阅读训练(day04)

A brief summary for :
Computational rationality: A converging paradigm for intelligence in brains, minds, and machines

Computational rationality based on three core ideas:

1. We aim to plan for actions for maximizing expected utility.
2. For real-world problems, we can only use effectively approximated rational algorithms.
3. Algortihms can be adpted for specific needs.(offline, online).

Computational tradeoffs in sequential decision-making:

1. Researchers have shown that the human brain might use an algorithm like MCTS to sovle spatial navigation problmes. 【interesting！！！】
2. Hybird model-free and model-based decision-making systems may be a promising route to explain how human decision-making in complex sequential taks can be so accurate and fast.

Rational decisions under bounded computational resources

1. The complexity of probabilistic inference in Bayesian networks has been shown to be in the nondeterministic polynomial-time (NP)–hard complexity class.
2. A tapestry of approximate inferential methods have been developed, include Monte Carlo simulation, modulating the complexity of models(changing the size or level of abstraction)