Rise from the ashes
Regular
Neuromorphic Artificial Intelligence Systems
05/25/22 - Modern AI systems, based on von Neumann architecture and classical neural networks, have a number of fundamental limitations in co...
BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
Hi FF,
You are confusing me with someone who knows stuff.
My forte is knowing where to find stuff:
https://towardsdatascience.com/coun...-in-deep-learning-models-by-hand-8f1716241889
This article by Rami Karim explains that, if you need to compress the arrangement, counting the before and after layers of the NN can be useful.
That said, my serious engagement with software ended at the Fortran IV stage. Since then, I've only dabbled at the edge. You'd be better off applying to @uiux .
By the way, did I tell you about the natty program for characterizing the radiation pattern of a 12 element yagi antenna I wrote in Fortran that largely followed the yet to be created OOP* structure?
*OOP = singular of OOPS.
Last edited:
SERA2g
Founding Member
[40] Anup Vanarse, Adam Osseiran, Alexander Rassau, and Peter van der Made. A hardware-deployable neuromorphic solution for encoding and classification of electronic nose data. Sensors, 19(22), 2019. 1, 9
Boab
I wish I could paint like Vincent
Edit Page 9
5.1. Target Hardware StereoSpike has resolutely been developed in the philosophy of spiking neural networks. As a result, it is essentially implementable on dedicated neuromorphic hardware, such as Intel Loihi [5], IBM TrueNorth [2]. These chips can leverage the binarity and sparsity of spike tensors navigating through the network. In addition, we believe that our model being feedforward and requiring a reset on all of its neurons at each timestep is not a problem, because resetting membrane potentials is actually less costly than applying a leak. Therefore, statelessness can be seen as an advantage over recurrence in spiking models with similar performances. However, we are aware that current neuromorphic chips are initially designed for the implementation of stateful units, and acknowledge that we do not leverage this feature. Consequently, we believe that it rather fits to dedicated hardware for stateless models with sparse quantized activations. We therefore consider that Brainchip’s Akida chip [40] is a good fit. As it imposes weights to take at most 8 bit, we quantized StereoSpike’s weights us
Page 9, bottom right.
SERA2g
Founding Member
Fact Finder
Top 20
Where it says:
“5.1. Target Hardware
StereoSpike has resolutely been developed in the philos- ophy of spiking neural networks. As a result, it is essen- tially implementable on dedicated neuromorphic hardware, such as Intel Loihi [5], IBM TrueNorth [2]. These chips can leverage the binarity and sparsity of spike tensors nav- igating through the network. In addition, we believe that our model being feedforward and requiring a reset on all of its neurons at each timestep is not a problem, because resetting membrane potentials is actually less costly than applying a leak. Therefore, statelessness can be seen as an advantage over recurrence in spiking models with similar performances. However, we are aware that current neuro- morphic chips are initially designed for the implementation of stateful units, and acknowledge that we do not leverage this feature. Consequently, we believe that it rather fits to dedicated hardware for stateless models with sparse quan- tized activations. We therefore consider that Brainchip’s Akida chip [40] is a good fit. As it imposes weights to take at most 8 bit, we quantized StereoSpike using PyTorch natively available post-training static quantiza- tion. The process resulted in an even lighter model with 8 bit wide unsigned integer weights, for the price of a minor performance drop (i.e., MDE of 17.1 cm on indoorflying split 1). Presumably, quantization-aware training would do even better. This demonstrates the efficient deployability on such hardware. Finally, we would like to emphasize that our class of model with sparse binary activations and less constrained weights provides a good compromise between Spiking Neural Networks (SNNs) and Binary Neural Net- works (BNNs).”
alwaysgreen
Top 20
Hey @Diogenese Are you interested in doing the IT for my business as a side hustle?
Can pay you in BRN shares if interested haha
EDIT: I'm actually serious.
Last edited:
Fact Finder
Top 20
Sorry doubled up.
Rise from the ashes
Regular
But what do you do on a phone for keyword search when "find in page" is not available?
Fact Finder
Top 20
There are a couple of different approaches which I have used:
1. If near a body of water challenge yourself to see how far you can throw it;
2. Swear loudly then sob uncontrollably - swearing can be in any language seems to work equally well in this regard.
Hope these help.
Regards
FF
AKIDA BALLISTA
wilzy123
Founding Member
There should be some function to that effect available.
equanimous
Norse clairvoyant shapeshifter goddess
The ASIC executives will retire by May 2024This could be of interest to some.
Inquiry into ASIC's capacity and capability to respond to reports of alleged misconduct
On 27 October 2022, the Parliamentary Joint Committee on Corporations and Financial Services began an inquiry into ASIC’s capacity and capability to respond to reports of alleged misconduct. The committee will call for written submissions in due course. &nbwww.aph.gov.au
Inquiry into ASIC's capacity and capability to respond to reports of alleged misconduct
On 27 October 2022, the Parliamentary Joint Committee on Corporations and Financial Services began an inquiry into ASIC’s capacity and capability to respond to reports of alleged misconduct.
The committee will call for written submissions in due course.
The committee currently intends to table a report in both Houses of the Parliament by June 2024.
Committee Secretariat contact:
Committee Secretary
Parliamentary Joint Committee on Corporations and Financial Services
PO Box 6100
Parliament House
Canberra ACT 2600
Phone: +61 2 6277 3583
Fax: +61 2 6277 5719
corporations.joint@aph.gov.au
Rise from the ashes
Regular
Pretty sure most PDF files I open don't have that option. When searching a webpage it's always and option for me.
Attachments
wilzy123
Founding Member
Bummer. When I open it using 'Google Drive', I get the option. Can't vouch for other applications tho.
HopalongPetrovski
I'm Spartacus!
Just wondering when our next announcement may drop?
Will it be a slow and steady rise following some boring doorbell thingy or will it be another out of the blue, meteoric hellcat whammo surprise like we got in January, courtesy of MB.
Just what will we get first, hmmmm....some cool new AI glasses, a sexy new space Alexa or will some new medical testing devise nose out the boring old PCR and RAT tests ready for the next wave?
Apparently there's about 46000 of us all standing at ground zero, just waiting for the flash.
Bring it, Brainchip!
Will it be a slow and steady rise following some boring doorbell thingy or will it be another out of the blue, meteoric hellcat whammo surprise like we got in January, courtesy of MB.
Just what will we get first, hmmmm....some cool new AI glasses, a sexy new space Alexa or will some new medical testing devise nose out the boring old PCR and RAT tests ready for the next wave?
Apparently there's about 46000 of us all standing at ground zero, just waiting for the flash.
Bring it, Brainchip!
Fact Finder
Top 20
Bravo
If ARM was an arm, BRN would be its biceps💪!
Tesla autonomous tech investigated after two new fatal crashes last month
Tesla’s controversial advanced driver assistance tech is in the spotlight again after appearing in the US road safety authority's fatal crash data twice last month.Jordan Mulach
07:0422 November 2022
The peak road safety authority in the US is investigating two fatal crashes involving Tesla electric cars last month.
Since July 2021, the National Highway Safety Traffic Administration (NHTSA) has required car-makers with ‘Level Two’ advanced driver assistance systems to report crashes involving the advanced technology within 24 hours of the incident occurring – followed by a subsequent updated report up to nine days later.
'Level Two' advanced driver assistance systems (an industry ranking to indicate various stages of autonomous technology) include functions such as adaptive cruise control and lane-keeping assistance – however drivers are always legally required to remain in control. The tech is intended as a fail safe rather than something that should be relied upon.
As reported by Automotive News, two Tesla Model 3 electric cars were involved in separate crashes resulting in a fatality between September 16 and October 15 – although it is not yet clear whether the car-maker’s advanced driver assistance systems were to blame.
Across the 30-day period, NHTSA recorded 18 fatal crashes in the US involving cars fitted with advanced driver assistance systems – the two Tesla incidents accounted for more than 10 per cent of the overall data over the period.
While both Tesla crashes happened in California, Automotive News claims the NHTSA data has been redacted or classified as confidential.
The latest fatal crashes have added to the scrutiny of Tesla’s advanced driver assistance systems.
As previously reported, NHTSA data from July 2021 to June 2022 noted 392 crashes – involving 12 different car brands – involving vehicles equipped with advanced driver assistance systems. In that sample, Tesla vehicles accounted for 273 – or 70 per cent – of the incidents.
Of the six fatalities where advanced driver assistance systems were considered a factor, Tesla cars accounted for 83 per cent of such crashes.
Last month, overseas reports revealed the US Department of Justice had launched a criminal investigation into Tesla in 2021, following more than a dozen motor vehicle accidents involving the US auto giant’s cars.
The investigation alleges drivers have placed too much reliance on Tesla's driver assistance technology because the names of the systems exaggerate their true capabilities and ignore their limitations.
Tesla advertises its three products as ‘Autopilot’, ‘Enhanced Autopilot’, and ‘Full Self-Driving Capability’ – each providing various Level Two advanced driver assistance functions.
Autopilot is similar to a number of other advanced driver assistance systems offered by several car makers, providing adaptive cruise control and lane-keeping assistance.
Enhanced Autopilot adds functions such as automatic navigation, lane change and parking, while also allowing the car to be summoned by its owner.
While the only addition to Full Self-Driving is the ability to detect and stop for traffic signs and lights, it is the most controversial of Tesla’s advanced driver assistance systems.
Released in September 2021, Full Self-Driving – also known as ‘FSD’ – is a ‘beta’ service which is tested in real-time by the electric car giant’s customers on public roads, with Tesla gathering the data to improve the system.
In 2020, Tesla CEO Elon Musk claimed Full Self-Driving would be approved for US roads by the end of the year. The executive reiterated the comments in 2021 and earlier this year.
Last month, Mr Musk announced Tesla’s next level of driver assistance systems would not be permitted for use on US roads before the end of 2022, although the technology would be made available to the electric-car giant’s customers.
Tesla autonomous tech investigated after two new fatal crashes last month
Tesla’s controversial advanced driver assistance systems are in the spotlight again after appearing in the US road safety authority's fatal crash data twice last month. However, the safety group is yet to determine whether the technology caused the deaths.
Last edited:
Similar threads
