Machine Learning for Predictive Maintenance: A Multiple Classifiers Approach

In this paper a multiple classifier machine learning methodology for Predictive Maintenance (PdM) is presented. PdM is a prominent strategy for dealing with maintenance issues given the increasing need to minimize downtime and associated costs. One of the challenges with PdM is generating so called ’health factors’ or quantitative indicators of the status of a system associated with a given maintenance issue, and determining their relationship to operating costs and failure risk. The proposed PdM methodology allows dynamical decision rules to be adopted for maintenance management and can be used with high-dimensional and censored data problems. This is achieved by training multiple classification modules with different prediction horizons to provide different performance trade-offs in terms of frequency of unexpected breaks and unexploited lifetime and then employing this information in an operating cost based maintenance decision system to minimise expected costs. The effectiveness of the methodology is demonstrated using a simulated example and a benchmark semiconductor manufacturing maintenance problem.

Collectively enhanced optomechanical coupling in periodic arrays of scatterers

We investigate the optomechanical properties of a periodic array of identical scatterers placed inside an optical cavity and extend our previous results. We show that operating at the points where the array is transmissive results in linear optomechanical coupling strengths between the cavity field and collective motional modes of the array that may be several orders of magnitude larger than is possible with an equivalent reflective ensemble. We describe and interpret these effects in detail and investigate the nature of the scaling laws of the coupling strengths for the different transmissive points in various regimes. © 2013 American Physical Society.

Visible-Wavelength Survey of Jupiter-Family Cometary Nuclei as Part of SEPPCoN

We present observations of a statistically-significant number of Jupiter-family cometary nuclei as part of SEPPCoN (Survey of the Ensemble Physical Properties of Cometary Nuclei). We present preliminary results on distributions of albedos and shapes.

Visible-wavelength Observations of Jupiter-family Comet Nuclei as Part of Seppcon.

We present the latest analysis and results from SEPPCoN (Survey of Ensemble Physical Properties of Cometary Nuclei). This on-going survey involves studying 100 JFCs - about 25% of the known population - at both mid-infrared and visible wave-lengths to constrain the distributions of sizes, shapes, spins, and albedos of this population. Having earlier reported results from measuring thermal emissions of our sample nuclei [1,2,3,4], we report here progress on the visible-wavelength observations that we have obtained at many ground-based facilities in Chile, Spain, and the United States. To date we have attempted observations of 91% of our sample of 100 JFCs, and at least 64 of those were successfully detected. In most cases the comets were at heliocentric distances between 3.0 and 6.5 AU so as to decrease the odds of a comet having a coma. Of the 64 detected comets, 48 were apparently bare, having no extended emission. Our datasets are further augmented by archival data and photometry from the NEAT program [5]. An important goal of SEPPCoN is to accumulate a large comprehensive set of high quality physical data on cometary nuclei in order to make accurate statistical comparisons with other minor-body populations such as Trojans, Centaurs, and Kuiper-belt objects. Information on the size, shape, spin-rate, albedo and color distributions is critical for understanding their origins and evolutionary processes affecting them.

High accuracy android malware detection using ensemble learning

With over 50 billion downloads and more than 1.3 million apps in Google’s official market, Android has continued to gain popularity amongst smartphone users worldwide. At the same time there has been a rise in malware targeting the platform, with more recent strains employing highly sophisticated detection avoidance techniques. As traditional signature based methods become less potent in detecting unknown malware, alternatives are needed for timely zero-day discovery. Thus this paper proposes an approach that utilizes ensemble learning for Android malware detection. It combines advantages of static analysis with the efficiency and performance of ensemble machine learning to improve Android malware detection accuracy. The machine learning models are built using a large repository of malware samples and benign apps from a leading antivirus vendor. Experimental results and analysis presented shows that the proposed method which uses a large feature space to leverage the power of ensemble learning is capable of 97.3 % to 99% detection accuracy with very low false positive rates.

Integrative analysis of chemo-transcriptomic profiles for drug-feature specific gene expression signatures

One of the major challenges in systems biology is to understand the complex responses of a biological system to external perturbations or internal signalling depending on its biological conditions. Genome-wide transcriptomic profiling of cellular systems under various chemical perturbations allows the manifestation of certain features of the chemicals through their transcriptomic expression profiles. The insights obtained may help to establish the connections between human diseases, associated genes and therapeutic drugs. The main objective of this study was to systematically analyse cellular gene expression data under various drug treatments to elucidate drug-feature specific transcriptomic signatures. We first extracted drug-related information (drug features) from the collected textual description of DrugBank entries using text-mining techniques. A novel statistical method employing orthogonal least square learning was proposed to obtain drug-feature-specific signatures by integrating gene expression with DrugBank data. To obtain robust signatures from noisy input datasets, a stringent ensemble approach was applied with the combination of three techniques: resampling, leave-one-out cross validation, and aggregation. The validation experiments showed that the proposed method has the capacity of extracting biologically meaningful drug-feature-specific gene expression signatures. It was also shown that most of signature genes are connected with common hub genes by regulatory network analysis. The common hub genes were further shown to be related to general drug metabolism by Gene Ontology analysis. Each set of genes has relatively few interactions with other sets, indicating the modular nature of each signature and its drug-feature-specificity. Based on Gene Ontology analysis, we also found that each set of drug feature (DF)-specific genes were indeed enriched in biological processes related to the drug feature. The results of these experiments demonstrated the pot- ntial of the method for predicting certain features of new drugs using their transcriptomic profiles, providing a useful methodological framework and a valuable resource for drug development and characterization.

Predictive molecular classifiers in colorectal cancer

The introduction of predictive molecular markers has radically enhanced the identification of which patients may benefit from a given treatment. Despite recent controversies, KRAS mutation is currently the most recognized molecular predictive marker in colorectal cancer (CRC), predicting efficacy of anti-epidermal growth factor receptor (anti-EGFR) antibodies. However, other relevant markers have been reported and claimed to identify patients that will benefit from anti-EGFR therapies. This group of markers includes BRAF mutations, PI3KCA mutations, and loss of PTEN expression. Similarly, molecular markers for cytotoxic agents' efficacy also may predict outcome in patients with CRC. This review aims to summarize the most important predictive molecular classifiers in patients with CRC and further discuss any inconsistent or conflicting findings for these molecular classifiers.

Cascaded multimodal biometric recognition framework

A practically viable multi-biometric recognition system should not only be stable, robust and accurate but should also adhere to real-time processing speed and memory constraints. This study proposes a cascaded classifier-based framework for use in biometric recognition systems. The proposed framework utilises a set of weak classifiers to reduce the enrolled users' dataset to a small list of candidate users. This list is then used by a strong classifier set as the final stage of the cascade to formulate the decision. At each stage, the candidate list is generated by a Mahalanobis distance-based match score quality measure. One of the key features of the authors framework is that each classifier in the ensemble can be designed to use a different modality thus providing the advantages of a truly multimodal biometric recognition system. In addition, it is one of the first truly multimodal cascaded classifier-based approaches for biometric recognition. The performance of the proposed system is evaluated both for single and multimodalities to demonstrate the effectiveness of the approach.

Identification of diagnostic biomarkers for infection in premature neonates

Infection is a leading cause of neonatal morbidity and mortality worldwide. Premature neonates are particularly susceptible to infection because of physiologic immaturity, comorbidity, and extraneous medical interventions. Additionally premature infants are at higher risk of progression to sepsis or severe sepsis, adverse outcomes, and antimicrobial toxicity. Currently initial diagnosis is based upon clinical suspicion accompanied by nonspecific clinical signs and is confirmed upon positive microbiologic culture results several days after institution of empiric therapy. There exists a significant need for rapid, objective, in vitro tests for diagnosis of infection in neonates who are experiencing clinical instability. We used immunoassays multiplexed on microarrays to identify differentially expressed serum proteins in clinically infected and non-infected neonates. Immunoassay arrays were effective for measurement of more than 100 cytokines in small volumes of serum available from neonates. Our analyses revealed significant alterations in levels of eight serum proteins in infected neonates that are associated with inflammation, coagulation, and fibrinolysis. Specifically P- and E-selectins, interleukin 2 soluble receptor alpha, interleukin 18, neutrophil elastase, urokinase plasminogen activator and its cognate receptor, and C-reactive protein were observed at statistically significant increased levels. Multivariate classifiers based on combinations of serum analytes exhibited better diagnostic specificity and sensitivity than single analytes. Multiplexed immunoassays of serum cytokines may have clinical utility as an adjunct for rapid diagnosis of infection and differentiation of etiologic agent in neonates with clinical decompensation.

Statistical Tests for Joint Analysis of Performance Measures

Recently there has been an increasing interest in the development of new methods using Pareto optimality to deal with multi-objective criteria (for example, accuracy and architectural complexity). Once one has learned a model based on their devised method, the problem is then how to compare it with the state of art. In machine learning, algorithms are typically evaluated by comparing their performance on different data sets by means of statistical tests. Unfortunately, the standard tests used for this purpose are not able to jointly consider performance measures. The aim of this paper is to resolve this issue by developing statistical procedures that are able to account for multiple competing measures at the same time. In particular, we develop two tests: a frequentist procedure based on the generalized likelihood-ratio test and a Bayesian procedure based on a multinomial-Dirichlet conjugate model. We further extend them by discovering conditional independences among measures to reduce the number of parameter of such models, as usually the number of studied cases is very reduced in such comparisons. Real data from a comparison among general purpose classifiers is used to show a practical application of our tests.

Face Recognition in the Scrambled Domain via Salience-Aware Ensembles of Many Kernels

With the rapid development of internet-of-things (IoT), face scrambling has been proposed for privacy protection during IoT-targeted image/video distribution. Consequently in these IoT applications, biometric verification needs to be carried out in the scrambled domain, presenting significant challenges in face recognition. Since face models become chaotic signals after scrambling/encryption, a typical solution is to utilize traditional data-driven face recognition algorithms. While chaotic pattern recognition is still a challenging task, in this paper we propose a new ensemble approach – Many-Kernel Random Discriminant Analysis (MK-RDA) to discover discriminative patterns from chaotic signals. We also incorporate a salience-aware strategy into the proposed ensemble method to handle chaotic facial patterns in the scrambled domain, where random selections of features are made on semantic components via salience modelling. In our experiments, the proposed MK-RDA was tested rigorously on three human face datasets: the ORL face dataset, the PIE face dataset and the PUBFIG wild face dataset. The experimental results successfully demonstrate that the proposed scheme can effectively handle chaotic signals and significantly improve the recognition accuracy, making our method a promising candidate for secure biometric verification in emerging IoT applications.

Monte Carlo studies of phase transitions and adsorption : application to n-6 Lennard-Jones, C60 and zeolite systems

The solid-fluid transition properties of the n - 6 Lennard-Jones system are studied by means of extensive free energy calculations. Different values of the parameter n which regulates the steepness of the short-range repulsive interaction are investigated. Furthermore, the free energies of the n < 12 systems are calculated using the n = 12 system as a reference. The method relies on a generalization of the multiple histogram method that combines independent canonical ensemble simulations performed with different Hamiltonians and computes the free energy difference between them. The phase behavior of the fullerene C60 solid is studied by performing NPT simulations using atomistic models which treat each carbon in the molecule as a separate interaction site with additional bond charges. In particular, the transition from an orientationally frozen phase at low temperatures to one where the molecules are freely rotating at higher temperatures is studied as a function of applied pressure. The adsorption of molecular hydrogen in the zeolite NaA is investigated by means of grand-canonical Monte Carlo, in a wide range of temperatures and imposed gas pressures, and results are compared with available experimental data. A potential model is used that comprises three main interactions: van der Waals, Coulomb and induced polarization by the permanent electric field in the zeolite.

Modern Music Ensemble December 4, 2015

Concert Program

Baroque Ensemble February 24, 2015

Concert Program