# What is flowchart of Temporal memory

**URL:** <https://discourse.numenta.org/t/what-is-flowchart-of-temporal-memory/3235>\
**Category:** Numenta Theory\
**Tags:** temporal-memory\
**Created:** [January 4, 2018, 2:17am UTC](https://discourse.numenta.org/t/what-is-flowchart-of-temporal-memory/3235 "2018-01-04T02:17:59Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Rodi](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/rodi/32/3720_2.png) [@Rodi](https://discourse.numenta.org/u/Rodi)\
**Post date:** [January 4, 2018, 2:17am UTC](https://discourse.numenta.org/t/what-is-flowchart-of-temporal-memory/3235/1 "2018-01-04T02:17:59Z")

</div>

Hi All.

Can you tell me what is flowchart of memory?  
I see in spatial pooler, there is flowchart include initialization, overlap, inhibition, learning.  
I read Temporal memory code, Just I want summary steps like spatial pooler.

Thank you.

---

<div class="post-metadata">

**Author:** ![mrcslws](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/mrcslws/32/4961_2.png) [@mrcslws](https://discourse.numenta.org/u/mrcslws)\
**Post date:** [January 4, 2018, 6:58pm UTC](https://discourse.numenta.org/t/what-is-flowchart-of-temporal-memory/3235/2 "2018-01-04T18:58:26Z")

</div>

[BAMI](https://numenta.com/biological-and-machine-intelligence/) describes the TM algorithm in depth. You might find the “Temporal Memory Algorithm Pseudocode” section useful.

> **[Temporal-Memory-Algorithm-Details.pdf](https://numenta.com/assets/pdf/temporal-memory-algorithm/Temporal-Memory-Algorithm-Details.pdf)**
>
> 329.84 KB

---

<div class="post-metadata">

**Author:** ![bkutt](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/bkutt/32/1574_2.png) [@bkutt](https://discourse.numenta.org/u/bkutt)\
**Post date:** [January 4, 2018, 11:01pm UTC](https://discourse.numenta.org/t/what-is-flowchart-of-temporal-memory/3235/3 "2018-01-04T23:01:39Z")

</div>

Hi @Rodi,

The “Materials and Methods” section of the paper “Why Neurons Have Thousands of Synapses, a Theory of Sequence Memory in Neocortex” might also clear this up for you. The input to the TM algorithm is a set of activated columns (output of the spatial pooler) and the output is a new predicted network state (predicted cells to become active next) for the next timestep. At a high level, computation follows this order for a given timestep:

Output is gathered from the spatial pooler -\> Cells that should become active based on previous predictions and column activation are activated -\> New predicted cells are decided based on what just got activated -\> Synaptic weights are updated for learning purposes.