BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
Meanwhile, Gen AI gets a kick along, but steps back from the edge:
https://asia.nikkei.com/Business/Te...num=6&si=44daa0b3-d70e-4506-aab2-d3b8890a93a5
...
Forming the core of the pact is the mutual use of foundational models powering generative AI.
Generative AI is beginning to be used in research settings for recording minutes and producing documents. By machine learning from experimental data, research papers and similar sources, AI may be able to come up with scientific hypotheses and experimental designs. That would help to significantly improve efficiency in research that will lead to new discoveries.
Riken, the research institute under the ministry, started developing generative AI to speed scientific research last fiscal year. In the U.S., the DOE's Argonne National Laboratory is engaged in the same activity.
For the U.S. foundational model, the goal is to incorporate over 1 trillion parameters. Riken plans to reach around 100 billion parameters initially for its model, then expand the scale further on.
Who picks up the tab for electricity?
At least their security is in good hands:
https://brainchip.com/brainchip-and-quantum-ventura-partner-to-develop-cyber-threat-detection/
https://asia.nikkei.com/Business/Te...num=6&si=44daa0b3-d70e-4506-aab2-d3b8890a93a5
Japan, U.S. team up on generative AI to modernize research
TOKYO -- The Japanese and U.S. governments will form a cooperative framework that will make use of generative artificial intelligence to take scientific research to the next level....
Forming the core of the pact is the mutual use of foundational models powering generative AI.
Generative AI is beginning to be used in research settings for recording minutes and producing documents. By machine learning from experimental data, research papers and similar sources, AI may be able to come up with scientific hypotheses and experimental designs. That would help to significantly improve efficiency in research that will lead to new discoveries.
Riken, the research institute under the ministry, started developing generative AI to speed scientific research last fiscal year. In the U.S., the DOE's Argonne National Laboratory is engaged in the same activity.
For the U.S. foundational model, the goal is to incorporate over 1 trillion parameters. Riken plans to reach around 100 billion parameters initially for its model, then expand the scale further on.
Who picks up the tab for electricity?
At least their security is in good hands:
https://brainchip.com/brainchip-and-quantum-ventura-partner-to-develop-cyber-threat-detection/
To make this usable at the edge, the model would need to be broken up into sub-modules covering different topics.
An example of a system which does this is the International Patent Classification system (IPC):
https://ipcpub.wipo.int/?notion=sch...&initial=A&cwid=none&tree=no&searchmode=smart
A HUMAN NECESSITIES
B PERFORMING OPERATIONS; TRANSPORTING
C CHEMISTRY; METALLURGY
D TEXTILES; PAPER
E FIXED CONSTRUCTIONS
F MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
G PHYSICS
H ELECTRICITY
The IPC has several levels of sub-categories to facilitate targeted searching.
DingoBorat
Slim
So sounds like they haven't incorporated AKIDA, into the cupcake server yet, as was previously thought?
Food4 thought
Regular
Did he say watch the financials go to shit down downI also remember Sean said “watch the financials” and not “watch panicking anonymous investors in a forum to understand the market”or should I? Or did he say “what’s the financials?” ? You decide! 🫵🫵🫵
Or did he say watch them go up up ????
What’s happened so far regarding them ….
Time will tell
Stable Genius
Regular
Hi all,
Another Rob T like dot join.
This company has tiny ml technology with a NN.
It wasn’t clear if it’s their NN or if they’re using someone else’s IP?
My voice A.I.
Another Rob T like dot join.
This company has tiny ml technology with a NN.
It wasn’t clear if it’s their NN or if they’re using someone else’s IP?
My voice A.I.
MY VOICE AI | My Voice
My Voice AI’s first product, NanoVoiceTM uses tinyML to verify speakers in real time, even on ultra-low power edge AI platforms. Our technology is patented, with our world class speech scientists developing the next generation of voice AI innovation, beyond identity.
www.myvoice.ai
Can we spam this guy as he is talking about Analog computers and says it uses less power.
https://youtu.be/6Y6FJVqzivc?si=3ybF6QbVUO0XR6Mt
https://youtu.be/6Y6FJVqzivc?si=3ybF6QbVUO0XR6Mt
Hi SG,
My Voice does not use Akida in their patents, but that's not to say that Akida would not make a big improvement:
US2023186896A1 SPEAKER VERIFICATION METHOD USING NEURAL NETWORK 20211215
[0115] The cloud network device 430 stores a neural network model 434 , and the device 420 stores a neural network model 424 . The neural network model 434 and the neural network model 424 are the same, except that the neural network model 434 on the cloud device includes a second fully-connected layer after the first fully-connected layer and a softmax function applied to the output of the second fully-connected layer. The neural network 434 on the cloud device is trained using training data 432 . An exemplary method of training a neural network is by using backpropagation, and this is discussed further below. Once trained, the learned weights 438 are extracted from the trained neural network 436 . The weights corresponding to the second fully-connected layer, and the softmax function are discarded at this stage.
[0105] With reference to the steps 400 shown in FIG. 4 , below we discuss a system for training a neural network according to the invention. The system comprises a cloud network device and a device. As shown by step 410 , the network device is configured to train a neural network, such as the neural network shown in FIG. 2 . Methods of training a neural network are known. For example, the neural network may be trained by backpropagation. Other methods could be used, as long as the value of weights connecting the nodes of the neural network are learned, as shown by step 420 . Referring to FIG. 2 , the second connected layer 280 and the softmax function 290 may be used during training the neural network. However, the weights related to these layers are not useful for generating a vocal signature. They are therefore discarded once training is complete. At step 440 , the weights are then sent to the device that performs the methods of generating a vocal signature and performing speaker verification.
[0110] Once the device receives the learned weights, they are stored, as shown by step 450 , and then the device initializes the neural network on the device with the learned weights at step 450 . To be clear, the neural network trained on the cloud network device has the same architecture as the neural network implemented on the device, and initially the neural network implemented on the device is not trained. By using the learned weights to initialize the untrained neural network, the neural network on the device is trained. After being trained, the neural network of the device is able to perform the methods of generating a vocal signature and speaker verification discussed above.
[0111] Training a neural network requires considerable computational, time and power resources. Therefore, by training the neural network in the cloud network device, the training is performed without any practical restriction on the time, or computational and power resources available. The weights are therefore learned with a high degree of precision and accuracy, which could not feasibly be achieved if the training was performed on the device.
[0112] It is emphasized that only training the neural network on the cloud network device is not enough to implement on device speaker verification. Having a neural network model with a reduced profile, and that compresses the data as described, by performing two convolutions, and then max pooling after each convolution is also necessary. The neural network model is required to ensure that the method can be performed without using an excessive amount of memory, on the device. Specifically, the entirety of the neural network model and the instructions for performing inference are embodied within a footprint of between 0 and 512 kilobytes in size. On top of that, the resulting vocal signature can be stored in a relatively small amount of memory, for example 1 kilobyte, and can be generated quickly, without consuming a large amount of electrical power. This is a direct result of the specific neural network architecture used to generate the vocal signature.
US2022405363A1 METHODS FOR IMPROVING THE PERFORMANCE OF NEURAL NETWORKS USED FOR BIOMETRIC AUTHENTICATION 20210618
[0014] In the example of speaker authentication, using an overly large data set can lead to a neural network overfitting to the training data, especially if certain categories of speakers are overrepresented in a training data set. On the other hand, it is important to ensure that a range of samples are used to train a neural network in order that the neural network is able to distinguish between a wide variety of different speakers and to ensure that a trained neural network model is able to handle environmental noise in speech samples during authentication.
[0018] As explained above, existing methods of biometric enrolment are unable to create a biometric signature that is robust enough to allow user recognition in a variety of environments. In contrast, embodiments of the first aspect of the invention allow for a biometric signature of a user to be determined that is robust to unimportant variations, which could include changes in the user's appearance or voice due to tiredness or illness and changes in environmental conditions, such as changes in lighting or ambient sounds.
[0019] This is achieved by taking multiple biometric samples from a user and using these to generate a single biometric signature of that user, which means that the biometric signature captures consistent identifying features of the user. By focussing on those identifying features that are consistent across the biometric samples, the biometric signature is robust to unimportant variations and noise and the accuracy of the neural network during authentication is therefore improved.
My Voice does not use Akida in their patents, but that's not to say that Akida would not make a big improvement:
US2023186896A1 SPEAKER VERIFICATION METHOD USING NEURAL NETWORK 20211215
[0115] The cloud network device 430 stores a neural network model 434 , and the device 420 stores a neural network model 424 . The neural network model 434 and the neural network model 424 are the same, except that the neural network model 434 on the cloud device includes a second fully-connected layer after the first fully-connected layer and a softmax function applied to the output of the second fully-connected layer. The neural network 434 on the cloud device is trained using training data 432 . An exemplary method of training a neural network is by using backpropagation, and this is discussed further below. Once trained, the learned weights 438 are extracted from the trained neural network 436 . The weights corresponding to the second fully-connected layer, and the softmax function are discarded at this stage.
[0105] With reference to the steps 400 shown in FIG. 4 , below we discuss a system for training a neural network according to the invention. The system comprises a cloud network device and a device. As shown by step 410 , the network device is configured to train a neural network, such as the neural network shown in FIG. 2 . Methods of training a neural network are known. For example, the neural network may be trained by backpropagation. Other methods could be used, as long as the value of weights connecting the nodes of the neural network are learned, as shown by step 420 . Referring to FIG. 2 , the second connected layer 280 and the softmax function 290 may be used during training the neural network. However, the weights related to these layers are not useful for generating a vocal signature. They are therefore discarded once training is complete. At step 440 , the weights are then sent to the device that performs the methods of generating a vocal signature and performing speaker verification.
[0110] Once the device receives the learned weights, they are stored, as shown by step 450 , and then the device initializes the neural network on the device with the learned weights at step 450 . To be clear, the neural network trained on the cloud network device has the same architecture as the neural network implemented on the device, and initially the neural network implemented on the device is not trained. By using the learned weights to initialize the untrained neural network, the neural network on the device is trained. After being trained, the neural network of the device is able to perform the methods of generating a vocal signature and speaker verification discussed above.
[0111] Training a neural network requires considerable computational, time and power resources. Therefore, by training the neural network in the cloud network device, the training is performed without any practical restriction on the time, or computational and power resources available. The weights are therefore learned with a high degree of precision and accuracy, which could not feasibly be achieved if the training was performed on the device.
[0112] It is emphasized that only training the neural network on the cloud network device is not enough to implement on device speaker verification. Having a neural network model with a reduced profile, and that compresses the data as described, by performing two convolutions, and then max pooling after each convolution is also necessary. The neural network model is required to ensure that the method can be performed without using an excessive amount of memory, on the device. Specifically, the entirety of the neural network model and the instructions for performing inference are embodied within a footprint of between 0 and 512 kilobytes in size. On top of that, the resulting vocal signature can be stored in a relatively small amount of memory, for example 1 kilobyte, and can be generated quickly, without consuming a large amount of electrical power. This is a direct result of the specific neural network architecture used to generate the vocal signature.
US2022405363A1 METHODS FOR IMPROVING THE PERFORMANCE OF NEURAL NETWORKS USED FOR BIOMETRIC AUTHENTICATION 20210618
[0014] In the example of speaker authentication, using an overly large data set can lead to a neural network overfitting to the training data, especially if certain categories of speakers are overrepresented in a training data set. On the other hand, it is important to ensure that a range of samples are used to train a neural network in order that the neural network is able to distinguish between a wide variety of different speakers and to ensure that a trained neural network model is able to handle environmental noise in speech samples during authentication.
[0018] As explained above, existing methods of biometric enrolment are unable to create a biometric signature that is robust enough to allow user recognition in a variety of environments. In contrast, embodiments of the first aspect of the invention allow for a biometric signature of a user to be determined that is robust to unimportant variations, which could include changes in the user's appearance or voice due to tiredness or illness and changes in environmental conditions, such as changes in lighting or ambient sounds.
[0019] This is achieved by taking multiple biometric samples from a user and using these to generate a single biometric signature of that user, which means that the biometric signature captures consistent identifying features of the user. By focussing on those identifying features that are consistent across the biometric samples, the biometric signature is robust to unimportant variations and noise and the accuracy of the neural network during authentication is therefore improved.
Sirod69
bavarian girl ;-)
BrainChip
9 Min. •
2024 STATE OF EDGE AI REPORT: Delve into the cutting-edge insights revealing the game-changing impact of Edge AI. With its ability to bring computation closer to data sources, Edge #AI offers unparalleled advantages, from reduced latency to heightened security. Witness the evolution of AI beyond the confines of centralized systems. Stay ahead of the curve with the latest intelligence on Edge AI adoption and its transformative potential. https://lnkd.in/dEzB2e5u via Wevolver
9 Min. •
2024 STATE OF EDGE AI REPORT: Delve into the cutting-edge insights revealing the game-changing impact of Edge AI. With its ability to bring computation closer to data sources, Edge #AI offers unparalleled advantages, from reduced latency to heightened security. Witness the evolution of AI beyond the confines of centralized systems. Stay ahead of the curve with the latest intelligence on Edge AI adoption and its transformative potential. https://lnkd.in/dEzB2e5u via Wevolver
Introduction | 2024 State of Edge AI Report
Exploring the Dynamic World of Edge AI Applications Across Industries
www.wevolver.com
cosors
👀
I also remember Sean said “watch the financials” and not “watch panicking anonymous investors in a forum to understand the market”
Last edited:
Sorry there is poor connectivity.
Here is the link mate. Its an old article.
https://www.electronicsmedia.info/2...evices-for-artificial-intelligence-of-things/
BrainShit
Regular
Akida is optional.... like the others as well.
BrainShit@thestockexchange:~$ sudo su
update msg prev DingoBorat
Last edited:
Anil backin it up with a two fa....
Qualcomm Acquires foundries.io, Announces Products For IoT, Edge
The announcements come at a time when Qualcomm plans to expand the portfolio of its current IoT offerings.
www.electronicsb2b.com
Last edited:
Integrating sensors with edge AI for enhanced IoT solutions
The proliferation of IoT devices and applications has generated massive data. | Edge AI refers to processing data closer to its source, reducing latency, and enhancing real-time decision-making.
Edge AI Capabilities
Edge AI brings machine learning and deep learning models to IoT devices, enabling them to process and analyze data locally. This approach provides several advantages, including reduced latency, improved privacy, and more efficient use of network resources.
Depending on the AI models deployed, the capabilities of edge AI extend to real-time object detection, anomaly detection, predictive maintenance, and even natural language processing.
This proves particularly advantageous when low-latency responses are crucial, such as autonomous vehicles, industrial automation, and healthcare. Additional benefits include remote or resource-constrained environments where continuous cloud connectivity may not be feasible.
As this technology continues to evolve, it will unlock new opportunities across various domains, reshaping the way we interact with our connected world.
Last edited:
Good morning all,
Check out the link below, someone may have already posted this I'm not sure to be honest, but it's great to see that Brainchip
is confirmed as a Gold Sponsor, there are a couple of photos of Rob at last years conference.
On another point, I was in a shop yesterday and I was told that one of Brainchip's customers (singular) is in our next generation, I
walked away and thought, did I just hear that right ? I was talking about shares and this person whom I didn't recognize blurted that out.
Strange, very strange, no company was identified.
Rumors good old rumors. Tech
aihwedgesummit.com
Check out the link below, someone may have already posted this I'm not sure to be honest, but it's great to see that Brainchip
is confirmed as a Gold Sponsor, there are a couple of photos of Rob at last years conference.
On another point, I was in a shop yesterday and I was told that one of Brainchip's customers (singular) is in our next generation, I
walked away and thought, did I just hear that right ? I was talking about shares and this person whom I didn't recognize blurted that out.
Strange, very strange, no company was identified.
Rumors good old rumors. Tech
AI Infra Summit 2025
The only event bringing together the AI & ML ecosystem under one roof.
Similar threads
