Probability diagram Model 7: Inference and Flow

1. Inference mode

There are three types of inference models for probabilistic graph models:

causal reasoning/causal reasoning : That is, when you know a cause of an event, we change the estimate of the outcome of the event.

evidential reasoning/Evidence Reasoning : That is, when we know the outcome of the event, we change the judgment of the cause of the event.

intercausal reasoning/Reasoning : Some people translate it into causal reasoning, which I think is prone to misunderstanding. Intercausal reasoning means that there are multiple reasons for an event to occur, and if we already know the outcome of the event, then the observation of one of these causes will change our judgment on the other.

The following is an introduction to the Bayesian network in Figure 1.1.

Figure 1.1: A simple Bayesian network

1.1 Causal Reasoning

As I said earlier, this mode of reasoning is a way of knowing the cause of the event and we will change the estimate of the outcome of the event. This is a pattern of reasoning that best fits our human perceptions. For example, in Figure 1.1, we can know that P (L1) ≈0.5 P (l^1) \approx 0.5, the specific method, is based on the result of factorization:
P (L=L1) =∑d,i,g,sp (D,I,G,S,L=L1) =∑d,i,g,sp (D) p (I) p (g| I,D) P (s| I) P (l=l1| G) (1.1) p (l=l^1) =\sum_{d,i,g,s} p (d,i,g,s,l=l^1) =\sum_{d,i,g,s}p (D) p (I) p (g| I,D) P (s| I) P (l=l^1| G) \tag{1.1}
Similarly, we can also obtain P (l1|i0) ≈0.39 P (l^1|i^0) \approx 0.39:
P (L=L1) =∑d,g,sp (D,I=I0,G,S,L=L1) =∑d,g,sp (D) p (i=i0) p (g| I=I0,D) P (S