Generative Maximum Entropy Learning for Multiclass Classification

Maximum entropy approach to classification is very well studied in applied statistics and machine learning and almost all the methods that exists in literature are discriminative in nature. In this paper, we introduce a maximum entropy classification method with feature selection for large dimensional data such as text datasets that is generative in nature. To tackle the curse of dimensionality of large data sets, we employ conditional independence assumption (Naive Bayes) and we perform feature selection simultaneously, by enforcing a `maximum discrimination' between estimated class conditional densities. For two class problems, in the proposed method, we use Jeffreys (J) divergence to discriminate the class conditional densities. To extend our method to the multi-class case, we propose a completely new approach by considering a multi-distribution divergence: we replace Jeffreys divergence by Jensen-Shannon (JS) divergence to discriminate conditional densities of multiple classes. In order to reduce computational complexity, we employ a modified Jensen-Shannon divergence (JS(GM)), based on AM-GM inequality. We show that the resulting divergence is a natural generalization of Jeffreys divergence to a multiple distributions case. As far as the theoretical justifications are concerned we show that when one intends to select the best features in a generative maximum entropy approach, maximum discrimination using J-divergence emerges naturally in binary classification. Performance and comparative study of the proposed algorithms have been demonstrated on large dimensional text and gene expression datasets that show our methods scale up very well with large dimensional datasets.

From Selective-ID to Full-ID IBS without Random Oracles

Since its induction, the selective-identity (sID) model for identity-based cryptosystems and its relationship with various other notions of security has been extensively studied. As a result, it is a general consensus that the sID model is much weaker than the full-identity (ID) model. In this paper, we study the sID model for the particular case of identity-based signatures (IBS). The main focus is on the problem of constructing an ID-secure IBS given an sID-secure IBS without using random oracles-the so-called standard model-and with reasonable security degradation. We accomplish this by devising a generic construction which uses as black-box: i) a chameleon hash function and ii) a weakly-secure public-key signature. We argue that the resulting IBS is ID-secure but with a tightness gap of O(q(s)), where q(s) is the upper bound on the number of signature queries that the adversary is allowed to make. To the best of our knowledge, this is the first attempt at such a generic construction.

Voice source characterization using pitch synchronous discrete cosine transform for speaker identification

A characterization of the voice source (VS) signal by the pitch synchronous (PS) discrete cosine transform (DCT) is proposed. With the integrated linear prediction residual (ILPR) as the VS estimate, the PS DCT of the ILPR is evaluated as a feature vector for speaker identification (SID). On TIMIT and YOHO databases, using a Gaussian mixture model (GMM)-based classifier, it performs on par with existing VS-based features. On the NIST 2003 database, fusion with a GMM-based classifier using MFCC features improves the identification accuracy by 12% in absolute terms, proving that the proposed characterization has good promise as a feature for SID studies. (C) 2015 Acoustical Society of America

Modeling harvest rates and numbers from age and sex ratios: A demonstration for elephant populations

Illegal harvest rates of wildlife populations are often unknown or difficult to estimate from field data due to under-reporting or incomplete detection of carcasses. This is especially true for elephants that are killed for ivory or in conflicts with people. We describe a method to infer harvest rates from coarse field data of three population parameters, namely, adult female to male ratio, male old-adult to young-adult ratio, and proportion of adult males in the population using Jensen's (2000) 2-sex, density-dependent Leslie matrix model. The specific combination of male and female harvest rates and numbers can be determined from the history of harvest and estimate of population size. We applied this technique to two populations of elephants for which data on age structure and records of mortality were available-a forest-dwelling population of the Asian elephant (at Nagarahole, India) and an African savannah elephant population (at Samburu, Kenya) that had experienced male-biased harvest regimes over 2-3 decades. For the Nagarahole population, the recorded numbers of male and female elephants killed illegally during 1981-2000 were 64% and 88% of the values predicted by the model, respectively, implying some non-detection or incomplete reporting while for the Samburu population the recorded and modeled numbers of harvest during 1990-1999 closely matched. This technique, applicable to any animal population following logistic growth model, can be especially useful for inferring illegal harvest numbers of forest elephants in Africa and Asia. (C) 2013 Elsevier Ltd. All rights reserved.

Power Allocation in MIMO Wiretap Channel with Statistical CSI and Finite-Alphabet Input

In this paper, we consider the problem of power allocation in MIMO wiretap channel for secrecy in the presence of multiple eavesdroppers. Perfect knowledge of the destination channel state information (CSI) and only the statistical knowledge of the eavesdroppers CSI are assumed. We first consider the MIMO wiretap channel with Gaussian input. Using Jensen's inequality, we transform the secrecy rate max-min optimization problem to a single maximization problem. We use generalized singular value decomposition and transform the problem to a concave maximization problem which maximizes the sum secrecy rate of scalar wiretap channels subject to linear constraints on the transmit covariance matrix. We then consider the MIMO wiretap channel with finite-alphabet input. We show that the transmit covariance matrix obtained for the case of Gaussian input, when used in the MIMO wiretap channel with finite-alphabet input, can lead to zero secrecy rate at high transmit powers. We then propose a power allocation scheme with an additional power constraint which alleviates this secrecy rate loss problem, and gives non-zero secrecy rates at high transmit powers.

Effect of strain rate response and pin diameter on mechanically mixed layer formation and wear mechanisms in a Ti6Al4V-SS316L pair

Three mechanisms operate during wear of materials. These mechanisms include the Strain Rate Response (SRR - effect of strain rate on plastic deformation), Tribo-Chemical Reactions (TCR) and formation of Mechanically Mixed Layers (MML). The present work investigates the effect of these three in context of the formation of MML. For this wear experiments are done on a pin-on-disc machine using Ti64 as the pin and SS316L as the disc. It is seen that apart from the speed and load, which control the SRR and TCR, the diameter of the pin controls the formation of MML, especially at higher speeds.

A Preliminary Check-List of the Marine Algae of the Moss Landing Jetty: An Annotated Floristic Compilation. Annual Report, Part 7, 1973

(PDF contains 55 pages)

An early-warning system for TNE

This report for Jisc1 is based on feedback from the UK higher education (HE) sector on current (2014) transnational education (TNE) activities and future plans, including the locations of such activity. The exercise includes feedback on current and future TNE delivery modes. It is further based on feedback of a more technical nature, for example, on what the network is used for in TNE and how such IT operations are managed abroad. The resulting narrative is a synthesis of these two distinct voices from within UK higher education institutions (HEIs).

A snapshot of digitisation and OCR Developments

One of the goals of Knowledge Exchange (www.knowledge-exchange.info) is to help partners to share knowledge and expertise and facilitate the build of expert networks. In the area of digitisation, including non-textual and 3-D-digitisation, a first step is to provide a snapshot of current activities and challenges in the KE partner countries. This paper is a synthesis of the information gathered in a questionnaire that was sent to 15 infrastructure institutions, e.g. libraries and also funders, within the five partner countries of Knowledge Exchange: Denmark (DK), Finland (FIN), Germany (GER), the Netherlands (NL) and the United Kingdom (UK). The paper is based on the answers provided by 6 respondents from four countries:  DK o Danish Agency of Culture (Henrik Jarl Hansen) o State and University Library (Tonny Skovgård Jensen)  GER o German research foundation, DFG (Franziska Regner)  NL o Royal Library (Hildelies Balk) o Leiden University Library (Saskia van Bergen)  UK o Jisc (Paola Marchionni, Peter Findlay) The absence of Finnish responses may be due to Finland participating in the recent Enumerate Core Survey II that also addressed digitisation. We have included some of the outcomes of this survey to present a richer picture.

NOAA Coastal Change Analysis Program (C-CAP): Guidance for Regional Implementation

EXECUTIVE SUMMARY: The Coastal Change Analysis Programl (C-CAP) is developing a nationally standardized database on landcover and habitat change in the coastal regions of the United States. C-CAP is part of the Estuarine Habitat Program (EHP) of NOAA's Coastal Ocean Program (COP). C-CAP inventories coastal submersed habitats, wetland habitats, and adjacent uplands and monitors changes in these habitats on a one- to five-year cycle. This type of information and frequency of detection are required to improve scientific understanding of the linkages of coastal and submersed wetland habitats with adjacent uplands and with the distribution, abundance, and health of living marine resources. The monitoring cycle will vary according to the rate and magnitude of change in each geographic region. Satellite imagery (primarily Landsat Thematic Mapper), aerial photography, and field data are interpreted, classified, analyzed, and integrated with other digital data in a geographic information system (GIS). The resulting landcover change databases are disseminated in digital form for use by anyone wishing to conduct geographic analysis in the completed regions. C-CAP spatial information on coastal change will be input to EHP conceptual and predictive models to support coastal resource policy planning and analysis. CCAP products will include 1) spatially registered digital databases and images, 2) tabular summaries by state, county, and hydrologic unit, and 3) documentation. Aggregations to larger areas (representing habitats, wildlife refuges, or management districts) will be provided on a case-by-case basis. Ongoing C-CAP research will continue to explore techniques for remote determination of biomass, productivity, and functional status of wetlands and will evaluate new technologies (e.g. remote sensor systems, global positioning systems, image processing algorithms) as they become available. Selected hardcopy land-cover change maps will be produced at local (1:24,000) to regional scales (1:500,000) for distribution. Digital land-cover change data will be provided to users for the cost of reproduction. Much of the guidance contained in this document was developed through a series of professional workshops and interagency meetings that focused on a) coastal wetlands and uplands; b) coastal submersed habitat including aquatic beds; c) user needs; d) regional issues; e) classification schemes; f) change detection techniques; and g) data quality. Invited participants included technical and regional experts and representatives of key State and Federal organizations. Coastal habitat managers and researchers were given an opportunity for review and comment. This document summarizes C-CAP protocols and procedures that are to be used by scientists throughout the United States to develop consistent and reliable coastal change information for input to the C-CAP nationwide database. It also provides useful guidelines for contributors working on related projects. It is considered a working document subject to periodic review and revision.(PDF file contains 104 pages.)