BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
McHale
Regular
Great post @Quiltman, and you've highlighted some really important points "What is a neuromorphic chip anyway" revealed how ignorant many analysts and commentators were about a technology that is now getting significant tech sector acceptance and rapidly growing publicity.
What most of us didn't understand 5 years ago was that it would take more than 5 years for a new technology to work through the cycle of assessment, design, engineering, procurement, product manufacture and then marketing; before you arrive at a place where the tech in this case Akida, is able to generate steady ongoing revenues/income. It has been a long haul but it looks like we're finally at the edge of seeing something really beginning to move at BRN.
Last year CES delivered some excellent news, and revealed the growth and reach of the "eco-system" Sean and the BRN team have been building, I anticipate there will be more at CES this year, and I don't think that is an overly ambitious position to take on the basis of what has developed since CES 2024.
I think the run up in SP in late Jan early Feb was catalyzed by the outcomes at CES, this was off a low of 15c on 24Jan to a high of 53.5c on 23Feb. The current breakout is off the back of the 2 recent anns. and the fact that Sean has delivered on target of making good anns before the end of 24.
From the perspective of charts, the shape of the current breakout, and also the price action of BRN prior to this breakout is far more conducive to a continued uptrend in SP than the chart of Jan 2023. So going into 2025 my feeling for BRN is bullish on both a fundamental level and a technical analysis level.
I want to thank all of those here who contribute so much great research, I really appreciate what has been shared here over 2024, and wish genuine contributors here a wonderful and prosperous 2025, and don't forget the year of the dragon aint over yet. More to come soon, have to say I am really pleased with the way Sean has delivered on what he said he wanted to, great work Sean and team. Go BRN.
Pom down under
Top 20
Can anyone access the link.
Attachments
Tony puts up with a lot of crap so thick skin is a must.Happy New Year Everyone from AKIDA
Well, the Podcast was beautifully timed in my opinion, including a nice new graphic which looks great, love how Peters
spiking neuron logo is front and centre.
Sean delivered, his demeanour remains unchanged, with the 4th quarter now done and dusted, we should expect to see
a healthier 4C which will be delivered towards the end of January 2025.
Now, having met Tony not long after he started with Peter in St. Georges Terrace, his tone has always been the same, calm,
not over excitable, his delivery in all the Podcasts has been very consistent, calm and clear questioning, polite even, considering
he is speaking with his boss as such.
I have thought about what one person posted with regards Tony and the recent Podcast, I have decided to post this private email,
which I'm pretty confident won't upset Tony......it sums things up beautifully, so lets please all move on.
Hi Chris,
As for anonymous people making “nasty comments” on internet forums, I’m pretty much immune to that stuff after 3 1/2 years in this job.
I can’t prevent people saying nasty, petty, ignorant, stupid things, all I can do is control how I react to those comments.
I ignore them.
I appreciate your efforts to support me.
Regards
Tony
You would want a decent salary to put up with it though.
CHIPS
Regular
No, unfortunately not. I am in Germany.
Can you please post the whole? Thanks and happy New Year!
Hi McH,
Stop me if you've heard this before -
In my opinion, the switch from Akida 1 to IP set back the adoption of our tech by a couple of years. Akida 1 had just been produced as a SoC when it was, shall we say, put on the back burner (although I have used the bathwater analogy before) in favour of IP licensing. As we have subsequently seen, there are several uses of Akida 1, but apparently the technology was moving so fast that customers wanted more functionality to transfer functionality from the processor (CPU) to the Akida silicon.
We also know that our shoestring of cash was severly frayed and some drastic measures were required to keep BRN afloat. On top of that, management knew that Akida 2 was in the pipeline and, to keep faith with the EAPs, BRN needed to advise them of Akida 2 TENNs - a classic dilemma - juggling razors on a tightrope. This also stalled some "nearly there" Akida 1 contracts.
Obviously (in retrospect), the company needed to maintain or strengthen its technological advantage in a rapidly evolving tech.
But, as we see with Frontgrade for example, Akida 1 is the ant's pants for some very high tech applications.
In any event, BRN has now adopted a much faster route to market with the addition of the algorithm/software product line. Our friends at Edge Impuse play an integral role in rapidly developing new models, essential for any NN application. This greatly shortens the time to market, whether we are licensing digital SNN simulation software, or licensing digital SNN IP.
So, if BRN had put all its behind the sofa cushion cash into producing more Akida 1 chips, we may not have ever got to Akida 2 TENNs.
So what happened to the taping-out of Akida 2 which Anil announced several months ago, before that was quashed by an announcement that a 3rd party was doing it?
We have a few friends who have done tape-outs: Socionext, MegaChips together with that FDSoI mob whose name eludes me for the moment ...
We also have friends who are capable of Tape-outs: ARM, Intel, ...
Also, I have always found Tony easy to deal with and prompt to reply.Happy New Year Everyone from AKIDA
Well, the Podcast was beautifully timed in my opinion, including a nice new graphic which looks great, love how Peters
spiking neuron logo is front and centre.
Sean delivered, his demeanour remains unchanged, with the 4th quarter now done and dusted, we should expect to see
a healthier 4C which will be delivered towards the end of January 2025.
Now, having met Tony not long after he started with Peter in St. Georges Terrace, his tone has always been the same, calm,
not over excitable, his delivery in all the Podcasts has been very consistent, calm and clear questioning, polite even, considering
he is speaking with his boss as such.
I have thought about what one person posted with regards Tony and the recent Podcast, I have decided to post this private email,
which I'm pretty confident won't upset Tony......it sums things up beautifully, so lets please all move on.
Hi Chris,
As for anonymous people making “nasty comments” on internet forums, I’m pretty much immune to that stuff after 3 1/2 years in this job.
I can’t prevent people saying nasty, petty, ignorant, stupid things, all I can do is control how I react to those comments.
I ignore them.
I appreciate your efforts to support me.
Regards
Tony
I have wondered at times whether BRN like most organisations have 'ignore' provisions when it comes to dealing with abusive morons.
Fullmoonfever
Top 20
Was trying to find any connection with this guys post but seems he is on a sabbatical and either worked for or still does, for this place:
sentrana.com
Just a blog post on neuromorphic but references / resources are just BRN, IBM amd Intel.
www.linkedin.com
Luis Mirantes
Neuromorphic computing represents a new paradigm in the design of computer systems, drawing inspiration from the structure and functionality of the human brain. Unlike traditional computers that process data sequentially using binary logic, neuromorphic systems replicate the brain's complex neural networks, allowing them to process information more efficiently, especially for tasks requiring real-time learning, sensory data processing, and pattern recognition.
In traditional von Neumann architecture, there’s a strict separation between memory and processing units. Information is processed in a step-by-step sequence, requiring multiple passes between the CPU and memory. This not only increases latency but also consumes significant energy. Neuromorphic computing, on the other hand, decentralizes processing and memory, allowing for highly efficient and parallelized information flow, much like how the brain manages complex tasks like vision, sound processing, and motor control.
Spiking Neural Networks (SNNs): Unlike traditional artificial neural networks (ANNs), where data flows continuously, SNNs send discrete “spikes” (inputs) of information between neurons. These spikes are analogous to the electrical impulses in the brain and allow for more biologically accurate modeling of neural processes.
Memristors: One of the key elements of neuromorphic hardware, memristors, are electrical components that can both store and process information. Memristors can "remember" the amount of charge that has passed through them, making them ideal for simulating synaptic behavior.
Event-Driven Processing: Neuromorphic systems often operate based on discrete events (inputs), rather than continuous data streams. In these systems, neurons only fire (execute a weighted function) when certain thresholds are reached, significantly reducing power consumption compared to the constant data flow in traditional computers.
Energy Efficiency: Neuromorphic systems are designed to use minimal energy, inspired by the human brain, which uses just 20 watts of power to perform highly complex tasks. In contrast, traditional supercomputers can require megawatts of power to perform similar operations.
Real-Time Learning and Adaptation: Neuromorphic chips can learn and adapt on the fly, thanks to their ability to strengthen or weaken connections between the neural network nodes, similar to how the brain learns through reinforcement. This is particularly useful for applications like robotics, autonomous systems, and intelligent assistants that need to make decisions in dynamic environments.
Superior Pattern Recognition: Neuromorphic systems excel in pattern recognition tasks, such as image and speech recognition. Since they can process sensory data in real time and adapt based on experience, they are highly effective in applications involving complex and unstructured data.
Parallel Processing: In neuromorphic computing, the decentralized architecture allows multiple processes to occur simultaneously. This leads to faster and more efficient processing for tasks that require the simultaneous integration of large amounts of data.
Autonomous Vehicles: Neuromorphic chips could greatly enhance the processing capabilities of self-driving cars. These vehicles rely on real-time data from cameras, radar, and other sensors to navigate complex environments. Neuromorphic systems can process this data with high speed and accuracy while using less power, making them ideal for autonomous vehicles that need to make split-second decisions.
Robotics: In robotics, neuromorphic computing can enable more advanced and adaptive behaviors. Robots could process sensory information and adapt their actions in real time, allowing for more seamless interaction with their environment and better problem-solving capabilities.
Healthcare and Neuroscience: In healthcare, neuromorphic systems can assist in brain-machine interfaces (BMIs), where artificial systems interact with the human nervous system. These systems could be used to create advanced prosthetics, restore lost sensory functions, or even develop more effective treatments for neurological disorders like epilepsy.
Edge Computing: Neuromorphic chips are perfect for edge computing environments, where power and computational resources are limited. For instance, in IoT devices or smart sensors that require real-time processing of data, neuromorphic systems offer a low-power, high-efficiency solution.
That said, ongoing research in neuromorphic computing is yielding exciting breakthroughs. Companies like Intel (Loihi), IBM (TrueNorth), and BrainChip are leading the development of neuromorphic processors, and the field is rapidly evolving.
In the future, neuromorphic systems could revolutionize artificial intelligence, robotics, and numerous other fields by providing energy-efficient, adaptive, and highly parallel computing systems. By mimicking the brain’s structure and functionality, neuromorphic computing is poised to push the boundaries of what computers can do, bringing us closer to building machines that think and learn like humans.
Neuromorphic computing offers a fundamentally different approach to computing, moving beyond the limitations of traditional architectures to create systems capable of real-time learning, efficient energy use, and unparalleled adaptability. By drawing inspiration from the brain, this emerging field holds the potential to revolutionize industries from healthcare to autonomous systems, offering a glimpse into the future of computing that is more intelligent, efficient, and capable than ever before.
Here are some key references you can explore for further reading on neuromorphic computing:
These resources provide a mix of foundational knowledge and current advancements in the field of neuromorphic computing.
Home
Breakthroughs Go fast with DeepCortex How fast can you turn your data into #rev_slider_25_1_wrapper rs-loader.spinner2{ background-color...
sentrana.com
Just a blog post on neuromorphic but references / resources are just BRN, IBM amd Intel.
Luis Mirantes - Sabbatical | LinkedIn
Experience: Sabbatical · Education: The Johns Hopkins University · Location: Chevy Chase · 500+ connections on LinkedIn. View Luis Mirantes’ profile on LinkedIn, a professional community of 1 billion members.
Neuromorphic Computing: A Revolution Inspired by the Human Brain
Luis Mirantes
Luis Mirantes
VP of Engineering | Full-Stack Software Platforms | ML/AI Lifecycle Engineering
Published Sep 7, 2024Neuromorphic computing represents a new paradigm in the design of computer systems, drawing inspiration from the structure and functionality of the human brain. Unlike traditional computers that process data sequentially using binary logic, neuromorphic systems replicate the brain's complex neural networks, allowing them to process information more efficiently, especially for tasks requiring real-time learning, sensory data processing, and pattern recognition.
What is Neuromorphic Computing?
Neuromorphic computing refers to computer architectures that simulate the brain’s neural structure. It leverages the way neurons and synapses in biological systems transmit and process information, using networks of artificial neurons to execute tasks. These systems are capable of performing computations in a highly parallel and distributed manner, much like the brain.In traditional von Neumann architecture, there’s a strict separation between memory and processing units. Information is processed in a step-by-step sequence, requiring multiple passes between the CPU and memory. This not only increases latency but also consumes significant energy. Neuromorphic computing, on the other hand, decentralizes processing and memory, allowing for highly efficient and parallelized information flow, much like how the brain manages complex tasks like vision, sound processing, and motor control.
The Building Blocks of Neuromorphic Systems
Neuromorphic chips contain artificial ‘neurons’ that simulate the way biological neurons process information. These neurons communicate with each other through artificial ‘synapses’, which strengthen or weaken (weighted) connections based on experience—mirroring the way the human brain learns. Some of the core components of these architecture include:Spiking Neural Networks (SNNs): Unlike traditional artificial neural networks (ANNs), where data flows continuously, SNNs send discrete “spikes” (inputs) of information between neurons. These spikes are analogous to the electrical impulses in the brain and allow for more biologically accurate modeling of neural processes.
Memristors: One of the key elements of neuromorphic hardware, memristors, are electrical components that can both store and process information. Memristors can "remember" the amount of charge that has passed through them, making them ideal for simulating synaptic behavior.
Event-Driven Processing: Neuromorphic systems often operate based on discrete events (inputs), rather than continuous data streams. In these systems, neurons only fire (execute a weighted function) when certain thresholds are reached, significantly reducing power consumption compared to the constant data flow in traditional computers.
Advantages of Neuromorphic Computing
Neuromorphic computing offers several advantages over conventional computing, particularly in areas that demand high energy efficiency and fast processing of unstructured data.Energy Efficiency: Neuromorphic systems are designed to use minimal energy, inspired by the human brain, which uses just 20 watts of power to perform highly complex tasks. In contrast, traditional supercomputers can require megawatts of power to perform similar operations.
Real-Time Learning and Adaptation: Neuromorphic chips can learn and adapt on the fly, thanks to their ability to strengthen or weaken connections between the neural network nodes, similar to how the brain learns through reinforcement. This is particularly useful for applications like robotics, autonomous systems, and intelligent assistants that need to make decisions in dynamic environments.
Superior Pattern Recognition: Neuromorphic systems excel in pattern recognition tasks, such as image and speech recognition. Since they can process sensory data in real time and adapt based on experience, they are highly effective in applications involving complex and unstructured data.
Parallel Processing: In neuromorphic computing, the decentralized architecture allows multiple processes to occur simultaneously. This leads to faster and more efficient processing for tasks that require the simultaneous integration of large amounts of data.
Applications of Neuromorphic Computing
Neuromorphic computing has broad potential in a variety of fields, particularly in applications where traditional computers struggle to meet real-time processing requirements.Autonomous Vehicles: Neuromorphic chips could greatly enhance the processing capabilities of self-driving cars. These vehicles rely on real-time data from cameras, radar, and other sensors to navigate complex environments. Neuromorphic systems can process this data with high speed and accuracy while using less power, making them ideal for autonomous vehicles that need to make split-second decisions.
Robotics: In robotics, neuromorphic computing can enable more advanced and adaptive behaviors. Robots could process sensory information and adapt their actions in real time, allowing for more seamless interaction with their environment and better problem-solving capabilities.
Healthcare and Neuroscience: In healthcare, neuromorphic systems can assist in brain-machine interfaces (BMIs), where artificial systems interact with the human nervous system. These systems could be used to create advanced prosthetics, restore lost sensory functions, or even develop more effective treatments for neurological disorders like epilepsy.
Edge Computing: Neuromorphic chips are perfect for edge computing environments, where power and computational resources are limited. For instance, in IoT devices or smart sensors that require real-time processing of data, neuromorphic systems offer a low-power, high-efficiency solution.
Challenges and Future Prospects
Despite its promise, neuromorphic computing faces several challenges. The development of neuromorphic chips and systems is still in its early stages, and creating hardware that can truly mimic the brain’s complexity is a massive engineering challenge. Additionally, building efficient learning algorithms for these systems that can match the brain’s versatility is a work in progress.That said, ongoing research in neuromorphic computing is yielding exciting breakthroughs. Companies like Intel (Loihi), IBM (TrueNorth), and BrainChip are leading the development of neuromorphic processors, and the field is rapidly evolving.
In the future, neuromorphic systems could revolutionize artificial intelligence, robotics, and numerous other fields by providing energy-efficient, adaptive, and highly parallel computing systems. By mimicking the brain’s structure and functionality, neuromorphic computing is poised to push the boundaries of what computers can do, bringing us closer to building machines that think and learn like humans.
Neuromorphic computing offers a fundamentally different approach to computing, moving beyond the limitations of traditional architectures to create systems capable of real-time learning, efficient energy use, and unparalleled adaptability. By drawing inspiration from the brain, this emerging field holds the potential to revolutionize industries from healthcare to autonomous systems, offering a glimpse into the future of computing that is more intelligent, efficient, and capable than ever before.
Here are some key references you can explore for further reading on neuromorphic computing:
- Intel Labs - Loihi: Neuromorphic Computing for AI Intel’s Loihi chip represents a significant advancement in neuromorphic computing, emphasizing event-based computing and spiking neural networks. Intel Labs on Loihi.
- IBM Research - TrueNorth: Neuromorphic System IBM's TrueNorth chip is another major development, designed to replicate the brain’s architecture with millions of artificial neurons. IBM's TrueNorth Chip.
- Nature - Neuromorphic Computing and Its Impact on AI This article outlines the scientific and technical aspects of neuromorphic systems and their applications. Nature Article on Neuromorphic Computing.
- IEEE Spectrum - How Neuromorphic Computing Will Shape AI This article delves into the technological impact of neuromorphic computing on artificial intelligence and robotics. IEEE Spectrum on Neuromorphic Computing.
- BrainChip - Neuromorphic Processor Technology BrainChip’s Akida processor represents the application of neuromorphic principles in edge computing and AI. BrainChip's Neuromorphic Technology.
These resources provide a mix of foundational knowledge and current advancements in the field of neuromorphic computing.
McHale
Regular
Hi Barrel Sitting Diogenous one, I don't think that I could stop you - (but what do you keep in that barrel anyway, and does it have any relationship with all those strange symbols you're in a habit of posting here), even though I believe I have heard this line of reasoning from you before, or a similar version thereof, or at least part of it. Which, I might add, all sounds quite - to very reasonable.
It seems that Akida 1 does strike back, Edge Impulse just keep on coming up, but like you and a number of others I would love to know more about the tapeout of Akida 2 who will it be ?
But whatever happened with SocioNext ? sorry I can't think of who fabricated Akida 1.5 on fullydepletedsilicononinsulator either.
I didn't quite say earlier, how much, that I really like the look of the BRN chart taking shape since late September. I can't think of a time it ever looked better.
Yes - it's my Mortein answer.
I'm guessing that Socionext is tied in with TSMC, and, while there's nothing wrong with TSMC per se, the political sutiation and the US Chips Bill make it expedient for BRN to have a US-based backup plan.
One consequence of dropping TSMC is that Akida 2 TENNs tapeout will not be as susceptible to adverse actions, although I assume TSMC has all the tapeouts in a burn box.
Global Foundries has the FDSoI (the 1500 pie in which MegaChips claimed to have a finger). I would think BRN would do well not to spread the tapeout trade secrets around too liberally. every time there is a redesign, there are performance improvements (efficiency), quite apart from technical advances such as TENNs.
Last edited:
Fullmoonfever
Top 20
Appears this young lady has been working with Akida at GMAC for the last several months. Now interning at Amex.
Which was the small partner that wound up?...for some reason I thought it was GMAC but maybe not?
• Implemented software for IoT products, including cameras for license plate recognition and drive-through speech recognition
• Utilizing the Akida Neuromorphic ML Framework to train highly accurate and power-efficient algorithms on Edge AI platforms within IoT devices
• Migrating digit classification and audio detection models from Convolutional Neural Networks (CNN) to Spiking Neural Networks (SNN) using TensorFlow Keras API, incorporating quantization and knowledge distillation techniques
Which was the small partner that wound up?...for some reason I thought it was GMAC but maybe not?
Akshika Chawade
Computer Science Student at Cornell University
American Express Cornell University
Braintree, Massachusetts, United States
Machine Learning Intern
GMAC Intelligence
Apr 2024 - Oct 2024 7 months• Implemented software for IoT products, including cameras for license plate recognition and drive-through speech recognition
• Utilizing the Akida Neuromorphic ML Framework to train highly accurate and power-efficient algorithms on Edge AI platforms within IoT devices
• Migrating digit classification and audio detection models from Convolutional Neural Networks (CNN) to Spiking Neural Networks (SNN) using TensorFlow Keras API, incorporating quantization and knowledge distillation techniques
DingoBorat
Slim
Is this the first time, we've had external media report on an investor podcast?
Media Alert: BrainChip CEO Addresses Shareholders’ Top Questions on Investor Podcast
LAGUNA HILLS, Calif., December 31, 2024--BrainChip CEO Addresses Shareholders’ Top Questions on Investor Podcastfinance.yahoo.com
...
Fullmoonfever
Top 20
Something I personally haven't seen before. It appears to be from a 2022 IEEE conference.
Maybe others are aware of it though?
You might recognise a couple of the authors (but are all from TCS Research India) and a couple of the references used that I pasted below
Is a purchase only one I can find so far, so abstract only, but curious as to whether it has been continued in development.
Cite This
Arun M George; Andrew Gigie; A Anil Kumar; Sounak Dey; Arpan Pal; K Aditi
Brainchip unveils the akidatm development environment, 2019, [online] Available: https://www.brainchipinc.com/news-media/press-releases/ detail/61/brainchip-unveils-the-akida-development-environment.
S. Dey, A. Mukherjee, G. Bzard and D. McLelland, "Human gesture recognition using spiking input on akida neuromorphic platform", Neural Information Processing Systems (NeurIPS), 2019.
Maybe others are aware of it though?
You might recognise a couple of the authors (but are all from TCS Research India) and a couple of the references used that I pasted below
Is a purchase only one I can find so far, so abstract only, but curious as to whether it has been continued in development.
EchoWrite-SNN: Acoustic Based Air-Written Shape Recognition Using Spiking Neural Networks
Publisher: IEEECite This
Arun M George; Andrew Gigie; A Anil Kumar; Sounak Dey; Arpan Pal; K Aditi
Abstract:
In this paper, we propose EchoWrite-SNN, a robust edge compatible air-writing recognition system (used in applications such as AR/VR, HRI etc.) based on principles of SONAR and neuromorphic computing. The bare finger movements in air are captured by a pair of commonly available speaker-microphone pair. A new tracking algorithm based on windowed difference cross-correlation and ESPRIT is employed which shows better tracking accuracy compared to state-of-the-art methods with a median tracking error of only 3.31mm. To classify these air-written shapes, a 5-layer CNN is trained and then converted to a Spiking Neural Network (SNN) using ANN-to-SNN conversion technique to reap the benefits of low power neuromorphic computing on edge. Experimental results show that the converted SNN achieves 92% accuracy (a mere 3% less than the CNN) while showing 4.4 × reduction in number of operations compared to CNN resulting in further energy benefit when run on actual neuromorphic computation platforms.Brainchip unveils the akidatm development environment, 2019, [online] Available: https://www.brainchipinc.com/news-media/press-releases/ detail/61/brainchip-unveils-the-akida-development-environment.
S. Dey, A. Mukherjee, G. Bzard and D. McLelland, "Human gesture recognition using spiking input on akida neuromorphic platform", Neural Information Processing Systems (NeurIPS), 2019.
Hi FMF,Something I personally haven't seen before. It appears to be from a 2022 IEEE conference.
Maybe others are aware of it though?
You might recognise a couple of the authors (but are all from TCS Research India) and a couple of the references used that I pasted below
Is a purchase only one I can find so far, so abstract only, but curious as to whether it has been continued in development.
EchoWrite-SNN: Acoustic Based Air-Written Shape Recognition Using Spiking Neural Networks
Publisher: IEEE
Cite This
Arun M George; Andrew Gigie; A Anil Kumar; Sounak Dey; Arpan Pal; K Aditi
Abstract:
In this paper, we propose EchoWrite-SNN, a robust edge compatible air-writing recognition system (used in applications such as AR/VR, HRI etc.) based on principles of SONAR and neuromorphic computing. The bare finger movements in air are captured by a pair of commonly available speaker-microphone pair. A new tracking algorithm based on windowed difference cross-correlation and ESPRIT is employed which shows better tracking accuracy compared to state-of-the-art methods with a median tracking error of only 3.31mm. To classify these air-written shapes, a 5-layer CNN is trained and then converted to a Spiking Neural Network (SNN) using ANN-to-SNN conversion technique to reap the benefits of low power neuromorphic computing on edge. Experimental results show that the converted SNN achieves 92% accuracy (a mere 3% less than the CNN) while showing 4.4 × reduction in number of operations compared to CNN resulting in further energy benefit when run on actual neuromorphic computation platforms.
Brainchip unveils the akidatm development environment, 2019, [online] Available: https://www.brainchipinc.com/news-media/press-releases/ detail/61/brainchip-unveils-the-akida-development-environment.
S. Dey, A. Mukherjee, G. Bzard and D. McLelland, "Human gesture recognition using spiking input on akida neuromorphic platform", Neural Information Processing Systems (NeurIPS), 2019.
I have a vague recollection of seeing this (or a similar) acoustic gesture tracking method before, but I can't put my finger on it.
Fullmoonfever
Top 20
Maybe wave your finger in the air and it might come to you
Fullmoonfever
Top 20
That's the presso fwiw. We get our small mention Pg 5Something I personally haven't seen before. It appears to be from a 2022 IEEE conference.
Maybe others are aware of it though?
You might recognise a couple of the authors (but are all from TCS Research India) and a couple of the references used that I pasted below
Is a purchase only one I can find so far, so abstract only, but curious as to whether it has been continued in development.
EchoWrite-SNN: Acoustic Based Air-Written Shape Recognition Using Spiking Neural Networks
Publisher: IEEE
Cite This
Arun M George; Andrew Gigie; A Anil Kumar; Sounak Dey; Arpan Pal; K Aditi
Abstract:
In this paper, we propose EchoWrite-SNN, a robust edge compatible air-writing recognition system (used in applications such as AR/VR, HRI etc.) based on principles of SONAR and neuromorphic computing. The bare finger movements in air are captured by a pair of commonly available speaker-microphone pair. A new tracking algorithm based on windowed difference cross-correlation and ESPRIT is employed which shows better tracking accuracy compared to state-of-the-art methods with a median tracking error of only 3.31mm. To classify these air-written shapes, a 5-layer CNN is trained and then converted to a Spiking Neural Network (SNN) using ANN-to-SNN conversion technique to reap the benefits of low power neuromorphic computing on edge. Experimental results show that the converted SNN achieves 92% accuracy (a mere 3% less than the CNN) while showing 4.4 × reduction in number of operations compared to CNN resulting in further energy benefit when run on actual neuromorphic computation platforms.
Brainchip unveils the akidatm development environment, 2019, [online] Available: https://www.brainchipinc.com/news-media/press-releases/ detail/61/brainchip-unveils-the-akida-development-environment.
S. Dey, A. Mukherjee, G. Bzard and D. McLelland, "Human gesture recognition using spiking input on akida neuromorphic platform", Neural Information Processing Systems (NeurIPS), 2019.
Dallas
Regular
There’re two algorithms that scale indefinitely: learning and search. | Jim Fan
There’re two algorithms that scale indefinitely: learning and search. May you learn well and find your own green tree in 2025! | 73 comments on LinkedIn
Similar threads
