A PAC-Bayesian margin bound for linear classifiers

We present a bound on the generalization error of linear classifiers in terms of a refined margin quantity on the training sample. The result is obtained in a probably approximately correct (PAC)-Bayesian framework and is based on geometrical arguments in the space of linear classifiers. The new bound constitutes an exponential improvement of the so far tightest margin bound, which was developed in the luckiness framework, and scales logarithmically in the inverse margin. Even in the case of less training examples than input dimensions sufficiently large margins lead to nontrivial bound values and-for maximum margins-to a vanishing complexity term. In contrast to previous results, however, the new bound does depend on the dimensionality of feature space. The analysis shows that the classical margin is too coarse a measure for the essential quantity that controls the generalization error: the fraction of hypothesis space consistent with the training sample. The practical relevance of the result lies in the fact that the well-known support vector machine is optimal with respect to the new bound only if the feature vectors in the training sample are all of the same length. As a consequence, we recommend to use support vector machines (SVMs) on normalized feature vectors only. Numerical simulations support this recommendation and demonstrate that the new error bound can be used for the purpose of model selection.     

50 years of the IEEE Microwave Theory and Techniques Society

The early history of the IEEE Microwave Theory and Techniques Society (IEEE MTT-S) is summarized since its founding in 1952, and all administrative committee members and presidents are listed. Some of the more recent changes resulting from growth and multinational participation are described. Publications are discussed with editors listed for this Transactions, the IEEE Microwave and Wireless Components Letters, the IEEE Microwave Newsletter, and IEEE Microwave Magazine. The chronological evolution of the IEEE MTT-S's awards is presented, including a listing of all award winners. Distinguished lecturers and microwave symposia sites and chairpersons are also discussed. Early technology trends are described     

A robust classification procedure based on mixture classifiers and nonparametric weighted feature extraction

There are many factors to consider in carrying out a hyperspectral data classification. Perhaps chief among them are class training sample size, dimensionality, and distribution separability. The intent of this study is to design a classification procedure that is robust and maximally effective, but which provides the analyst with significant assists, thus simplifying the analyst's task. The result is a quadratic mixture classifier based on Mixed-LOOC2 regularized discriminant analysis and nonparametric weighted feature extraction. This procedure has the advantage of providing improved classification accuracy compared to typical previous methods but requires minimal need to consider the factors mentioned above. Experimental results demonstrating these properties are presented.     

Characterization of residual information for SeaWinds quality control

Recent work has shown the important properties of the wind inversion residual or maximum-likelihood estimator (MLE) for quality Control (QC) of QuikSCAT Hierarchical Data Format (HDF) observations. Since March 2000, the QuikSCAT near-real-time (NRT) Binary Universal Format Representation (BUFR) product is available. As this product is used for numerical weather prediction (NWP) assimilation purposes, a QC procedure for the BUFR product is needed. We study the behavior of the MLE in order to determine whether the HDF QC procedure is appropriate for BUFR data. A comparison using real HDF and BUFR data reveals that the MLE distributions of HDF and BUFR differ and are actually poorly correlated. One important difference between BUFR and HDF is the amount of signal averaging prior to wind inversion. The averaging reduces the number of observations used in the wind retrieval for the BUFR product as compared to HDF. We show with a simple example that different MLE distributions are indeed expected due to this averaging. We also run a simulation in order to link theory and reality and better understand the behavior of the MLE. Despite the different MLE behavior in BUFR and HDF, the quality of the retrieved winds, as compared with the European Centre for Medium-Range Weather Forecasts winds, is very similar. We develop an MLE-based QC procedure for BUFR, similarly to the one in HDF, and we compare both. The skill of the QC in BUFR is again very similar to the one in HDF, showing that despite the different MLE behavior in both formats, the properties of the MLE as a QC indicator remain very similar.     

Cryptanalysis of Polly Cracker

An attack on the public key cryptosystem Polly Cracker is described, that reveals the complete secret key σ ∈ F_q ⁿ by means of n (nonadaptively) chosen "fake" ciphertexts     

General properties of antennas

In antenna design there are some fundamental relationships based on reciprocity. The equivalence of antenna pattern in transmission and reception is well known. Less well known is the time-derivative relationship going from reception to transmission. These relationships are derived here and expressed in various useful forms. Electric and magnetic dipoles are given special consideration, and the combined form constructed as a terminated TEM transmission line (the BTW antenna) is discussed for its transmission and reception properties     

Evaluation of a compound probability model with tower-mountedscatterometer data

Six months of data from the YSCAT94 experiment conducted at the CCIW WAVES research platform on Lake Ontario, Canada, are analyzed to evaluate a compound probability model. YSCAT was an ultrawideband small footprint (≈1 m) microwave scatterometer that operated at frequencies of 2-18 GHz, incidence angles from 0° to 60°, both h-pol and v-pol, and which tracked the wind using simultaneous weather measurements. The probability distribution function of the measured instantaneous backscattered amplitude (p(a)) is compared to theoretical distributions developed from-the composite model and a simple wave spectrum. Model parameters of the resulting Rayleigh/generalized lognormal distribution probability density function (pdf) (C, a₁ , and a₂) are derived directly from the data and are found to demonstrate relationships with wind speed, incidence angle, and radar frequency     

High-frequency small signal AC and noise modeling of MOSFETs for RFIC design

High-frequency (HF) AC and noise modeling of MOSFETs for radio frequency (RF) integrated circuit (IC) design is discussed. A subcircuit RF model incorporating the HF effects of parasitics is presented. This model is compared with the measured data for both y parameter and f_T characteristics. Good model accuracy is achieved against measurements for a 0.25 μm RF CMOS technology. The HF noise predictivity of the model is also examined with measured data. Furthermore, a methodology to extract the channel thermal noise of MOSFETs from HF noise measurements is presented. By using the extracted channel thermal noise, any thermal noise models can be verified directly. Several noise models including the RF model discussed in this paper have been examined, and the results show that the RF model can predict the channel thermal noise better than the other models     

Capacity-achieving sequences for the erasure channel

This paper starts a systematic study of capacity-achieving (c.a.) sequences of low-density parity-check codes for the erasure channel. We introduce a class A of analytic functions and develop a procedure to obtain degree distributions for the codes. We show various properties of this class which help us to construct new distributions from old ones. We then study certain types of capacity-achieving sequences and introduce new measures for their optimality. For instance, it turns out that the right-regular sequence is c.a. in a much stronger sense than, e.g., the Tornado sequence. This also explains why numerical optimization techniques tend to favor graphs with only one degree of check nodes.