Catálogo de publicaciones - libros

HTTP request exploitations take substantial portion of network-based attacks. This paper presents a novel anomaly detection framework, which uses data mining technologies to build four independent detection models. In the training phase, these models mine specialty of every web program using web server log files as data source, and in the detection phase, each model takes the HTTP requests upon detection as input and calculates at least one anomalous probability as output. All the four models totally generate eight anomalous probabilities, which are weighted and summed up to produce a final probability, and this probability is used to decide whether the request is malicious or not. Experiments prove that our detection framework achieves close to perfect detection rate under very few false positives.

This paper introduces some widely used techniques related to nude image detection. By analyzing the merits and drawbacks of these techniques, a new nude image detection method is proposed. The proposed approach consists of two parts: the content-based approach, which aims to detect the nude image by analyzing whether it contains large mass of skin region, and the image-based approach, which extracts the color and spatial information of the image using the color histogram vector and color coherence vector, and makes classification based on the CHV and CCVs of the training samples. From the experimental results, our algorithm can achieve a classification accuracy of 85% with less than 10% false detection rate.

For natural and smooth movement of small scale fish robots, collision detection and direction changes are important. Typical obstacles are walls, rocks, water plants and other nearby robots for a group of small scale fish robots and submersibles that have been constructed in our lab. Two of 2-axes acceleration sensors are employed to measure the three components of collision angles, collision magnitudes, and the angles of robot propulsion. These data are integrated using fuzzy logic to calculate the amount of propulsion direction changes. Because caudal fin provides the main propulsion for a fish robot, there is a periodic swinging noise at the head of a robot. This noise provides a random acceleration effect on the measured acceleration data at the collision instant. We propose an algorithm based on fuzzy logic which shows that the MEMS-type accelerometers are very effective to provide information for direction changes.

In this paper, we present a new real-time visual servoing unit for laparoscopic surgery. This unit can automatically control a laparoscope manipulator through visual tracking of the laparoscopic surgical tool. For the tracking, we present a two-stage adaptive CONDENSATION (conditional density propagation) algorithm to detect the accurate position of the surgical tool tip from a surgical image sequence in real-time. This algorithm can be adaptable to abrupt changes of illumination. The experimental results show that the proposed visual tracking algorithm is highly robust.

Determining the causal structure of a domain is frequently a key task in the area of Data Mining and Knowledge Discovery. This paper introduces ensemble learning into linear causal model discovery, then examines several algorithms based on different ensemble strategies including Bagging, Adaboost and GASEN. Experimental results show that (1) Ensemble discovery algorithm can achieve an improved result compared with individual causal discovery algorithm in terms of accuracy; (2) Among all examined ensemble discovery algorithms, BWV algorithm which uses a simple Bagging strategy works excellently compared to other more sophisticated ensemble strategies; (3) Ensemble method can also improve the stability of parameter estimation. In addition, Ensemble discovery algorithm is amenable to parallel and distributed processing, which is important for data mining in large data sets.

The major problems of currently used approaches for extracting symbolic rules from trained neural networks are analyzed. The lack of efficient heuristic information is the fundamental reason that causes the low effectiveness of currently used approaches. In this paper, a new measurement method of the classification power of attributes on the basis of differential information of the trained neural networks is proposed, which is suitable for both continuous attributes and discrete attributes. Based on this new measurement method, a new approach for rule extraction from trained neural networks and classification problems with continuous attributes is proposed. The performance of the new approach is demonstrated by several computing cases. The experimental results prove that the approach proposed can improve the validity of the extracted rules remarkably compared with other rule extracting approaches, especially for the complicated classification problems.

Schema matching, the problem of finding semantic correspondences between elements of source and warehouse schemas, plays a key role in data warehousing. Currently, schema mapping is largely determined manually by domain experts, thus a labor-intensive process. In this paper, we propose a structural matching algorithm based on semantic similarity propagation. Experimental results on several real-world domains are presented, and show that the algorithm discovers semantic mappings with a high degree of accuracy.

Frequent itemsets play an essential role in many data mining tasks that try to find interesting patterns from databases. A new algorithm based on the lattice theory and bitmap index for mining frequent itemsets is proposed in this paper. Firstly, the algorithm converts the origin transaction database to an itemsets-lattice (which is a directed graph) in the preprocessing, where each itemset vertex has a label to represent its support. So we can change the complicated task of mining frequent itessets in the database to a simpler one of searching vertexes in the lattice, which can speeds up greatly the mining process. Secondly, Support counting in the association rules mining requires a great I/O and computing cost. A bitmap index technique to speed up the counting process is employed in this paper. Saving the intact bitmap usually has a big space requirement. Each bit vector is partitioned into some blocks, and hence every bit block is encoded as a shorter symbol. Therefore the original bitmap is impacted efficiently. At the end experimental and analytical results are presented.

In this paper, a support vector machine (SVM) multi-user receiver based on competition learning (CL) strategy is proposed. The new algorithm adopts a heuristic approach to extend standard SVM algorithm for multiuser classification problem, and also a clustering analysis is applied to reduce the total amount of computation. In implementation of multi-user receiver, an asymptotical iterative algorithm is used to guide the learning of the input sample pattern. The digital result shows that the new multi-user detector scheme has a relatively good performance comparing with the conventional MMSE detector especially under the heavy interference environment.

In this note, a robust adaptive tracking control problem is discussed for a class of affine nonlinear systems in the strict-feedback form with unknown nonlinearities. A unified and systematic procedure is developed to derive a novel robust adaptive fuzzy controller. Compared with most results reported in the literature, the proposed control algorithm has several advantages: 1) the controller singularity problem is avoided perfectly; 2) the online computation burden is kept to minimum; 3) exponential tracking to the reference trajectory up to an ultimately bounded error is achieved; 4) the controllers are particularly suitable for parallel processing and hardware implementation.