From the Google Ai DeepMind team's Alphago paper algorithm extraction learning methods: Brain Logic structure Diagram _ AI

The Bayes theorem considers P (a| b) is the likelihood that a occurs when B occurs. In reality, event A is affected by multiple events and may also be affected by the last event a itself (feedback).

Can be written in artificial intelligence general feedback Formula Y=f (x,y1), where Y1=f (x). It is the foundation of intensive learning and self-learning.

An artificial intelligence learning method extracted from Alphago's thesis algorithm

The first level of value judgment

Important----cream, remove 99% useless.

Second-level quick response (simple) and deep understanding (complex)

Simple: Common, fixed logic

Complexity: Convolution neural network (layered split calculation, seek infinite proximity value) + Monte Carlo Tree Search (select important node backward inference, get the best value)

Third-tier intensive learning

The next step in decision-making incentives

Fourth tier expert network

Incorporation into existing cognitive networks (categorized)

Fifth level self improvement

Reverse Update Self Learning

It's not like our brains.

You go out and turn around and don't jot down everything like a camera, but you choose the attention you think is important (value judgment).

If you walk the road countless times, you will not get lost, you will quickly find the shop you want to visit (quick response), this road is familiar again (intensive learning).

But if you come to a street you've never walked, you meet different people, different houses, different plants, and so on, you might be interested. At the same time, you will try to understand these features (deep understanding), label them, and deposit them in your existing cognitive system (expert network).

The more sensitive you are around, the more you care, the deeper you think (rules, algorithms, layers, the finer the split, the longer the computation time), the more the label, the more complete the things to remember (deep understanding).

When you meet again next time, you will quickly (quickly respond) to know that the person I have met before.

If this new street, the first time to give you a lot of fun (reward and punishment incentive), you must first come here.

Your value judgment automatically upgrades the street to the most worthwhile street (reverse update, updated at five levels). At the same time you are constantly recalling this happy time (self-study).