Finite Factored Sets: Inferring Time
post by Scott Garrabrant · 2021-08-31T21:18:36.189Z · LW · GW · 5 commentsContents
6.1. Factored Set Models 6.2. Examples None 5 comments
The fundamental theorem [LW · GW] of finite factored sets [? · GW] tells us that (conditional [? · GW]) orthogonality [? · GW] data can be inferred from probabilistic data. Thus, if we can infer temporal data from orthogonality data, we will be able to combine these to infer temporal data purely from probabilistic data. In this section, we will discuss the problem of inferring temporal data from orthogonality data, mostly by going through a couple of examples.
6.1. Factored Set Models
We'll begin with a sample space, .
Naively, one might except that temporal inference in this paradigm involves inferring a factorization of . What we'll actually be doing, however, is inferring a factored set model of . This will allow for the possibility that some situations are distinct without being distinct in —that there can be latent structure not represented in .
Definition 38 (model). Given a set , a model of is a pair , where is a finite factored set and is a function from the set of to .
Definition 39. Let and be sets, and let be a function from to .
Given a , we let .
Given an , we let .
Given an , we let be given by .
Definition 40 (orthogonality database). Given a set , an orthogonality database on is a pair , where and are both subsets of .
Definition 41. Given an orthogonality database on a set , and partitions , we write if , and we write if .
Definition 42. Given a set , a model of , and an orthogonality database on , we say models if for all ,
- if then , and
- if then .
Definition 43. An orthogonality database on a set is called consistent if there exists a model of such that models .
Definition 44. An orthogonality database on a set is called complete if for all , either or .
Definition 45. Given a set , an orthogonality database on , and , we say if for all models of that model , we have .
6.2. Examples
Example 1. Let be the set of all bit strings of length . For , let be the event that the first bit is , and let be the event that the second bit is i. Let and let .
Let be the event that the two bits are equal, let be the event that the two bits are unequal, and let .
Let , where and .
Proposition 33. In Example 1, is consistent.
Proof. First observe that is a factored set, and so is a model of , where is the identity on . It suffices to show that models .
Indeed , and , so , so .
Further, it is not the case that , since . Thus it is not the case that .
Thus satisfies all of the conditions to model , so is consistent.
Proposition 34. In Example 1, .
Proof. Let be any model of that models . Let . For any , let . Our goal is to show that is a strict subset of .
First observe that , so for any , if and , then and , so , so . Thus .
It follows that . However, since , we have that , so .
By swapping and in the argument above, we also get that . Since , we have that . Thus contains some element . Observe that , but . Thus is a strict subset of , so .
Since was an arbitrary model of that models , this implies that .
Example 2. Let be the set of all bit strings of length . For , let be the event that the first bit is , let be the event that the second bit is , and let be the event that the third bit is . Let , let , and let .
Let be the event that the first two bits are equal, let be the event that the first two bits are unequal, and let .
Let , where and .
Proposition 35. In Example 2, is consistent.
Proof. Let be the set of all bit strings of length either or .
For , let be the event that the first bit is , and let .
For , let be the event that the second bit is , and let .
Let be the event that the first two bits are equal, let be the event that the first two bits are unequal, and let .
For , let be the event that the third bit exists and is , let be the event that there are only two bits, and let .
Let . Clearly, is a finite factored set.
Let be given by if , , , , and , so copies the last bit on inputs of length , and otherwise leaves the bit string alone. We will show that models .
First, observe that , , , and .
It is easy to verify that , and . From this, we get that holds, but and do not hold.
Next, observe that for , . It is easy to verify that .
Also, observe that , and observe that . It is easy to verify that .
From this, we get that and hold, and does not hold.
Thus, models , so is consistent.
Proposition 36. In Example 2, .
Proof. Let be any model of that models . Let . For any , let . Our goal is to show that is a strict subset of and that is a strict subset of .
First observe that , so , so . Since , , so . Symmetrically, , so .
Similarly, , so . Thus .
We also know that and are nonempty, because and .
Thus is a strict subset of , so .
Let be arbitrary such that and are both nonempty. Fix some and .
Since , there must exist such that for all , but not . Thus it is not the case that . Without loss of generality, assume that and .
Similarly, since , there must exist such that for all , but not . Again, without loss of generality, assume that and .
For , let .
Next, observe that , so , so . Similarly, , so . Thus, if , and if , .
Further, observe that , since and agree on all factors other than and . In particular, this means that . Similarly, since , we have that .
We will use this to show that for any and , either , or . This is because , so , so by the above argument, if is nonempty, then , which since is nonempty means is nonempty, so , so . Symmetrically, we also have that if is nonempty, then . Thus, if is nonempty, then either or is nonempty, so .
Note that for any , two of the elements among the four defined above are in , and those two elements are in different parts in , so has at least two parts, so is nonempty. However, . Thus, , so , so . Symmetrically, .
In particular, this means that , since .
Since , there exists some . Since , there exist such that for all , but it is not the case that . Without loss of generality, assume that and . Let .
Let be an arbitrary element of . Since , there exist such that for all , but it is not the case that . Without loss of generality, assume that and .
Consider . Since , . Since is also in , . However, since , we have , so .
If were in , we would similarly have , which would contradict the fact that it is not the case that . Thus .
Next, consider . Since , . Since is also not in , . However, since , we have , so .
Thus, it is not the case that . However, we constructed and such that for all . Thus . Since was arbitrary in , we have that . Finally, we need to show that this subset relation is strict.
Since , there is some such that . Let be any element of . Since , . However, . Therefore . Thus is a strict subset of , so .
In the next post, we'll discuss applications and future research directions.
5 comments
Comments sorted by top scores.
comment by Rohin Shah (rohinmshah) · 2021-09-06T09:24:46.023Z · LW(p) · GW(p)
For Example 2 / Prop 35, would this model also work?
Define to be the factor corresponding to the question "are the second and third bits equal or not?" Then is a model of . I believe this is consistent with :
For :
We have and for the first condition.
We have and for the second condition.
We have and for the third condition.
For .
We have for the first and second conditions.
We have for the third condition.
Replies from: Scott Garrabrant↑ comment by Scott Garrabrant · 2021-09-06T10:54:04.712Z · LW(p) · GW(p)
I think that works, I didn't look very hard. Yore histories of X given Y and V given Y are wrong, but it doesn't change the conclusion.
Replies from: rohinmshah↑ comment by Rohin Shah (rohinmshah) · 2021-09-06T14:21:42.821Z · LW(p) · GW(p)
Yeah, both of those should be , if I'm not mistaken (a second time).
Replies from: Scott Garrabrant↑ comment by Scott Garrabrant · 2021-09-06T20:13:06.957Z · LW(p) · GW(p)
Yeah, also note that the history of given is not actually a well defined concept. There is only the history of given for . You could define it to be the union of all of those, but that would not actually be used in the definition of orthogonality. In this case , , and are all independent of choice of , but in general, you should be careful about that.
Replies from: rohinmshah