#201805 English Freetalk

05-29

來自專欄意識型數據結構啟動子（MDSS）

##How Do We Define the Similarity?

Meme is a well-known psychological conception that describe the stable and similar parts that deeply dwelled in different cultures that we could easily identify. Intuitively, this noun was created by imitating the word "gene" in the biological territory. This implies that memes could be inherited and interfered just like the gene fragments on the DNA molecules.

But remember not to put collective consciousness equivalent to it. Memes are so identical that when youre talking about them, youve clearly recognized theyre some memes. Sometimes, they look like to be created and casted deliberately by some prophets.

So, as studying on genes could help us to understand human-bodies, studying on memes is certainly helpful to explore the unknown aspects of human-minds.

Respectively, culture itself and meme are playing two opponent roles of social level mind-interaction. Culture tends to isolate different societies, while meme keeps the ground consistency of them. This statement is a little bit similar to the babel story recorded in the Bibel.

Speaking of ancient legends, recently Ive read about the prehistory mythology of Japan called "古事記、日本書紀". Society like Yayoi Japan is a typical hybrid but isolated community. You can discover a large amount of typical memes in the creation myths of Yayoi Japan. For instances, the "banana" legend and "Dont-look-back" legend.

The "banana" legend is also known as "Adams apple" legend, so itve been widely spread from east to west as long as the society has a basic explanation of the creation of lifeforms. Technically, it reflected the consistent anxious originated from the awakening of self-consciousness. Meanwhile, "Dont-look-back" legend also occupied a massive proportion of creation myths, it could be interpreted as a caveat of betraying the elementary contraction of the community.

Well, this passage is not merely written for this topic, were gonna go further about the background of those memes. As weve discussed, memes are measuring the over-all similarity of cultural environments. It is somehow like authenticators of blockchain. Perhaps we can take it as a metaphor of similarity metrics of cultures.

Here we might naturally recall the basic metrics of similarity -- the cosine similarity. The most effective context when this metrics works is to process some embedded corpus-type data or some coordinate location informations. That means it has a strong dependence to datas geometric specification -- the orthogonality.

If we expand the freedom degree of testing data, to some deconvoluted images or time-sequences, we also have got some experienced widely-applied functions to calculate the signal similarity. As for Images, we have the structural similarity, which is more favorable to photographers to measure photos distortion. As for languages, we have BLEU score which have been successfully integrated into todays neuro-translation.

But what if data has a far-more complicated structure than all above? we had to abandon all traditional static metrics and start to build a reinforced network to interpret the similarity of this hierarchy. Ironically, supervised learning here might give little help due to itself always need some metrics to create gradients for back propagation.

Finally, we turned way back to cultures and memes, if we tried to build a truly "brainish" stuff to interpret behaviors of a whole society, what should we do when facing this potential of complexity? Keep designing a brand-new static metrics for supervised learning, or straightly perform an astronomical-hyperparametered unsupervised learning, or try to reconstruct an institute to integrated them organically together?

This might be interesting if we gonna spend some attentions on.

(Background Image From: 映畫「天使のタマコ」）