About the “why” model in tutorial 3 of this morning’s videos, an entropy function is used to form an optimization problem. The entropy of the exponential curve is found to be the maximum so it tells why the interval data histogram is best fitted by an exponential function.

I find this a bit “occasional” to adopt this “entropy” way to form an optimal “why” question. There might be many different angles to form such opt questions. Could you shed some light on other “why” models to better my understanding, and ideas to build them?

Here they only compare three models: Exponential, Unimodel and deterministic. When they calculate the entropy, they didn’t use the raw data. Therefore, it’s the entropy of different types of distribution, not for the data. Based on that, they make conclusion that exponential distribution gives more information than the others, even more than the empirical distribution based on the raw data.

Interesting that the exponential model has the maximum entropy, and is better than the real data.

Video this afternoon explains the “why” model further.

I think “reinforcement learning model” is a kind of “why” model which uses rewarding system to give a reason of animals behaviours. It is also a bit “how” if the model uses neuron-graphs to show how animals think with and without rewarding.