Gradient-based approaches to direct policy search in reinforcement learning have received much recent attention as a means to solve problems of partial observability and to avoid some of the problems associated with policy degradation in value-function methods. In this paper we introduce GPOMDP,...

There are two important rules in a patent race: what an innovator must accomplish to receive the patent and the allocation of the benefits that flow from the innovation. Most patent races end before R&D is completed and the prize to the innovator is often less than the social benefit of the...

There exist many opportunities for deploying autonomic computing in an IT environment. The highest-value opportunities are going to be where we can reduce human decision-making complexity for systems administrators. To identify these opportunities, we need a model of decision complexity for...

A credit derivative is a financial instrument whose value depends on the credit risk of an underlying asset or assets. Credit risk is the possibility that the obligor fails to honor any payment obligation. This thesis proposes four new computational methods for the valuation of credit...

We propose a new numerical method to solve stochastic models that combines the parameterized expectations (PEA) and the Smolyak algorithms. This method is especially convenient to address problems with occasionally binding constraints (a feature inherited from PEA) and/or a large number of state...