Any Neuromorphic Hardware available for supporting HTM?


#1

Does anyone know if there is any Neuromorphic Hardware, like an IBM True North chip or a Qualcomm Snapdragon 835 which could be integrated into a PC motherboard, perhaps using a graphics port? I read that neuromorphic hardware is out there and some may already be commercialized. Where would one find more information to purchase some?


#2

https://en.wikipedia.org/wiki/Field-programmable_gate_array
https://www.micronautomata.com/
https://www.parallella.org/2016/10/05/epiphany-v-a-1024-core-64-bit-risc-processor/


#3

@BrainConstellation I think SpiNNaker will be most suitable due to high performance and flexibility


#4

Thank you, Sean, for your quick tip. FPGAs seem to be a very broad field with XILINX, ALTERA, HUMANDATA etc. all providing different platforms. Do have any configuration experience with these HW platforms in a PC environment? And if yes, is there any HTM work that can be re-used or viewed as a reference?

I would not mind embarking on such a project but would like to know what is already out there in the HTM community before re-doing the same effort.


#5

Thanks for the tip. I just checked out their website from the UK. I will try to read up on this. Do you have any previous experience with SpiNNaker and HTM?


#6

@BrainConstellation: No, we do not have any SpiNNaker, but I know that NUPIC/CLA is ported somehows into Hardware in Human Brain Project, and maybe there are experiments with SpiNNaker and NuPIC …Let’s ask in HBP community


#7

To my knowledge HTM has not yet been ported to SpiNNaker but I think @thanh-binh.to is correct, the architecture should be very suitable. We did do another HBP project with the Heidelberg Neuromorphic Computing Platform. That work is written up here:

https://arxiv.org/abs/1505.02142

It would be great if someone asked the HBP community - there may be implementations we’re unaware of.


#8

Next week, I will come to a workshop of HBP community and will ask them about the current state?


#9

Thanks a lot for bringing this question with you to the HBP workshop.

Joe


#10

HBP recently presented at CERN a talk about “Brain inspired hardware architectures”.
At slide 30 they talk about the Brainscales project (what Subutai’s article concerns) and at slide 52 they mention HTM.


#11

Wow, that’s cool, @Oblynx! Thanks for sharing. For those of you who didn’t dive into the PDF, have a look at the slides on HTM:


#12

I was looking for someone else’s take on HTM, from outside this community, and am glad they noticed! Are you at Numenta continuing the work of Subutai’s paper on porting HTM to the Brainscales system? Karlheinz Meier, who made that presentation, is at the core of Brainscales.


#13

There are no current HTM efforts with that system that I am aware of. Sebastian (who really did most of the work on the paper :slight_smile: ) was a graduate student of Karlheinz Meier and did the research while in our offices. I believe Sebastian graduated, and another graduate student or collaborator will need to take it on. (We don’t have the hardware expertise to do these projects ourselves.)


#14

No further activity on Htm/cla in HBP consortium. Someone want to use a mini-column model into SpiNNaker, but i do not know in details