To maintain competitive advantages in today's global market, US companies are challenging higher education institutions to bridge competency gaps between industry workforce needs and what is provided by engineering education programs. To meet this challenge, the Mechanical Engineering program at the University of Cincinnati proposes the EXPLORES (Experiential and Problem-based Learning within Opportunities for Real-world Engineering Settings) model where students from freshman year onwards are exposed to real-world industrial problems that have a direct bearing on fundamental engineering concepts. The EXPLORES model is implemented in a learner-centered, knowledge-centered, assessment-centered, and community-centered student learning environment. Although inclusion of real-world industrial problems into engineering curriculum is not a new invention, the EXPLORES model is unique in the following aspects: (1) the problems span multiple courses so students need to have the ability to synthesize knowledge acquired from different courses in order to develop complete solutions; (2) industry partners are an integral part of the education process by providing problem design, periodic feedback, and expert solution; and (3) an on-line virtual company framework is used to facilitate information sharing and motivate student learning.

Dimensionality reduction is an important step in knowledge discovery in databases. Intrinsic dimension indicates the number of variables necessary to describe a data set. Two methods, box-counting dimension and correlation dimension, are commonly used for intrinsic dimension estimation. However, the robustness of these two methods has not been rigorously studied. This paper demonstrates that correlation dimension is more robust with respect to data sample size. In addition, instead of using a user selected distance d, we propose a new approach to capture all log-log pairs of a data set to more precisely estimate the correlation dimension. Systematic experiments are conducted to study factors that influence the computation of correlation dimension, including sample size, the number of redundant variables, and the portion of log-log plot used for calculation. Experiments on real-world data sets confirm the effectiveness of intrinsic dimension estimation with our improved method. Furthermore, a new supervised dimensionality reduction method based on intrinsic dimension estimation was introduced and validated.

Feature selection is a critical preprocessing step in machine learning. It contributes to cost-effective model building and improvement of model prediction performance. Generally, a feature selection algorithm requires a dependency measure and a search strategy. Extant dependency measures are mostl y based on pair-wise correlation analysis, which cannot detect feature interaction. To overcome this problem, we developed a unified dependency criterion called inference correlation. The inference correlation between a set of predictor variables and a response variable can be efficiently calculated. The variables could be discrete, continuous, or mixed. Therefore, inference correlation can be applied to select features for both classification and regression problems. A feature selection algorithm using sequential floating forward search based on inference correlation is presented. Experiments of the algorithm on synthetic datasets and real-world problems confirm the effectiveness of the feature selection approach when compared to extant feature selection methods.

One-Time Password (OTP) is always used as the strongest authentication scheme among all password-based solutions. Currently, consumer devices such as smart card have implemented OTP based two-factor authentications for secure access controls. Such solutions are economically sound without support of timestamp mechanisms. Therefore, synchronization of internal parameters in OTP models, such asmoving factor or counter, between the client and server is the key challenge. Recently, a novel phenomenon shows that two interacting neural networks, called Tree Parity Machines (TPM), with common inputs can finally synchronize their weight vectors through finite steps of output-based mutual learning. The improved secure TPM can well be utilized to synchronize parameters for OTP schemes. In this paper, TPM mutual learning scheme is introduced, then two TPM-based novel OTP solutions are proposed. One is a full implementation model including initialization and rekeying, while the other is light-weight and efficient suitable for resource-constrained embedded environment. Security and performance on the proposed protocols are at final discussed.

This paper presents a robust and efficient method to discover knowledge for classification problems through data summarization. It discretizes continuous features and then summarizes the data using a contingency table. Inconsistency rate for different subsets of features can then be easily calculated from the contingency table. Sequential search is then used to find the best feature subset. After the number of features is reduced to a certain extent, easy-to-understand knowledge can be intuitively derived from data summary. Another desirable feature of the proposed method is its capability to learn incrementally; namely, knowledge can be updated quickly whenever new data are obtained. Moreover, the proposed method is capable of handling missing values when used for prediction. The method is applied on two benchmark data sets showing its effectiveness on selecting discriminative features. The practical usefulness of this method in manufacturing is demonstrated through an application on welding fault diagnosis.

Knowledge acquisition is the process of collecting domain knowledge, documenting the knowledge, and transforming it into a computerized representation. Due to the difficulties involved in eliciting knowledge from human experts, knowledge acquisition was identified as a bottleneck in the development of knowledge-based system. Over the past decades, a number of automatic knowledge acquisition techniques have been developed. However, the performance of these techniques suffers from the so called curse of dimensionality, i.e., difficulties arise when many irrelevant (or redundant) parameters exist. This paper presents a heuristic approach based on statistics and greedy search for dimensionality reduction to facilitate automatic knowledge acquisition. The approach deals with classification problems. Specifically, Chi-square statistics are used to rank the importance of individual parameters. Then, a backward search procedure is employed to eliminate parameters (less important parameters first) that do not contribute to class separability. The algorithm is very efficient and was found to be effective when applied to a variety of problems with different characteristics.

Rule-based systems have been the forerunner among contemporary techniques for systems engineering. A doyen of the intelligent systems, they have undergone philosophical and methodological transformations and have proven to be very successful in applications such as robotics, geometric information systems, medical instrumentation, image processing, natural language processing and interactive systems. However, due to the knowledge acquisition bottleneck, the development of rule-based systems is time-consuming, carries high costs, and has high risks. In this paper, we propose a methodology that aims at overcoming the bottleneck by automatically generating intelligible and robust rules based on empirical data. With statistics as its foundation and bolstered by heuristic search, the, proposed methodology has been successful in cutting through the maze of complex systems in generating more than a plausible solution. This was confirmed by real-world engineering applications. Techniques that can automatically generate and refine fuzzy rules do exist in the literature. A popular method is the adaptive neural-fuzzy inference system (ANFIS) that refines fuzzy rules obtained using clustering techniques. However, ANFIS utilizes the Sugeno type fuzzy rules. The rules generated are not easy to comprehend, since the consequents are a crisp functions rather than fuzzy linguistic terms. Moreover, the approach cannot deal with discrete outputs. Our methodology generates the more popular, versatile, and easily understood Mamdani type fuzzy rules. Neural network-type learning algorithms are then used to optimize parameters of fuzzy membership functions and relieve the defuzzification impediment.

The advent of artificial neural networks has stirred the imagination of many in the field of knowledge acquisition. There is an expectation that neural networks will play an important role in automating knowledge acquisition and encoding, however, the problem solving knowledge of a neural network is represented at a subsymbolic level and hence is very difficult for a human user to comprehend. One way to provide an understanding of the behavior of neural networks is to extract their problem solving knowledge in terms of rules that can be provided to users. Several papers which propose extracting rules from feedforward neural networks can be found in the literature, however, these approaches can only deal with networks with binary inputs. Furthermore, certain approaches lack theoretical support and their usefulness and effectiveness are debatable. Upon carefully analyzing these approaches, we propose a method to extract fuzzy rules from networks with continuous-valued inputs. The method was tested using a real-life problem (decision-making by pilots involving combat situations) and found to be effective.