# Self organising maps

**URL:** <https://community.konduit.ai/t/self-organising-maps/116>\
**Category:** DL4J\
**Created:** [January 31, 2020, 9:03am UTC](https://community.konduit.ai/t/self-organising-maps/116 "2020-01-31T09:03:55Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![raviteja777](https://yyz1.discourse-cdn.com/flex035/user_avatar/community.konduit.ai/raviteja777/32/269_2.png) [@raviteja777](https://community.konduit.ai/u/raviteja777)\
**Post date:** [January 31, 2020, 9:03am UTC](https://community.konduit.ai/t/self-organising-maps/116/1 "2020-01-31T09:03:55Z")

</div>

Is there anyway to create self organising maps in dl4j ?

---

<div class="post-metadata">

**Author:** ![treo](https://yyz1.discourse-cdn.com/flex035/user_avatar/community.konduit.ai/treo/32/47_2.png) [@treo](https://community.konduit.ai/u/treo)\
**Post date:** [February 3, 2020, 9:54am UTC](https://community.konduit.ai/t/self-organising-maps/116/2 "2020-02-03T09:54:23Z")

</div>

There is no out of the box support for self organizing maps in DL4J.  
Even though they are often considered to be a type of neural network, the way that they actually work and are trained is very different from how a classical neural network is trained.

If you want to implement them yourself, you might still be interested in using ND4J for it.

---

<div class="post-metadata">

**Author:** ![S6Regen](https://avatars.discourse-cdn.com/v4/letter/s/838e76/32.png) [@S6Regen](https://community.konduit.ai/u/S6Regen)\
**Post date:** [March 20, 2020, 11:29pm UTC](https://community.konduit.ai/t/self-organising-maps/116/3 "2020-03-20T23:29:40Z")

</div>

This is a very useful book for such things:  
[https://archive.org/details/SelfOrganizationAndAssociativeMemory/mode/2up](https://archive.org/details/SelfOrganizationAndAssociativeMemory/mode/2up)

It is also one of the few places I’ve seen any discussion of the error correction effects of the weighted sum when used as a recall device. The variance equation for linear combinations of random variables also tells you that and the central limit theorem. Overparameterization then can lead to error correction instead of overfitting as an example consequence.  
Basically mild hashing before the weighted sum, together with some training algorithm allows it to act as a general associative memory.  
And so what?  
Well it seems to me that would make a nice sort of memory for a Neural Turing Machine, that is a neural network with an external memory bank.