My thoughts on Rotating Neurons -- RDBMS & AI - memory operations, neural networks

In this blog post,  I want to share my thoughts on a new subject. 

Actually, an article impressed me and I want to share it with you. 

These ideas may be considered for new developments on the Oracle side. 

Well, the article is based on rotating memories. I read it on quantamagazine. This article made me realize that, insiring from the nature is important in computer science. We already see that in the developments done for AI. (not in all of them ofcouse..) 

Getting the information from neuroscience and bringing it to the computer science and supporting the implementation with the engineering brings lots of good things and/or revolutions to our world. 

This understanding brings lots of advantages when we imitiate in certain scales. Or when we take advantage from it to enhance our processes in the functional , in behaviorual levels.. 

The work/article that I mentioned is also made in this context.. 

It is basically done for understanding how the brain works. It is promising..

https://www.quantamagazine.org/the-brain-rotates-memories-to-save-them-from-new-sensations-20210415 

This rotation that is mentioned in the article seems to be done on a representation.. This representation is pretty low level.. 

So at first glance, I thought about the database world, and the in memory stores that we have in Oracle Database ( Oracle Database In Memory - seperate stores for row and column formats , for OLTP and OLAP...) But then I though that, this memory concepts in Oracle Database is a high level representation, so the thing that is given in the article is directly not applicable to the DB side. 

On the other hand, as the article suggests, with this info ( I mean rotating neurons), a new neural network design can be made. 

As we are going to a target where we have embedded AI in Oracle Database, this is still in the context.. Well, this new neural network may be done by having directions in addition to the weights. (in terms of vectoral directions -- as an extra parameter) -- by designing a proper vector arithmetic and implementing it in a way that makes the directions to be considered for the syncrohizations... 

This may be an interesting idea and it is space efficient. The advantage is having a compact space -- less memory and more efficient multi tasking.. 

As for the high level representation that we have in the database layer, one might think enhancing the block format or having a transformation mechanism while reading the blocks in to the memory pages.. 

I mean a dynamic and intelligent format maybe.. Some new relations in the block level... This may bring lots of good opportunities to us ( in addition to the faster buffer scan and faster buffer access. ) 

This task, I mean getting it to the high level is ofcourse very hard -- no matter how much we abstract the subject.. 

Maybe the solution is to change the low level on which the high level is based. It can use a certain rotation angle at low level for the operations to be done at the high level. 

This ofcourse touches the foundation.. Anyways, interesting subject.. I wanted to share it with you, maybe we may see that in the development of  database, in the new generations...