Recently a number of algorithms under the theme of 'unbiased learning-to-rank' have been proposed, which can reduce position bias, the major type of bias in click data, and train a highperformance ranker with click data in learning-to-rank. Most of the existing algorithms, based on the inverse propensity weighting (IPW) principle, first estimate the click bias at each position, and then train an unbiased ranker with the estimated biases using a learning-to-rank algorithm. However, there has not been a method for unbiased pairwise learning-to-rank that can simultaneously conduct debiasing of click data and training of a ranker using a pairwise loss function. In this paper, we propose a novel framework to accomplish the goal and apply this framework to the state-ofthe-art pairwise learning-to-rank algorithm, LambdaMART. Our algorithm named Unbiased LambdaMART can jointly estimate the biases at click positions and the biases at unclick positions, and learn an unbiased ranker. Experiments on benchmark data show that Unbiased LambdaMART can significantly outperform existing algorithms by large margins. In addition, an online A/B Testing at a commercial search engine shows that Unbiased LambdaMART can effectively conduct debiasing of click data and enhance relevance ranking.

No abstract

Thermodynamics and dynamics of DNA hybridization/dehybridization at a terminal of a DNA duplex are investigated using steady-state fluorescence and fluorescence correlation spectroscopy (FCS). We introduce two pairs of dyes with different characteristic distances in fluorescence resonance energy transfer (FRET) to investigate the same process. The phenomenal discrepancy in the experimental observations between our two FRET pairs is incompatible with the traditional two-state model. We propose a so-called stretched exponential zipper (SEZ) model to successfully analyze the experimental data, through which the fundamental behavior of base-by-base hybridization/dehybridization is revealed. The dynamic parameters of the activation energy of single base-pair reaction derived from the two FRET pairs are consistent. The enthalpy change and entropy change of single base-pair formation are in agreement with theoretical prediction.

Membrane filtration operations (ultra-, microfiltration) are now extensively used for concentrating or separating an ever-growing variety of colloidal dispersions. However, the phenomena that determine the efficiency of these operations are not yet fully understood. This is especially the case when dealing with colloids that are soft, deformable, and permeable. In this paper, we propose a methodology for building a model that is able to predict the performance (flux, concentration profiles) of the filtration of such objects in relation with the operating conditions. This is done by focusing on the case of milk filtration, all experiments being performed with dispersions of milk casein micelles, which are sort of ″natural″ colloidal microgels. Using this example, we develop the general idea that a filtration model can always be built for a given colloidal dispersion as long as this dispersion has been characterized in terms of osmotic pressure Π and hydraulic permeability k. For soft and permeable colloids, the major issue is that the permeability k cannot be assessed in a trivial way like in the case for hard-sphere colloids. To get around this difficulty, we follow two distinct approaches to actually measure k: a direct approach, involving osmotic stress experiments, and a reverse-calculation approach, that consists of estimating k through well-controlled filtration experiments. The resulting filtration model is then validated against experimental measurements obtained from combined milk filtration/SAXS experiments. We also give precise examples of how the model can be used, as well as a brief discussion on the possible universality of the approach presented here.