knowledge distillation. first...

The incorporation of waterborne polyurethane (WPU) into bacterial cellulose (BC) fibers significantly improved the tensile strength resulting WPU/BC composite film, achieving an enhancement 19.4 times. formation hydrogen bonds between WPU and BC effectively eliminates cavities within matrix, significant plasticization toughening. Compared with pure film (WPU/BC-0), elastic modulus WPU/BC-5 is reduced by 97.5%, surface hardness decreased 96.9%. When integrated a flexible EGaIn electrode,...

This paper presents an approach to build Sparse Large Margin Classifiers (SLMC) by adding one more constraint the standard Support Vector Machine (SVM) training problem. The added explicitly controls sparseness of classifier and is provided solve formulated When considering dual this it can be seen that building SLMC equivalent constructing SVM with a modified kernel function. Further analysis function indicates proposed essentially finds discriminating subspace spanned small number vectors,...

This paper presents a Local Learning Projection (LLP) approach for linear dimensionality reduction. We first point out that the well known Principal Component Analysis (PCA) essentially seeks projection has minimal global estimation error. Then we propose reduction algorithm leads to with local error, and elucidate its advantages classification tasks. also indicate LLP keeps information in sense value of each can be estimated based on neighbors their values. Experimental results are provided...

Web search logs contain extremely sensitive data, as evidenced by the recent AOL incident. However, storing and analyzing can be very useful for many purposes (i.e. investigating human behavior). Thus, an important research question is how to privately sanitize logs. Several log anonymization techniques have been proposed with concrete privacy models. in all of these solutions, output utility only evaluated rather than being maximized any fashion. Indeed, effective anonymization, it...

This work is oriented toward the task of open-set Human Object Interaction (HOI) detection. The challenge lies in identifying completely new, out-of-domain relationships, as opposed to in-domain ones which have seen improvements zero-shot HOI To address this challenge, we introduce a simple Disentangled Detection (DHD) model for detecting novel relationships by integrating an object detector with Visual Language Model (VLM). We utilize disentangled image-text contrastive learning metric...

Discounted cumulative gain (DCG) is widely used for evaluating ranking functions. It therefore natural to learn a function that directly optimizes DCG. However, DCG non-smooth, rendering gradient-based optimization algorithms inapplicable. To remedy this, smoothed versions of have been proposed but with only partial success. In this paper, we first present analysis shows it ineffective using the gradient drive algorithm. We then propose novel approach, SHF-SDCG, smoothing by hinge functions...

The knowledge graph with relational abundant information has been widely used as the basic data support for retrieval platforms. Image and text descriptions added to enrich node information, which accounts advantage of multi-modal graph. In field cross-modal platforms, graphs can help improve accuracy efficiency because provided by graphs. representation learning method is significant application This paper proposes a distributed collaborative vector platform (DCRL-KG) using multimodal...

Kernel methods have been applied successfully in many data mining tasks. Subspace kernel learning was recently proposed to discover an effective low-dimensional subspace of a feature space for improved classification. In this paper, we propose construct using the Hilbert-Schmidt Independence Criterion (HSIC). We show that optimal can be obtained efficiently by solving eigenvalue problem. One limitation existing formulations is and classification are independent may not optimally adapted To...

In this paper, a geometrical approach for building neural networks is proposed. With the proposed approach, it very easy to construct an efficient classifier solve handwritten Chinese character recognition problem, as well other pattern problems of large scale. Experiments are conducted evaluate performance and results obtained promising.

Graph node embedding aims at learning a vector representation for all nodes given graph. It is central problem in many machine tasks (e.g., classification, recommendation, community detection). The key graph lies how to define the dependence neighbors. Existing approaches specify (either explicitly or implicitly) certain dependencies on neighbors, which may lead loss of subtle but important structural information within and other among This intrigues us ask question: can we design model give...

Recently embedding-based retrieval or dense have shown state of the art results, compared with traditional sparse bag-of-words based approaches. This paper introduces a model-agnostic doc-level embedding framework through large language model (LLM) augmentation. In addition, it also improves some important components in training process, such as negative sampling, loss function, etc. By implementing this LLM-augmented framework, we been able to significantly improve effectiveness widely-used...