New data from the NIH reveal that the scientific return on its sponsored research reaches a maximum at around $400,000 of annual support per principal investigator. We discuss the implications of this 'sweet spot' for funding policy, and propose that the NIH should limit both the minimum and maximum amount of funding per researcher.

We examine two policies for reopening of nodes: never reopen (NR) and always reopen (AR). While there are circumstances where each policy is beneficial, we observed empirically that NR is usually faster. However, NR may fail to return a solution of the desired quality in two scenarios: (1) in a bounded suboptimal search when inconsistent heuristics are used and (2) in a bounded cost setting. To...

Policy gradient methods have been widely applied in reinforcement learning. For reasons of safety and cost, learning is often conducted using a simulator. However, learning in simulation does not traditionally utilise the opportunity to improve learning by adjusting certain environment variables – state features that are randomly determined by the environment in a physical setting but controlla...

Firm cash holdings increased substantially from 1980 to 2013. The overall distribution of firm cash holdings changed in the same period. We study the implications of these changes for monetary policy. We use Compustat data and a model with financial frictions that allows the calculation of the monetary policy effects according to the distribution of cash holdings. We find that the interest rate...

We consider a resource allocation problem, where a rational agent has to decide how to share a limited amount of resources among different companies that might be facing financial difficulties. The objective is to minimize the total long term cost incurred by the economy due to default events. Using the framework of multiarmed restless bandits and, assuming a two-state evolution of the default ...

Policy search is a successful approach to reinforcement learning. However, policy improvements often result in the loss of information. Hence, it has been marred by premature convergence and implausible solutions. As first suggested in the context of covariant or natural policy gradients, many of these problems may be addressed by constraining the information loss. In this paper, we continue th...

This year sees the 40th anniversary of the first policy paper regarding the use of computers in higher education in the United Kingdom. The publication of this paper represented the beginning of the field of learning technology research and practice in higher education. In the past 40 years, policy has at various points drawn from different communities and provided the roots for a diverse field...

E agree about the goals of price stability, low unemployment and stable economic growth, but they disagree about the policies to achieve these goals. The disagreement is particularly heated over discretionary countercyclical Keynesian fiscal policy. After the poor macroeconomic performance of the 1970s and critical policy evaluations of the Keynesian approach— ranging from Robert E. Lucas and T...

Policy gradient methods have been widely applied in reinforcement learning. For reasons of safety and cost, learning is often conducted using a simulator. However, learning in simulation does not traditionally utilise the opportunity to improve learning by adjusting certain environment variables – state features that are randomly determined by the environment in a physical setting but controlla...