BRN Discussion Ongoing

jtardif999

Regular
If it's got neurons you would assume synapses follow

My understanding is that there's a relationship between neurons and synapses, like a sliding scale. You can decrease the number of neurons for more synapses and vice versa.

Maybe Graimatter see more value in advertising # parameters vs # of synapses




I see zero mention of learning rules associated with the device. It can't perform on-chip learning or one-shot learning..
And there will be no learning if the tech doesn’t have some form of synapses but I take your point that they may not be advertising it that way. Perhaps @Diogenese may know from any Graimatter patents whether they do?
 
  • Like
Reactions: 6 users

Xray1

Regular
Hi Guys,
remember this as well.

BrainChip Holdings Ltd

Appendix 4C and Quarterly Activities Report

For the Period Ended 30 June 2022.


Financial Update.-

Brainchip incurred $0.77M in third party licenses and hardware related to the development of

next-generation Akida engineering samples




cheers
AKIDA BALLISTA UBQTS
I wonder exactly what are the third part licenses and hardware and where they were purchased from ??
 
  • Like
Reactions: 11 users

uiux

Regular
I wonder exactly what are the third part licenses and hardware and where they were purchased from ??

Very good question
 
  • Like
Reactions: 3 users
D

Deleted member 118

Guest
Very good question
C0DFF1EA-665E-453C-B50F-F4756559AA9D.png


5B8B32EA-6BAA-452C-BC5F-84F3BF4E49C9.png
 
  • Like
Reactions: 2 users

Boab

I wish I could paint like Vincent
  • Like
Reactions: 8 users

Xray1

Regular
Very good question
If we only knew the answer to that, it would be exceptional news .....imo, I think it will have something to do with intergrating additional optionals with the new Akida 2000 that others like ARM, Megachips etc etc would like to see incorporated into same... but in the back of my mind I suppose that an expenditure of only some $0.77mill isn't really such a huge figure in the big scheme of things.
 
  • Like
Reactions: 5 users
Hi all,

Non Brainchip question so feel free to skip over it:

I can finally move a portion of my Super into a SMSF at the end of November. It’s been agonising waiting for this to occur, watching Brainchip and a few others go up whilst unable to put my super into it!

To balance my portfolio I am asking if anyone has any other solid share tips they’d like to share please?

Obviously I’m going to do my own research which is why I’m asking now as it give me a few months to consider my options and investment strategy.

Rather than clog up this thread if you could message me with any advice/share tips it would be much appreciated.

Thanks in advance. :)
 
  • Like
Reactions: 7 users
D

Deleted member 118

Guest
  • Like
Reactions: 5 users

uiux

Regular
If we only knew the answer to that, it would be exceptional news .....imo, I think it will have something to do with intergrating additional optionals with the new Akida 2000 that others like ARM, Megachips etc etc would like to see incorporated into same... but in the back of my mind I suppose that an expenditure of only some $0.77mill isn't really such a huge figure in the big scheme of things.


Is it related to memristor technology?

Another line of enquiry
 
  • Like
  • Thinking
Reactions: 5 users

Diogenese

Top 20
And there will be no learning if the tech doesn’t have some form of synapses but I take your point that they may not be advertising it that way. Perhaps @Diogenese may know from any Graimatter patents whether they do?
I'm pretty sure you can't have functioning neurons without synapses.

This grai matter patent has synapse table 23:
WO2020193689A1 NEUROMORPHIC PROCESSOR AND NEUROMORPHIC PROCESSING METHOD
In the embodiment shown in FIG. 1, the computation unit 30 is a time- shared computation unit for a plurality of neuromorphic elements. The state memory entries 10_1, , 10_i, 10_n, in the state memory 10 are individually addressable by the time-shared computation unit 30. In the embodiment of FIG. 1, the message exchange facility 20 comprises a network on chip (NoC) 21, an event-message queue 22 and a synapse table 23. Event messages provided to, and received from the network 21 are generally denoted herein as EVMout and EVMin respectively. The exchanged event-messages include messages EVMi,j, wherein i,j respectively are the identifications of the transmitting and receiving neuromorphic elements that are involved. In operation, the time-shared


1660272422643.png
 
  • Like
  • Fire
Reactions: 15 users

Taproot

Regular
Hi all,

Non Brainchip question so feel free to skip over it:

I can finally move a portion of my Super into a SMSF at the end of November. It’s been agonising waiting for this to occur, watching Brainchip and a few others go up whilst unable to put my super into it!

To balance my portfolio I am asking if anyone has any other solid share tips they’d like to share please?

Obviously I’m going to do my own research which is why I’m asking now as it give me a few months to consider my options and investment strategy.

Rather than clog up this thread if you could message me with any advice/share tips it would be much appreciated.

Thanks in advance. :)
Liontown ( LTR )
Greatest " No Brainer " stock that ever existed on the ASX and perfect for a superfund.
Of course: DYOR, starting with previous company presentations.
Management is AAA+
 
  • Love
  • Like
  • Fire
Reactions: 7 users

Bravo

If ARM was an arm, BRN would be its biceps💪!
Is it related to memristor technology?

Another line of enquiry


Could the 3rd party license fee incurred be relating to Cisco for LSTM? Something that @Fact Finder said back in May was (or so I thought) hinting at a possibility of CIsco and BrainChip working on LSTM together #11,285

Apologies if I have misinterpreted.
 
Last edited:
  • Like
  • Thinking
  • Love
Reactions: 14 users

White Horse

Regular
  • Like
  • Fire
Reactions: 13 users
Could someone who knows stuff please have a look at this and confirm ours is better?

Here's my uneducated take..

First off, AKIDA is now freely available, so why are they comparing their latest iteration, to a GPU?

18 million neurons, well and truly trumps AKIDAs 1.2 million, but it is the number of connections between neurons, synapses, that makes something "smart".
AKIDA claims 10 billion (but I understand the ratio of neurons and synapses are on a sliding scale?).

They are using 16 bit (vs AKIDA 1 to 4) activations, to remember the previous step (something we won't achieve, until LSTM?).
So that's the only thing I can see, that they are doing better, but they would be using a lot more power, than when we have next gen.

As others have said.
No learning (how can something even be considered "intelligent", if it can't learn?)

Not available as IP = huge disadvantage in the OEM marketplace.

It's competition, but not good competition, in my opinion.
 
Last edited:
  • Like
  • Love
  • Fire
Reactions: 25 users

uiux

Regular
Here's my uneducated take..

First off, AKIDA is now freely available, so why are they comparing their latest iteration, to a GPU?

48 million neurons, well and truly trumps AKIDAs 1.2 million, but it is the number of connections between neurons, synapses, that makes something "smart".
AKIDA claims 10 billion (but I understand the ratio of neurons and synapses are on a sliding scale?).

They are using 16 bit (vs AKIDA 1 to 4) activations, to remember the previous step (something we won't achieve, until LSTM?).
So that's the only thing I can see, that they are doing better, but they would be using a lot more power, than when we have next gen.

As others have said.
No learning (how can something even be considered "intelligent", if it can't learn?)

Not available as IP = huge disadvantage in the OEM marketplace.

It's competition, but not good competition, in my opinion.

18 million neurons

@DingoBorat I literally did the exact same thing, misreading the neurons/params
 
Last edited:
  • Like
Reactions: 6 users
18 million neurons

@DingoBorat I literally did the exact same thing, misreading the neurons/params
When you look at that, 48 million parameters, doesn't sound like a lot, when you have 18 million neurons..

At a guess, I'd say they have a lot less connections, but both Brainchip's and GrAI Matter's figures are ambiguous..
 
  • Like
Reactions: 6 users

gex

Regular
I'm guessing there's no set standard for benchmarking as yet?
 
  • Like
Reactions: 5 users
I'm guessing there's no set standard for benchmarking as yet?
I think it's a bit like rolling different shaped fruit, down a board 😛
 
Last edited:
  • Like
  • Haha
Reactions: 5 users

uiux

Regular
So say brainchip and nasa are working together to get the best design for Akida 1500, they would then be paying socionext a fee ontop of this too produce the chip. I think that’s my understanding. Or maybe the fee was something to do with this as brought up before




If anything - this could be relevant - it looks like it actually offers an improvement over the SRAM:

"MRAM which offers 2.5X to 3.5X density advantage over SRAM, 1000X better endurance over FLASH, high radiation tolerance above 100Krad to 1Mrad and ultra-low power standby leakage which is critical for long battery life between solar recharge is a big advantage for these critical SPACE missions"
 
  • Like
  • Love
Reactions: 8 users
Top Bottom