Structured Learning for Non-Smooth Ranking losses

Learning to rank from relevance judgment is an active research area. Itemwise score regression, pairwise preference satisfaction, and listwise structured learning are the major techniques in use. Listwise structured learning has been applied recently to optimize important non-decomposable ranking criteria like AUC (area under ROC curve) and MAP(mean average precision). We propose new, almost-lineartime algorithms to optimize for two other criteria widely used to evaluate search systems: MRR (mean reciprocal rank) and NDCG (normalized discounted cumulative gain)in the max-margin structured learning framework. We also demonstrate that, for different ranking criteria, one may need to use different feature maps. Search applications should not be optimized in favor of a single criterion, because they need to cater to a variety of queries. E.g., MRR is best for navigational queries, while NDCG is best for informational queries. A key contribution of this paper is to fold multiple ranking loss functions into a multi-criteria max-margin optimization.The result is a single, robust ranking model that is close to the best accuracy of learners trained on individual criteria. In fact, experiments over the popular LETOR and TREC data sets show that, contrary to conventional wisdom, a test criterion is often not best served by training with the same individual criterion.

Evaluation of mechanical losses in a linear motor pressure wave generator

A moving magnet linear motor compressor or pressure wave generator (PWG) of 2 cc swept volume with dual opposed piston configuration has been developed to operate miniature pulse tube coolers. Prelimnary experiments yielded only a no-load cold end temperature of 180 K. Auxiliary tests and the interpretation of detailed modeling of a PWG suggest that much of the PV power has been lost in the form of blow-by at piston seals due to large and non-optimum clearance seal gap between piston and cylinder. The results of experimental parameters simulated using Sage provide the optimum seal gap value for maximizing the delivered PV power.

Classification calibration dimension for general multiclass losses

We study consistency properties of surrogate loss functions for general multiclass classification problems, defined by a general loss matrix. We extend the notion of classification calibration, which has been studied for binary and multiclass 0-1 classification problems (and for certain other specific learning problems), to the general multiclass setting, and derive necessary and sufficient conditions for a surrogate loss to be classification calibrated with respect to a loss matrix in this setting. We then introduce the notion of \emph{classification calibration dimension} of a multiclass loss matrix, which measures the smallest `size' of a prediction space for which it is possible to design a convex surrogate that is classification calibrated with respect to the loss matrix. We derive both upper and lower bounds on this quantity, and use these results to analyze various loss matrices. In particular, as one application, we provide a different route from the recent result of Duchi et al.\ (2010) for analyzing the difficulty of designing `low-dimensional' convex surrogates that are consistent with respect to pairwise subset ranking losses. We anticipate the classification calibration dimension may prove to be a useful tool in the study and design of surrogate losses for general multiclass learning problems.

Extension of TSVM to multi-class and hierarchical text classification problems With general losses

Transductive SVM (TSVM) is a well known semi-supervised large margin learning method for binary text classification. In this paper we extend this method to multi-class and hierarchical classification problems. We point out that the determination of labels of unlabeled examples with fixed classifier weights is a linear programming problem. We devise an efficient technique for solving it. The method is applicable to general loss functions. We demonstrate the value of the new method using large margin loss on a number of multi-class and hierarchical classification datasets. For maxent loss we show empirically that our method is better than expectation regularization/constraint and posterior regularization methods, and competitive with the version of entropy regularization method which uses label constraints.

Surrogate Regret Bounds for Bipartite Ranking via Strongly Proper Losses

The problem of bipartite ranking, where instances are labeled positive or negative and the goal is to learn a scoring function that minimizes the probability of mis-ranking a pair of positive and negative instances (or equivalently, that maximizes the area under the ROC curve), has been widely studied in recent years. A dominant theoretical and algorithmic framework for the problem has been to reduce bipartite ranking to pairwise classification; in particular, it is well known that the bipartite ranking regret can be formulated as a pairwise classification regret, which in turn can be upper bounded using usual regret bounds for classification problems. Recently, Kotlowski et al. (2011) showed regret bounds for bipartite ranking in terms of the regret associated with balanced versions of the standard (non-pairwise) logistic and exponential losses. In this paper, we show that such (non-pairwise) surrogate regret bounds for bipartite ranking can be obtained in terms of a broad class of proper (composite) losses that we term as strongly proper. Our proof technique is much simpler than that of Kotlowski et al. (2011), and relies on properties of proper (composite) losses as elucidated recently by Reid and Williamson (2010, 2011) and others. Our result yields explicit surrogate bounds (with no hidden balancing terms) in terms of a variety of strongly proper losses, including for example logistic, exponential, squared and squared hinge losses as special cases. An important consequence is that standard algorithms minimizing a (non-pairwise) strongly proper loss, such as logistic regression and boosting algorithms (assuming a universal function class and appropriate regularization), are in fact consistent for bipartite ranking; moreover, our results allow us to quantify the bipartite ranking regret in terms of the corresponding surrogate regret. We also obtain tighter surrogate bounds under certain low-noise conditions via a recent result of Clemencon and Robbiano (2011).

Assessment of post-harvest losses in Nigeria fishery: the Kainji Lake model

The paper provides a description of a methodology used for quantitative assessment of post harvest losses in the Kainji Lake Fishery (Nigeria). The sample population was made up of 314 fisherfolk, 115 processors, 125 fish buyers and 111 fish sellers. For the determination of handling losses, 24,839 fishes weighing 2,389.31 kg belonging to 43 species were examined of which 10% by number and 9% by weight deteriorated at checking and 4% by number and 3% by weight at landing. Processing losses recorded 22% by number and 16% by weight deteriorated prior to and during smoking with the traditional 'Banda' kiln. During marketing, 16% of fish sold had deteriorated and 6% by weight of fish bought also deteriorated, mainly due to insect infestation during storage. Based on the 1995 yield estimate for Kainji Lake fishery, approximately 1000 tons of fish estimated at 80 million Naira were lost during handling alone. This figure would be much higher if the level of losses during processing and marketing are included. This assessment technique is recommended for use in obtaining quantifiable data on post harvest losses from other water bodies in Nigeria

A preliminary investigation into the post harvest losses of fish in Shiroro Lake area

The findings are presented of a study conducted to assess the post harvest losses in Shiroro Lake, Nigeria. The major objectives were to identify and quantify the types of losses, to provide recommendations that would enhance formulation of policy guidelines for utilization and exploitation of the declining fishery resources of the lake

Post harvest losses in the fisheries of Kainji Lake

An assessment is given of the post-harvest losses in the Lake Kainji fisheries of Nigeria. The study focussed on quantifiable information on post-harvest technology and post-harvest losses from fisherfolk, fish processors and fish traders operating within the Kainji Lake basin. The information was obtained from questionnaires sent to a total of 665 respondents, comprising 317 fishermen, 115 fish processors, 125 fish buyers, and 111 fish sellers in 45 fishing villages and collection centres within the lake basin. Considering the total catch from gillnets, longlines, traps and cast nets estimated at 14,000 in 1995 about 1,000 t of fish was either discarded or lost value due to spoilage during handling by fisherfolk. Assuming an average prices of 80 Naira/kg of fish, the loss to the economy amounted to 80 million Naira annually. Appropriate recommendations are made to significantly reduce post-harvest losses in the Kainji Lake fishery. (PDF contains 91 pages)

Economic Assessment of Forest Ecosystem Services Losses: Cost of Policy Inaction

52 p.

Modelling the Effect of Climate Change on Environmental Pollution Losses from Dairy Systems in the UK

21 p.

Electric and magnetic losses and gains in determining the sign of refractive index

Within the framework of classic electromagnetic theories, we have studied the sign of refractive index of optical medias with the emphases on the roles of the electric and magnetic losses and gains. Starting from the Maxwell equations for an isotropic and homogeneous media, we have derived the general form of the complex refractive index and its relation with the complex electric permittivity and magnetic permeability, i.e. n = root epsilon mu, in which the intrinsic electric and magnetic losses and gains are included as the imaginary parts of the complex permittivity and permeability, respectively, as epsilon = epsilon(r) + i(epsilon i) and mu = mu(r) + i mu(i). The electric and magnetic losses are present in all passive materials, which correspond, respectively, to the positive imaginary permittivity and permeability epsilon(i) > 0 and mu(i) > 0. The electric and magnetic gains are present in materials where external pumping sources enable the light to be amplified instead of attenuated, which correspond, respectively, to the negative imaginary permittivity and permeability epsilon(i) < 0 and mu(i) < 0. We have analyzed and determined uniquely the sign of the refractive index, for all possible combinations of the four parameters epsilon(r), mu(r), epsilon(i), and mu(i), in light of the relativistic causality. A causal solution requires that the wave impedance be positive Re {Z} > 0. We illustrate the results for all cases in tables of the sign of refractive index. One of the most important messages from the sign tables is that, apart from the well-known case where simultaneously epsilon < 0 and mu < 0, there are other possibilities for the refractive index to be negative n < 0, for example, for epsilon(r) < 0, mu(r) > 0, epsilon(i) > 0, and mu(i) > 0, the refractive index is negative n < 0 provided mu(i)/epsilon(i) > mu(r)/vertical bar epsilon(r)vertical bar. (c) 2006 Elsevier B.V. All rights reserved.