BRN Discussion Ongoing

Diogenese

Top 20
So, reading the paragraph after & relating to 180 it seems to me they have essentially replaced the phrase " wake word" and embedded or left open placeholders to be later embedded, words or phrases with weightings so the assistant recognises them....just like a wake word without specifically saying "hey...whoever" :rolleyes:
Just a note of caution about patents - they are usually not published until 18 months after filing, so they may have a revised version in the pipeline.
So, reading the paragraph after & relating to 180 it seems to me they have essentially replaced the phrase " wake word" and embedded or left open placeholders to be later embedded, words or phrases with weightings so the assistant recognises them....just like a wake word without specifically saying "hey...whoever" :rolleyes:
... but they also have the option of using the video input to determine the speaker's intention:

0045] In some examples, the reasoner implements a classifier that takes advantage of yet other input to make the classification. For example, video from the camera 103 monitoring the user's facial expressions, lip movement, etc., is provided to a visual classifier 157 of the reasoner 150 to aid in the classification. In a machine learning approach, such a video signal may be processed using a machine learning approach that is trained on video (i.e., image sequences) for users making system-directed and non-system-directed utterances. In some examples, the video is processed in conjunction with the audio, essentially combining functions of the acoustic classifier 152 and visual classifier 157 (e.g., to permit taking advantage of relationships between visual and audio cues that help determine when the user intends for an utterance to be system directed).
 
  • Like
Reactions: 6 users

TECH

Regular
I stand by my comment, until we have a signed deal then it’s nothing. Chat, partnerships, discussions, articles, and tweets mean diddly squat unless they sign, start using Akida, and revenue starts flowing. Until then they have zero commitment towards Brainchip. That’s just the way it is.

What you have stated is 100% how many would feel due mainly because of the lack of IP signings, and revenue streams having been
established as yet, to counter your view, I have been told that Mercedes Benz is a "client" and will do what they will do with our technology,
as is their right...read that as you wish.

Sean is certainly moving around the globe now, or appears to be, I personally think the name Brainchip and more importantly, AKIDA 2.0
is starting to be more freely talked about in the tech world and the word success can take many forms, and to this point we have had some
tremendous success but ultimately we all know that a business can't survive without revenue, have faith it's coming.

What makes me laugh somewhat is persons whom bag our company, but still feel comfortable holding, is that a modern definition of a
"fence sitter" or a politician ?

Tech in Melbourne 💎 AKD 3.0
 
  • Like
  • Love
  • Fire
Reactions: 46 users

Iseki

Regular
Ahhhhh, I see, you're looking for an argument.
That room is three doors on the left down the hall. 🤣
Not at all.
When you say that there are wonderful things happening only we can't know about them, you are causing a distrust of the board that will bring about an outcome that neither you or I want. Please desist. And let's not fight.
Instead let's find a new way that BRN can spread the word AND give shareholders something to hang on to.

eg - Let's give the akida IP away free for the first three medical devices that can utilize it. Something like that - win, win, win.
 
Last edited:
  • Like
  • Love
Reactions: 2 users

Diogenese

Top 20
Not sure of date on this article, but suggests DNN embedded and cloud?


CERENCE INTRODUCES NEW FEATURES IN CERENCE DRIVE, THE WORLD’S LEADING TECHNOLOGY AND SOLUTIONS PORTFOLIO FOR AUTOMAKERS AND CONNECTED CARS​

New capabilities such as enhanced voice recognition and synthetic speech serve as the foundation for a safer, more enjoyable journey for everyone
BURLINGTON, Mass. – Cerence Inc., AI for a world in motion, today introduced new innovations in Cerence Drive, its technology and solutions portfolio for automakers and IoT providers to build high-quality, intelligent voice assistant experiences and speech-enabled applications. Cerence Drive today powers AI-based, voice-enabled assistants in approximately 300 million cars from nearly every major automaker in the world, including Audi, BMW, Daimler, Ford, Geely, GM, SAIC, Toyota, and many more.
The Cerence Drive portfolio offers a distinct, hybrid approach with both on-board and cloud-based technologies that include voice recognition, natural language understanding (NLU), text-to-speech (TTS), speech signal enhancement (SSE), and more. These technologies can be deployed and tightly integrated with the wide variety of systems, sensors and interfaces found in today’s connected cars. The latest version of Cerence Drive includes a variety of new features to elevate the in-car experience:
> Enhanced, active voice recognition and assistant activation that goes beyond the standard push-to-talk buttons and wake-up words. The voice assistant is always listening for a relevant utterance, question or command, much like a personal assistant would, creating a more natural experience. In addition, Cerence’s voice recognition can run throughout the car, both embedded and in the cloud, distributing the technical load and delivering a faster user experience for drivers.
> New, deep neural net (DNN)-based NLU engine built on one central technology stack with 23 languages available both embedded and in the cloud. This streamlined approach creates new standards for scalability and flexibility between embedded and cloud applications and domains for simpler integration, faster innovation, and a more seamless in-car experience, regardless of connectivity.
> TTS and synthetic voice advancements that deliver new customizations, including a non-gender-specific voice for the voice assistant, and emotional output, which enables automakers to adjust an assistant’s speaking style based on the information delivered or tailored to a specific situation. In addition, the introduction of deep learning delivers a more natural and human-like voice with an affordable computational footprint.
> Improved, more intelligent speech signal enhancement that includes multi-zone processing with quick and simple speaker identification; passenger interference cancelation that blocks out background noise as well as voices from others in the car; and a deep neural net-based approach for greater noise suppression and better communication.
“Improving the experience for drivers and creating curated technology that feels unique and harmonious with our partners’ brands have been true motivators since we started our new journey as Cerence, and that extends to our latest innovations in Cerence Drive,” said Sanjay Dhawan, CEO, Cerence. “Cerence Drive, our flagship offering, is the driving force behind our promise of a truly moving in-car experience for our customers and their drivers, and our new innovations announced today are core to making that mission a reality. ”
Cerence Drive’s newest features are available now for automakers worldwide. To learn more about Cerence Drive, visit www.cerence.com/solutions.

Also a 2022 pdf spiel on their overall solutions package.

HERE

Guess we have to remember we also have a patent granted on 2018 on neuromorphic application via PVDM.


US-10157629-B2 - Low Power Neuromorphic Voice Activation System and Method​


Abstract
The present invention provides a system and method for controlling a device by recognizing voice commands through a spiking neural network. The system comprises a spiking neural adaptive processor receiving an input stream that is being forwarded from a microphone, a decimation filter and then an artificial cochlea. The spiking neural adaptive processor further comprises a first spiking neural network and a second spiking neural network. The first spiking neural network checks for voice activities in output spikes received from artificial cochlea. If any voice activity is detected, it activates the second spiking neural network and passes the output spike of the artificial cochlea to the second spiking neural network that is further configured to recognize spike patterns indicative of specific voice commands. If the first spiking neural network does not detect any voice activity, it halts the second spiking neural network.

This Cerence patent uses hybrid SoC/Cloud:

1686737304323.png


US11462216B2 Hybrid arbitration system
A method for selecting a speech recognition result on a computing device includes receiving a first speech recognition result determined by the computing device, receiving first features, at least some of the features being determined using the first speech recognition result, determining whether to select the first speech recognition result or to wait for a second speech recognition result determined by a cloud computing service based at least in part on the first speech recognition result and the first features.
Hi Dr E,

I too share your confusion about Cerence.

However, there are some points which mitigate my concerns somewhat.

A few weeks ago, Mercedes started talking about MBUX not needing the "Hey Mercedes" wake up when there was only one person in the car, the corollary being that they still need it when there are two or more people.

Another thing is I've looked at Cerence patents, and while they discuss the use of NNs, they do not describe or claim any NN circuitry.

As you say, Mercedes found Akida to be 5 to 10 times beter than other systems for "Hey Mercedes". They also used "Hey Mercedes" as an example of what Akida could do, and appeared to make reference to plural uses of Akida.

On top of that, Mercedes also stated their desire to standardize on the chips they use. Akida is sensor agnostic.

Then there's Valeo Scala 3 lidar due out shortly, which I think may contain Akida, leaving aside Luminar with their foveated lidar and who have stated that they expect to expand their cooperation with Mercedes from mid-decade. MB used Scala 2 to obtain Level 3 ADAS certification, (sub-60 kph), while Scala 3 is rated to 160 kph.

Luminar, like Cerence, talk about using AI, but do not describe its construction.

Standardizing on Akida would improve the efficiency of the MB design office as their engineers would all be signing off the same hymn sheet in close harmony.


https://www.bing.com/videos/search?&q=Comedian+Harmonists+Songs&view=detail&mid=44C7F5D2E5E5041230E244C7F5D2E5E5041230E2&FORM=VDRVRV&ru=/videos/search?&q=Comedian+Harmonists+Songs&FORM=VDRSCL&ajaxhist=0


https://www.bing.com/videos/search?...edian+Harmonists+Songs&FORM=VDRSCL&ajaxhist=0
 
  • Like
  • Haha
Reactions: 7 users

robsmark

Regular
What you have stated is 100% how many would feel due mainly because of the lack of IP signings, and revenue streams having been
established as yet, to counter your view, I have been told that Mercedes Benz is a "client" and will do what they will do with our technology,
as is their right...read that as you wish.

Sean is certainly moving around the globe now, or appears to be, I personally think the name Brainchip and more importantly, AKIDA 2.0
is starting to be more freely talked about in the tech world and the word success can take many forms, and to this point we have had some
tremendous success but ultimately we all know that a business can't survive without revenue, have faith it's coming.

What makes me laugh somewhat is persons whom bag our company, but still feel comfortable holding, is that a modern definition of a
"fence sitter" or a politician ?

Tech in Melbourne 💎 AKD 3.0
Hey Tech,

For what it’s worth, I think that Mercedes will likely be a future customer, but as I have repeatedly mentioned in my last few posts, find it very unlikely that they currently are as it hasnt been announced and the company has a continual disclosure as per ASX requirements to release this information as soon as it becomes available.

I have no issues with Sean travelling overseas to promote the product, in fact as a CEO I would expect it.

Cheers.
 
  • Like
  • Thinking
Reactions: 12 users

manny100

Regular
All the NDA talk also mean diddly squat too. It may or not be true..
That is the risk. If it all comes off as expected these prices will seem cheap. If not well...
Its just patience from here. There will be some who cannot stand the wait and sell. There will be others who maintain a hold or a hold and accumulate like myself.
BRN has at least a 2 plus year of first mover advantage so we need it to gather momentum before others catch up.
I note AKIDA 3 will be on the drawing board later this year.
 
  • Like
  • Fire
Reactions: 15 users
HI Dio,

This is more a question than statement. When Mercedes.mentioned their relationship with Brainchip a few of us did a deep dive into Cerence. Promising but nothing definitive. Bravo was one ring leader if I remember correctly. I have heard you state that, it takes about 18 months for patents to become public. It is now about 18 months since it became known about Mercedes and they would have been working on this for some time. It seems to me if Cerence were involved we should be seeing new patents coming from Cerence. Or are current patents enough as written to be working with us?

SC
 
  • Like
Reactions: 8 users
So, reading the paragraph after & relating to 180 it seems to me they have essentially replaced the phrase " wake word" and embedded or left open placeholders to be later embedded, words or phrases with weightings so the assistant recognises them....just like a wake word without specifically saying "hey...whoever" :rolleyes:

Thanks guys for your input re the Cerence/Akida matter.

I simply thought it was Akida the hardware; and Cerence the software.

Akida on all the sensors either sending signals to act on the info directly or feeding relevant info to Nvidia’s main control unit.

Akida on Valeo’s Lidar interpreting the info.

Thanks for the deep dive.

I have found this Brainchip journey to be quite intriguing and exciting!

Many chapters still to be written in this adventure; can’t wait for the next instalment.

:)
 
Last edited:
  • Like
  • Love
  • Fire
Reactions: 36 users

manny100

Regular
What happened to the chips (AKIDA 1000) we received from Socionext in August 2021?
 
  • Like
  • Thinking
  • Haha
Reactions: 7 users

Getupthere

Regular
  • Like
  • Thinking
Reactions: 3 users
What happened to the chips (AKIDA 1000) we received from Socionext in August 2021?

Hi @manny100

It’s my understanding they were reference chips to prove they worked in silicon.

1686744947828.png

You could buy them to test via Development Kits etc as per below.

1686745001952.png



I imagine the same will happen with Gen 2 variations although there shouldn’t be as much doubt over them working as essentially they are the same tech, just different sizes (large, medium and small) depending on use case.

I’m confident that Brainchip has listened to the customer and is providing them what they have asked for. As Sean said, they will be more easily consumed/used and be a better fit depending on the application.

Edit: in saying that the TENNS sounds like a huge technological advancement which raises the bar significantly so I shouldn’t ignore that new feature!

:)
 
  • Like
  • Love
Reactions: 33 users

manny100

Regular
Hi @manny100

It’s my understanding they were reference chips to prove they worked in silicon.

View attachment 38368
You could buy them to test via Development Kits etc as per below.

View attachment 38369


I imagine the same will happen with Gen 2 variations although there shouldn’t be as much doubt over them working as essentially they are the same tech, just different sizes (large, medium and small) depending on use case.

I’m confident that Brainchip has listened to the customer and is providing them what they have asked for. As Sean said, they will be more easily consumed/used and be a better fit depending on the application.

Edit: in saying that the TENNS sounds like a huge technological advancement which raises the bar significantly so I shouldn’t ignore that new feature!

:)
Thanks, cheers
 
  • Like
Reactions: 4 users
This Cerence patent uses hybrid SoC/Cloud:

View attachment 38365

US11462216B2 Hybrid arbitration system
A method for selecting a speech recognition result on a computing device includes receiving a first speech recognition result determined by the computing device, receiving first features, at least some of the features being determined using the first speech recognition result, determining whether to select the first speech recognition result or to wait for a second speech recognition result determined by a cloud computing service based at least in part on the first speech recognition result and the first features.



https://www.bing.com/videos/search?&q=Comedian+Harmonists+Songs&view=detail&mid=44C7F5D2E5E5041230E244C7F5D2E5E5041230E2&FORM=VDRVRV&ru=/videos/search?&q=Comedian+Harmonists+Songs&FORM=VDRSCL&ajaxhist=0


https://www.bing.com/videos/search?&q=Comedian+Harmonists+Songs&view=detail&mid=252FACB077B6AE44C21A252FACB077B6AE44C21A&FORM=VDRVRV&ru=/videos/search?&q=Comedian+Harmonists+Songs&FORM=VDRSCL&ajaxhist=0
@Diogenese

Recent job vacancy at Cerence.

Indication of some probable AI they using?

I note that TDNN was first developed in the late 1980s apparently :oops:



Product R&D Engineer - Acoustic Modelling

Cerence
Be the first to apply
Posted 11d ago


Job highlights
Strong background knowledge and hands-on experience in speech processing technologies,with particular familiarity in speech signal processing and speech enhancement techniques . Good understanding or experience in acoustic echo cancellation and noise reduction technologies


Who you are:

  • Strong background knowledge and hands-on experience in speech processing technologies, with particular familiarity in speech signal processing and speech enhancement techniques
  • Good understanding or experience in acoustic echo cancellation and noise reduction technologies
  • Familiarity with one or more deep learning technologies, such as TDNN, Deep Feed Forward Neural Network, and LSTM, would be an advantage
  • Proven industry or academic experience in developing large vocabulary speech recognition systems
  • Extensive knowledge and hands-on experience in statistical language modeling, including estimation methods, smoothing, pruning, efficient representation, and interpolation techniques
  • Experience in conducting accuracy experiments and systematically improving performance
  • Self-driven and diligent in solving real-world problems, with the ability to manage tasks and deliverables for multiple projects simultaneously
  • An understanding of information and application cyber s ecurity standards (secure coding, securing SDLC s, etc)
 
  • Like
  • Thinking
Reactions: 12 users

Diogenese

Top 20
@Diogenese

Recent job vacancy at Cerence.

Indication of some probable AI they using?

I note that TDNN was first developed in the late 1980s apparently :oops:



Product R&D Engineer - Acoustic Modelling

Cerence
Be the first to apply
Posted 11d ago


Job highlights
Strong background knowledge and hands-on experience in speech processing technologies,with particular familiarity in speech signal processing and speech enhancement techniques . Good understanding or experience in acoustic echo cancellation and noise reduction technologies


Who you are:

  • Strong background knowledge and hands-on experience in speech processing technologies, with particular familiarity in speech signal processing and speech enhancement techniques
  • Good understanding or experience in acoustic echo cancellation and noise reduction technologies
  • Familiarity with one or more deep learning technologies, such as TDNN, Deep Feed Forward Neural Network, and LSTM, would be an advantage
  • Proven industry or academic experience in developing large vocabulary speech recognition systems
  • Extensive knowledge and hands-on experience in statistical language modeling, including estimation methods, smoothing, pruning, efficient representation, and interpolation techniques
  • Experience in conducting accuracy experiments and systematically improving performance
  • Self-driven and diligent in solving real-world problems, with the ability to manage tasks and deliverables for multiple projects simultaneously
  • An understanding of information and application cyber s ecurity standards (secure coding, securing SDLC s, etc)
TDNN sounds a bit like LSTM, only different.

https://www.isca-speech.org/archive/pdfs/odyssey_2020/huang20_odyssey.pdf
 
Last edited:
  • Like
Reactions: 5 users

cassip

Regular
Mercedes One-Eleven

 
Last edited:
  • Like
Reactions: 4 users

GStocks123

Regular


Worth the watch- Sailesh from Renesas presenting on AI at endpoint/edge (specifically digital). I like those figures!
 
  • Like
  • Fire
Reactions: 10 users

GStocks123

Regular
  • Like
  • Love
  • Fire
Reactions: 22 users

Townyj

Ermahgerd
Last edited:
  • Like
  • Fire
  • Love
Reactions: 31 users

Tothemoon24

Top 20

TinyML computer vision is turning into reality with microNPUs (µNPUs)​

June 14, 2023 Elad Baram
Ubiquitous ML-based vision processing at the edge is advancing as hardware costs decrease, computation capability increases significantly, and new methodologies make it easier to train and deploy models. This leads to fewer barriers to adoption and increased use of computer vision AI at the edge.


Advertisement

Computer vision (CV) technology today is at an inflection point, with major trends converging to enable what has been a cloud technology to become ubiquitous in tiny edge AI devices. Technology advancements are enabling this cloud-centric AI technology to extend to the edge, and new developments will make AI vision at the edge pervasive.
There are three major technological trends enabling this evolution. New, lean neural network algorithms fit the memory space and compute power of tiny devices. New silicon architectures are offering orders of magnitude more efficiency for neural network processing than conventional microcontrollers (MCUs). And AI frameworks for smaller microprocessors are maturing, reducing barriers to developing tiny machine learning (ML) implementations at the edge (tinyML).
As all these elements come together, tiny processors at milliwatt scale can have powerful neural processing units that execute extremely efficient convolutional neural networks (CNNs)—the ML architecture most common for vision processing—leveraging a mature and easy-to-use development tool chain. This will enable exciting new use cases across just about every aspect of our lives.

The promise of CV at the edge

Digital image processing—as it used to be called—is used for applications ranging from semiconductor manufacturing and inspection to advanced driver assistance systems (ADAS) features such as lane-departure warning and blind-spot detection, to image beautification and manipulation on mobile devices. And looking ahead, CV technology at the edge is enabling the next level of human machine interfaces (HMIs).

HMIs have evolved significantly in the last decade. On top of traditional interfaces like the keyboard and mouse, we have now touch displays, fingerprint readers, facial recognition systems, and voice command capabilities. While clearly improving the user experience, these methods have one other attribute in common—they all react to user actions. The next level of HMI will be devices that understand users and their environment via contextual awareness.


Context-aware devices sense not only their users, but also the environment in which they are operating, all in order to make better decisions toward more useful automated interactions. For example, a laptop visually senses when a user is attentive and can adapt its behavior and power policy accordingly. This is already being enabled by Synaptics’ Emza Visual Sense technology, which OEMS can use to optimize power by adaptively dimming the display when a user is not watching it, reducing display energy consumption (figure 1). By tracking on-lookers’ eyeballs (on-looker detect) the technology can also enhance security by alerting the user and hiding the screen content until the coast is clear.

There are also endless use cases for visual sensing in industrial fields, ranging from object detection for safety regulation (i.e., restricted zones, safe passages, protective gear enforcement) up to anomaly detection for manufacturing process control. In agritech, crop inspections, and status and quality monitoring enabled by CV technologies, are all critical.

Whether it’s in laptops, consumer electronics, smart building sensors or industrial environments, this ambient computing capability is enabled when tiny and affordable microprocessors, tiny neural networks, and optimized AI frameworks make devices more intelligent and power efficient.

Neural-network vision processing evolves

2012 marked the turning point when CV started to shift from heuristic CV methods to deep convolutional neural networks (DCNN), with the publication of AlexNet by Alex Krizhevsky and his colleagues. There was no turning back after the DCNN won the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) that year.

Since then, teams across the globe have continued to seek higher detection performance, but without much concern about the efficiency of the underlying hardware. So CNNs continued to be data- and compute-hungry. This focus on performance was fine for applications running in the cloud infrastructure.

In 2015, ResNet152 was introduced. It had 60 million parameters, required more than 11 gigaflops for single-inference operation and demonstrated 94% top-5 accuracy for the ImageNet data set. This continued to push the performance and accuracy of CNNs. But it wasn’t until 2017, with the publication of MobileNets by a group of researchers from Google, that we saw a push toward efficiency.

MobileNets—aimed at smartphones—was significantly lighter than existing neural network (NN) architectures at that time. MobileNetV2, as an example, had 3.5 million parameters and required 336 Mflops. This drastic reduction was achieved initially through hard labor—manually identifying layers in the deep-learning network that did not add much to accuracy. Later, automated architecture-search tools allowed further improvement in the number and organization of layers. Roughly 20x “lighter” than ResNet192, both in memory and computational load, MobileNetV2 demonstrated top-5 accuracy of 90%. A new set of mobile-friendly applications could now use AI.

And hardware evolves

With smaller NNs and with a clear understanding of the workloads involved, developers could now design optimized silicon for tiny AI. This led to the micro neural processing unit (micro NPU). By tightly managing memory organization and data flow, while exploiting massive parallelism, these small, dedicated cores can execute NN inference 10x or 100x faster than the unaided CPU in a typical MCU. An example is the Arm Ethos U55 micro NPU.

Let’s look at a specific example of the impact of microNPUs (µNPUs). One of the fundamental tasks in CV is object detection. Object detection in essence requires two tasks: localization, which determines where an object is located within the image, and classification, which identifies the detected object (figure 2).

Emza has implemented a face detection model on an Ethos U55 µNPU, training an object detection and classification model that is a lightweight version of the single shot detector, optimized by Synaptics for detecting just the class of faces. The results astonished us with model execution times of less than 5 milliseconds: this is comparable to the execution speed on a powerful smartphone application processor, like the Snapdragon 845. When executing this same model on the Raspberry Pi 3B using four Cortex A53 cores, the execution time is six times longer.

AI frameworks & democratization

Widespread adoption of any technology as complex as ML requires good development tools. TensorFlow Lite for Microcontrollers (TFLM) by Google is a framework designed for easier training and deployment of AI for tinyML. For a subset of the operators covered by the full TensorFlow, TFLM emits microprocessor C code for an interpreter and a model to run on a µNPU. The PyTorch Mobileframework and Glow compiler from Meta are also targeting this area. In addition, there are today quite a few AI automation platforms (known as AutoML) that can automate some aspects of AI deployment for tiny targets. Examples are Edge Impulse, Deeplite, Qeexo, and SensiML.

But to enable execution on specific hardware and µNPUs, compilers and tool chains must be modified. Arm has developed the Vela compiler that optimizes CNN model execution for the U55 µNPU. The Vela complier removes the complexities of a system that contains both a CPU and a µNPU by automatically splitting the model execution task between them.

More broadly, the Apache TVM is an open-source, end-to-end ML compiler framework for CPUs, GPUs, NPUs and accelerators. And TVM micro is targeting microcontrollers with the vision of running any AI model on any hardware. This evolution of AI frameworks, AutoML platforms, and compilers makes it easier for developers to leverage the new µNPUs for their specific needs.

Ubiquitous AI at the edge

The trend toward ubiquitous ML-based vision processing at the edge is clear. Hardware costs are decreasing, computation capability is increasing significantly, and new methodologies make it easier to train and deploy models. All of this is leading to fewer barriers to adoption, and to increased use of CV AI at the edge.

But even as we see increasingly ubiquitous tiny edge AI, there is still work to do. To make ambient computing a reality, we need to serve the long tail of use cases in many segments that can create a scalability challenge. In consumer products, factories, agriculture, retail and other segments, each new task requires different algorithms and unique data sets for training. The R&D investments and skillset needed to solve for each use case continue to be a major barrier today.

This gap can best be filled by AI companies up-levelling the software around their NPU offerings by developing rich sets of model examples—”model zoos”—and applications reference code. In doing so, they can enable a wider range of applications for the long tail while ensuring design success by having the right algorithms optimized to the target hardware to solve specific business needs, within the defined cost, size, and power constraints.
 
  • Like
Reactions: 10 users
Top Bottom