Gregory N. Hullender

Learning to rank using gradient descent

We investigate using gradient descent methods for learning ranking functions; we propose a simple probabilistic cost function, and we introduce RankNet, an implementation of these ideas using a neural network to model the underlying ranking function. We present test results on toy data and on data from a commercial internet search engine.

An efficient method for tuning handwriting parameters

Many problems in recognition involve making linear combinations of results from various experts. Computing the coefficients can be expensive because a single test run can take many minutes, hours, or even days, and many values need to be evaluated to find an optimum. This paper describes an algorithm that can make a good approximation to the optimum value for a parameter in a single pass over the tuning data and also outlines methods for tuning several parameters at once.