This paper introduces concepts and algorithms of feature selection, surveys existing selection for classification clustering, groups compares different with a categorizing framework based on search strategies, evaluation criteria, data mining tasks, reveals unattempted combinations, provides guidelines in selecting algorithms. With the framework, we continue our efforts toward-building an integrated system intelligent selection. A unifying platform is proposed as intermediate step. An...

Deep learning techniques based on neural networks have shown significant success in a wide range of AI tasks. Large-scale training datasets are one the critical factors for their success. However, when crowdsourced from individuals and contain sensitive information, model parameters may encode private information bear risks privacy leakage. The recent growing trend sharing publishing pre-trained models further aggravates such risks. To tackle this problem, we propose differentially approach...

Membership inference attacks seek to infer membership of individual training instances a model which an adversary has black-box access through machine learning-as-a-service API. In providing in-depth characterization privacy risks against learning models, this paper presents comprehensive study towards demystifying from two complimentary perspectives. First, we provide generalized formulation the development attack model. Second, characterize importance choice on vulnerability systematic...

Due to its great application value in the military and civilian fields, ship detection synthetic aperture radar (SAR) images has always attracted much attention. However, targets High-Resolution (HR) SAR show significant characteristics of multi-scale, arbitrary directions dense arrangement, posing enormous challenges detect ships quickly accurately. To address these issues above, a novel YOLO-based arbitrary-oriented detector using bi-directional feature fusion angular classification...

Stability is an important yet under-addressed issue in feature selection from high-dimensional and small sample data. In this paper, we show that stability of has a strong dependency on size. We propose novel framework for stable which first identifies consensus groups subsampling training samples, then performs by treating each group as single entity. Experiments both synthetic real-world data sets algorithm developed under effective at alleviating the problem size leads to more results...

Learning Rate (LR) is an important hyper-parameter to tune for effective training of deep neural networks (DNNs). Even the baseline a constant learning rate, it non-trivial choose good value DNN. Dynamic rates involve multi-step tuning LR values at various stages process and offer high accuracy fast convergence. However, they are much harder tune. In this paper, we present comprehensive study 13 rate functions their associated policies by examining range parameters, step update parameters....

Membership inference attacks seek to infer membership of individual training instances a model which an adversary has black-box access through machine learning-as-a-service API. In providing in-depth characterization privacy risks against learning models, this paper presents comprehensive study towards demystifying from two complimentary perspectives. First, we provide generalized formulation the development attack model. Second, characterize importance choice on vulnerability systematic...

Captioning the images with proper descriptions automatically has become an interesting and challenging problem. In this paper, we present one joint model AICRL, which is able to conduct automatic image captioning based on ResNet50 LSTM soft attention. AICRL consists of encoder decoder. The adopts convolutional neural network, creates extensive representation given by embedding it into a fixed length vector. decoder designed LSTM, recurrent network attention mechanism, selectively focus over...

Recently, integrating vision and language for in-depth video understanding e.g., captioning question answering, has become a promising direction artificial intelligence. However, due to the complexity of information, it is challenging extract feature that can well represent multiple levels concepts i.e., objects, actions events. Meanwhile, content completeness syntactic consistency play an important role in high-quality language-related understanding. Motivated by these, we propose novel...

In this paper we describe an extension of the information theoretical FCBF (Fast Correlation Based Feature Selection) algorithm. The extension, called FCBF#, enables to select any given size feature subset and it selects features in a different order than FCBF. We find out that extended algorithm results more accurate classifiers.

We propose a novel object detection framework for partially-occluded small instances, such as pedestrians in low resolution surveillance video, cells under microscope, flocks of animals (e.g. birds, fishes), or even tiny insects like honeybees and flies. These scenarios are very challenging traditional detectors, which typically trained on individual instances. In our approach, we first estimate the density map input image, then divide it into local regions. For each region, sliding window...

Hardware Malware Detectors (HMDs) have recently been proposed as a defense against the proliferation of malware. These detectors use low-level features, that can be collected by hardware performance monitoring units on modern CPUs to detect malware computational anomaly. Several aspects detector construction explored, leading with high accuracy. In this paper, we explore question how well evasive avoid detection HMDs. We show existing HMDs effectively reverse-engineered and subsequently...

Synthetic aperture radar (SAR) is an active earth observation system with a certain surface penetration capability and can be employed to observations all-day all-weather. Ship detection using SAR of great significance maritime safety port management. With the wide application in-depth learning in ordinary images good results, increasing number algorithms began entering field remote sensing images. image has characteristics small targets, high noise, sparse targets. Two-stage methods, such...

Highlights High resolution images are found to play an important role in observing maize growth and detecting tassels. Our super-resolution model (IESRGAN) produces higher-quality compared other SR models. IESRGAN largely improves the detection accuracy of tassels, increasing by 8.77%~25.71% different tests. Abstract. The Unmanned Aerial Vehicle (UAV) technology provides essential technical support for (Zea mays) cultivation smart agriculture. To address challenge low UAV due factors such as...

In 2017, Nikiforov defined the Aα-matrix of graph G as Aα(G)=αD(G)+(1−α)A(G),0≤α≤1, which merges diagonal degree matrix D(G) and adjacency A(G). this paper, we characterize graphs attain maximum Aα-index among triangle-free non-bipartite order n for 1/2<α<1.

High-dimensional data poses a severe challenge for mining. Feature selection is frequently used technique in pre-processing high-dimensional successful Traditionally, feature focused on removing irrelevant features. However, data, redundant features equally critical. In this paper, we provide study of redundancy and propose novel correlation-based approach to within the filter model. The extensive empirical using real-world shows that proposed efficient effective

Besides high accuracy, stability of feature selection has recently attracted strong interest in knowledge discovery from high-dimensional data. In this study, we present a theoretical framework about the relationship between and accuracy based on formal bias-variance decomposition error. The also suggests variance reduction approach for improving algorithms. Furthermore, propose an empirical framework, margin instance weighting, which weights training instances according to their influence...

The burgeoning success of deep learning has raised the security and privacy concerns as more tasks are accompanied with sensitive data. Adversarial attacks in have emerged one dominating threat to a range mission-critical systems applications. This paper takes holistic principled approach perform statistical characterization adversarial examples learning. We provide general formulation elaborate on basic principle for attack algorithm design. introduce easy hard categorization analyze...

Hardware Malware Detectors (HMDs) have recently been proposed to make systems more malware-resistant. HMDs use hardware features detect malware as a computational anomaly. Several aspects of the detector construction explored, leading detectors with high accuracy. In this article, we explore whether developers can modify avoid detection. We show that existing be effectively reverse-engineered and subsequently evaded. Next, retraining using evasive would help is limited. To address these...