Differences

This shows you the differences between two versions of the page.

--- metarelation_mining [2021/08/27 17:03]
ymkw
+++ metarelation_mining [2021/08/31 18:34] (current)
ymkw
@@ Line 1: / Line 1: @@
 ====== Metarelation Mining ======
 === Technology Overview ===
-　"Metarelation Mining" is a technique for "finding relations within relations. It is a technique for finding meta-relationships between "pairs of series." A "pair" is a relationship between multidimensional series that have no distinguishing attributes. A pair is an ordered set (similar to a list). What we are focusing on and working on is Equivalence Structure (ES) extraction, which finds equivalent meta-relations.
-----
+Metarelation mining is a process used to find “relations of relations.” We focus on relations between sequences where their roles (attributes) are unknown. In this case, metarelations of interest are relations between such relations. A technique for metarelation mining we focus on is called **equivalence structure (ES) extraction**, which allows for finding analogous metarelations (Fig. 1).
-{{ :figure-meta-relation_mining.png?600 |}}
- Figure: Example of equivalence structure extraction. From the series of #1 to #8 (series), we extract two pairs, 〈#1, #2, #3〉 and 〈#8, #7, #5〉, and verify whether there is a meta-relationship between these two pairs. In this example, the three subsequences (blue, red, and black) are very similar, so we consider them to be equivalent meta-relations.
-----
+{{ :fig1.png?600 |}}
- In equivalence structure extraction, when two or more pairs are compared and "contain similar subsequences" at times that are not necessarily simultaneous (asynchronous), a structure with a meta-relationship called "equivalence" is considered to have been discovered. This extraction has a feature that can be used even in cases where the attributes are unknown. An example of "asynchronous" and "unknown attribute" data is the task of finding equivalent relations in human and chimpanzee brain waves. Since such data is generally not synchronized, it is necessary to absorb the differences in time. Also, unknown attributes here means that we do not know how the attributes (functions and meanings) of each brain wave correspond to each other in humans and chimpanzees. Finding an equivalence structure between such data can help us find out which series of EEGs correspond to each other. Here, the determination of whether two pairs are equivalent is based on comparing the subsequences and whether they are similar enough. This allows us to discover relationships even when the relationships are asynchronous and the attributes are unknown.
+	Fig. 1: An illustration of ES extraction. Two three-dimensional sequences specified by tuples <#1, #2, #3> and <#8, #7, #5> are compared to validate if there is a metarelation between the three-dimensional sequences. Because a subsequence of one three-dimensional sequence shown by a blue, a red and a black box closely resembles a subsequence of the other 3d sequence, they are likely to be considered to have an analogous metarelation.
+In ES extraction, an analogous metarelation between two multidimensional sequences is considered found based on the comparisons of their subsequences. The implementation of such comparisons allows for finding relations between asynchronous sequences. Therefore, ES extraction can be applicable to data that are “asynchronous” and “the attribute of each sequence is unknown.” A task using such data can be a task to find metarelations between the brainwaves of a human and those of a chimpanzee. Such data are usually asynchronous, and it is unknown how the brainwaves of a human correspond to those of a chimp, which means that the attribute of each sequence is unknown. Metarelations found in such data can be useful to find correspondence relations between the brainwaves of a human and those of a chimpanzee.
+One difficulty in the ES extraction is that the brute-force search is usually not feasible. Given N sequences, the number of K-tuples is K-permutations of N, which causes a combinatorial explosion. Moreover, comparisons of subsequences are implemented for each comparison of two K-dimensional sequences specified by K-tuples. To reduce the processing time, we recently propose a method called ESIS, in which ESs are obtained increasing K. In the future, we plan to propose a yet faster method.
-　The major problem with ES extraction is its large computational complexity. If the number of all series is N and the size of a pair is K, the number of pairs to be searched is K permutations out of N. Furthermore, it is necessary to compare sub-sequences among them. We proposed a search method (ESIS) that sequentially increases the number of K permutations, instead of directly calculating K permutations, to reduce the computation time. We are currently working on further reducing the computation time in order to improve the practicality.
 === Related works ===
@@ Line 29: / Line 29: @@
   * Hiroshi, Yamakawa. “Hippocampal Formation Mechanism Will Inspire Frame Generation for Building an Artificial General Intelligence’’. In Proc. of the International Conference on Artificial General Intelligence (AGI), 2012. pp.362--371.
-=== 連絡先 ===
+=== Contact ===
-佐藤聖也(AIST) ,  山川宏(全脳アーキテクチャ・イニシアティブ)
+Seiya Sato (Tokyo Denki University) ,  Hiroshi Yamakawa (The Whole Brain Architecture Initiative)