ACM Comm 2010 May Technical Perspective Learning to act in Uncertain Environments (Notes)

"Beyond Total Capture: A Constructive Critique of Lifelogging" CACM May 2010

Technical Perspective Learning to act in Uncertain Environments by Peter L. Bartlett, p.98

Technical Perspective Learning to act in Uncertain Environments

"A Darwin Machine is by definition an uncertain environment."

People

1. Kuzman Ganchev
2. Yuriy Nevmyvaka
3. Michael Kearns
4. Jennifer Wortman Vaughan

Ideas

1. Uncertain Environments
   1. Information about the environment has an unknown quality.
   2. Information about the environment has an unknown age.
   3. The environment can not be known in totality.
   4. Enough about the environment can not be known to make a decision with total confidence.
2. Exploration/Exploitation Trade-Off Should an explored path be chosen or should a new path be chosen hoping for something better?
3. This leads to risk assessment. What is the cost for each path? Given a good outcome? Given a bad outcome?
4. Optimism Heuristic Treat uncertain outcomes as optimistically as the data allows: pick the alternative that, in the best possible world, is consistent with our experiences so far, and leads to the best outcome.
5. Kaplan-Meier Estimator[1] Is a non-parametric statistic used to estimate the survival function from lifetime data.

References

1. Censored Exploration and the Dark Pool Problem By Kuzman Ganchev, Yuriy Nevmyvaka, Michael Kearns, and Jennifer Wortman Vaughan, CACM May 2010.

Internal Links

Parent Article: Reading Notes