BRN Discussion Ongoing

[Diogenese]

Hi @Diogenese so does that mean we are not Neuromorphic Chip? As I dont have much knowledge in this technical field. As having invested much of my savings, this is second commentary on BRN ,Were we dupe so far :-( thanks

Hi Baisyet,

Not at all. For the purists, who insist on a spike, Akida still does 1-bit weights and activations for low power, high speed applications which are not too demanding on accuracy.

Possibly the other point of distinction for the purists who cut their teeth on analog, analog is much closer to a neuron in that it does sum voltages until they reach a threshold voltage value as does a neuron, whereas Akida counts the number of input bits until a threshold count is reached. On reflection, this is the more likely differentiator.

Many of the academic researchers have been toiling away at analog for decades and perhaps they take a proprietary interest in the term "neuromorphic".

The problem with analog is manufacturing variability, in that the dimensions cannot be strictly controlled so the electrical characteristics of the individual capacitors varies, which is significant when many thousands/millions of capacitors are involved ...

So it is odd as Jason said he did not want to emulate the neuron so much as to draw inspiration from it.

 

[Diogenese]

Hi Baisyet,

Not at all. For the purists, who insist on a spike, Akida still does 1-bit weights and activations for low power, high speed applications which are not too demanding on accuracy.

Possibly the other point of distinction for the purists who cut their teeth on analog, analog is much closer to a neuron in that it does sum voltages until they reach a threshold voltage value as does a neuron, whereas Akida counts the number of input bits until a threshold count is reached. On reflection, this is the more likely differentiator.

Many of the academic researchers have been toiling away at analog for decades and perhaps they take a proprietary interest in the term "neuromorphic".

The problem with analog is manufacturing variability, in that the dimensions cannot be strictly controlled so the electrical characteristics of the individual capacitors varies, which is significant when many thousands/millions of capacitors are involved ...

So it is odd as Jason said he did not want to emulate the neuron so much as to draw inspiration from it.

So really, Akida is the result of thinking outside the box.

 

[Diogenese]

Hi Baisyet,

Not at all. For the purists, who insist on a spike, Akida still does 1-bit weights and activations for low power, high speed applications which are not too demanding on accuracy.

Possibly the other point of distinction for the purists who cut their teeth on analog, analog is much closer to a neuron in that it does sum voltages until they reach a threshold voltage value as does a neuron, whereas Akida counts the number of input bits until a threshold count is reached. On reflection, this is the more likely differentiator.

Many of the academic researchers have been toiling away at analog for decades and perhaps they take a proprietary interest in the term "neuromorphic".

The problem with analog is manufacturing variability, in that the dimensions cannot be strictly controlled so the electrical characteristics of the individual capacitors varies, which is significant when many thousands/millions of capacitors are involved ...

So it is odd as Jason said he did not want to emulate the neuron so much as to draw inspiration from it.

Digital Spiking Neural Network (DSNN)? ... or is that already taken?

 

[TECH]

When Dave Whatshisname deprecated Akida on linkedin, Tony Lewis replied that he tought he had made it clear that it was "event based".

So maybe the distinction is between the 1-bit Akida (still an available mode) and the 2 and 4-bit mode which are more accurate/definitive (able to make finer distinctions).

I haven't listened to the podcast with Jason yet, but we all know we are Neuromorphic, so possibly his comments have been misunderstood..as he is currently contracted as a contributing scientist on our SAB with Peter and Andre, less than 3 months ago, so I would suggest that Peter would speak with him if he feels his comment/s were in anyway inappropriate or not aligned with our companys objectives.
Integrity resides at Brainchip..we are 100% Neuromorphic and always have been.
Very happy to be corrected by the man himself...no regerts
Tech x

Let’s Go Brainchip time to bust open that dam.

 

[Terroni2105]

When Dave Whatshisname deprecated Akida on linkedin, Tony Lewis replied that he tought he had made it clear that it was "event based".

So maybe the distinction is between the 1-bit Akida (still an available mode) and the 2 and 4-bit mode which are more accurate/definitive (able to make finer distinctions).

If they are now confused on their own identity then I don’t think it is a good look after all these years of telling us they are leaders in neuromorphic.
I have looked at news releases from Brainchip and they still refer to their ”neuromorphic AI” and still refer to being “the worlds first commercial producer of neuromorphic AI” on their website, so I’m either not understanding things or I kind of feel Jason K. Eshraghian who is on our Scientific Advisory Board is undermining BrainChip. Maybe I just don’t understand the technicality’s of whatever Jason is saying but if he is correct then I feel like it is a massive miscommunication on the part of BrainChip all these years because there shouldn’t be confusion when trying to market your product.