Joint Transferable Dictionary Learning and View Adaptation for Multi-view Human Action Recognition

Multi-view human action recognition remains a challenging problem due to large view changes. In this article, we propose a transfer learning-based framework called transferable dictionary learning and view adaptation (TDVA) model for multi-view human action recognition. In the transferable dictionary learning phase, TDVA learns a set of view-specific transferable dictionaries enabling the same actions from different views to share the same sparse representations, which can transfer features of actions from different views to an intermediate domain. In the view adaptation phase, TDVA comprehensively analyzes global, local, and individual characteristics of samples, and jointly learns balanced distribution adaptation, locality preservation, and discrimination preservation, aiming at transferring sparse features of actions of different views from the intermediate domain to a common domain. In other words, TDVA progressively bridges the distribution gap among actions from various views by these two phases. Experimental results on IXMAS, ACT4 2 , and NUCLA action datasets demonstrate that TDVA outperforms state-of-the-art methods.

Multimodal semantic analysis with regularized semantic autoencoder

The real-world data is multimodal and to classify them by machine learning algorithms, features of both modalities must be transformed into common latent space. The high dimensional common space transformation of features lose their locality information and susceptible to noise. This research article has dealt with this issue of a semantic autoencoder and presents a novel algorithm with distinct mapped features with locality preservation into a commonly hidden space. We call it discriminative regularized semantic autoencoder (DRSAE). It maintains the low dimensional features in the manifold to manage the inter and intra-modality of the data. The data has multi labels, and these are transformed into an aware feature space. Conditional Principal label space transformation (CPLST) is used for it. With the two-fold proposed algorithm, we achieve a significant improvement in text retrieval form image query and image retrieval from the text query.

A Parallel N-Dimensional Space-Filling Curve Library and Its Application in Massive Point Cloud Management

Because of their locality preservation properties, Space-Filling Curves (SFC) have been widely used in massive point dataset management. However, the completeness, universality, and scalability of current SFC implementations are still not well resolved. To address this problem, a generic n-dimensional (nD) SFC library is proposed and validated in massive multiscale nD points management. The library supports two well-known types of SFCs (Morton and Hilbert) with an object-oriented design, and provides common interfaces for encoding, decoding, and nD box query. Parallel implementation permits effective exploitation of underlying multicore resources. During massive point cloud management, all xyz points are attached an additional random level of detail (LOD) value l. A unique 4D SFC key is generated from each xyzl with this library, and then only the keys are stored as flat records in an Oracle Index Organized Table (IOT). The key-only schema benefits both data compression and multiscale clustering. Experiments show that the proposed nD SFC library provides complete functions and robust scalability for massive points management. When loading 23 billion Light Detection and Ranging (LiDAR) points into an Oracle database, the parallel mode takes about 10 h and the loading speed is estimated four times faster than sequential loading. Furthermore, 4D queries using the Hilbert keys take about 1~5 s and scale well with the dataset size.

A Robust Distributed Big Data Clustering-based on Adaptive Density Partitioning using Apache Spark

Unsupervised machine learning and knowledge discovery from large-scale datasets have recently attracted a lot of research interest. The present paper proposes a distributed big data clustering approach-based on adaptive density estimation. The proposed method is developed-based on Apache Spark framework and tested on some of the prevalent datasets. In the first step of this algorithm, the input data is divided into partitions using a Bayesian type of Locality Sensitive Hashing (LSH). Partitioning makes the processing fully parallel and much simpler by avoiding unneeded calculations. Each of the proposed algorithm steps is completely independent of the others and no serial bottleneck exists all over the clustering procedure. Locality preservation also filters out the outliers and enhances the robustness of the proposed approach. Density is defined on the basis of Ordered Weighted Averaging (OWA) distance which makes clusters more homogenous. According to the density of each node, the local density peaks will be detected adaptively. By merging the local peaks, final cluster centers will be obtained and other data points will be a member of the cluster with the nearest center. The proposed method has been implemented and compared with similar recently published researches. Cluster validity indexes achieved from the proposed method shows its superiorities in precision and noise robustness in comparison with recent researches. Comparison with similar approaches also shows superiorities of the proposed method in scalability, high performance, and low computation cost. The proposed method is a general clustering approach and it has been used in gene expression clustering as a sample of its application.

Label Embedding Based on Multi-Scale Locality Preservation

Label Distribution Learning (LDL) fits the situations well that focus on the overall distribution of the whole series of labels. The numerical labels of LDL satisfy the integrity probability constraint. Due to LDL's special label domain, existing label embedding algorithms that focus on embedding of binary labels are thus unfit for LDL. This paper proposes a specially designed approach MSLP that achieves label embedding for LDL by Multi-Scale Locality Preserving (MSLP). Specifically, MSLP takes the locality information of data in both the label space and the feature space into account with different locality granularity. By assuming an explicit mapping from the features to the embedded labels, MSLP does not need an additional learning process after completing embedding. Besides, MSLP is insensitive to the existing of data points violating the smoothness assumption, which is usually caused by noises. Experimental results demonstrate the effectiveness of MSLP in preserving the locality structure of label distributions in the embedding space and show its superiority over the state-of-the-art baseline methods.