 |
||||
| Decision and Reinforcement Learning |
In the Auton lab, we focus on machine learning, especially data mining, reinforcement learning, control and logistics. Today information is processed in a huge variety of places, ranging from humble thermostats, traffic lights and TV remote controls up to continuous-flow manufacturing processes, packaging machines, truck routers, car engine management systems and plant management consoles. Here's what we find exciting: Imagine a future where all these systems have the computational power of a thousand Deep Blue chess search engines. The hardware and infrastructure for this future are not unthinkably far off. We are concerned with the computer science of exploiting this power.
The basic computer science research in the Auton lab is inspired by a belief that realistic AI deployment over the coming decades face two outstanding issues:
Although there will always be an important role for statistics as a methodology that human experts use to analyze data, we will also see a growing need for autonomous systems which communicate with human non-experts, or which are embedded deep within control systems. They must learn patterns, trends and models from experience with very little outside help.
It is rarely the case that systems learn to predict as an end in itself. AI systems must make decisions, and learning is only a part of the decision process. Again, in some cases human engineers and managers can be involved. But there are increasingly many parts of a corporation in which the sheer complexity, the speed, and the 24-hour nature of operations necessitate autonomy.
|
|
||||
| Decision and Reinforcement Learning |