Tech

Unlock superior AI with new tools from Deci


You can’t attend Transform 2022? View all summit sessions in our on-demand library now! See here.


Edge devices must be able to process distributed data quickly and in real time. And, advanced AI applications are only efficient and scalable when they can produce highly accurate image predictions.

Get the important and complex task is automatic driving: All relevant objects in the driving scene must be taken into account – be it pedestrians, lanes, sidewalks, other vehicles or traffic signs and lights.

“For example, an autonomous vehicle driving through a crowded city must maintain high accuracy while operating in real time with very low latency; otherwise, the lives of drivers and pedestrians could be in danger,” said Yonatan Geifman, CEO and co-founder of the deep learning firm. Deci.

The key to this is semantic segmentation, or image segment. There is a catch, however: Semantic segmentation models are complex, often slowing down their performance.

Incident

MetaBeat 2022

MetaBeat will bring together thought leaders to deliver guidance on how metaverse technology will transform the way all industries communicate and do business on October 4 in San Francisco, CA.

register here

“There is often a trade-off between accuracy and speed and the size of these models,” says Geifman, whose company has released a set of semantic segmentation models, DeciSeg. know this week, to help solve this complex problem.

“This can be a barrier for real-time applications,” says Geifman. “Creating accurate and computationally efficient models is a real hit for deep learning engineers, who are working so hard to achieve both the accuracy and speed to meet their needs. task at hand.”

The power of the edge

Based on Allied market researchGlobal advanced AI (artificial intelligence) market size will reach nearly $39 billion by 2030, a compound annual growth rate (CAGR) of nearly 19% within 10 years. Meanwhile, Astute Analytica reports that the global AI software market will reach more than 8 billion dollars by 2027, a CAGR of nearly 30% from 2021.

“Edge computing with AI is a powerful combination that could offer promising applications for both consumers and businesses,” said Geifman.

For the end user, this means higher speed, improved reliability and a better overall experience, he said. Not to mention better data security, as the data used for processing remains on the local device – mobile phones, laptops, tablets – and does not have to be uploaded to services third-party cloud. For businesses with consumer applications, this means a significant reduction in cloud computing costs, says Geifman.

Another reason why competitive AI is so important: Communication bottlenecks. Much machine vision edge devices require heavy duty analysis for high resolution video streams. However, if the communication request is too large for the network capacity, some users will not get the necessary analysis. “Therefore, moving computation to the edge, even partially, will allow for large-scale operations,” says Geifman.

There are no significant trade-offs

Semantic segmentation is key to advanced AI and is one of the most widely used computer vision tasks across many business verticals: automotive, healthcare, agriculture, media, and entertainment location, consumer applications, smart cities, and other image-intensive deployments.

Many of these applications are “very important in the sense that getting accurate and real-time segmentation predictions can be a matter of life or death,” Geifman said.

Self-propelled vehicle, for one person; the other is the heart semantic segment. For this important task in MRI analysis, explains Geifman, the image is segmented into a number of anatomically significant segments that are used to estimate important factors such as myocardial mass and wall thickness.

Of course, there are examples beyond critical situations, such as virtual background features of video conferencing or smart photography.

Unlike image classification models – which are designed to identify and label an object in a given image – semantic segmentation models assign a label to each pixel in an image, Geifman explains. . They are usually designed using an encoder/decoder structure. The encoder gradually samples the input while increasing the number of feature maps, thus constructing informative spatial features. The decoder takes these features and gradually upscales them to a full resolution segment map.

And, although it is often required for many edge AI applications, there are significant barriers to running semantic segmentation models directly on edge devices. These include high latency and cannot deploy models due to their size.

Geifman explains that very accurate segmentation models are not only much larger than classifiers, they are often applied on larger input images, which “quadratic” increases in complexity. their computational complexity. This translates into slower inference performance.

For example, Error checking systems running on a production line must maintain high accuracy to reduce false alarms, but speed cannot be sacrificed in the process, says Geifman.

Lower latency, higher accuracy

DeciSeg models are automatically generated using Deci’s Automated Neural Architecture Construction (AutoNAC) technology. The Tel Aviv-based company says these models are “significantly better” than existing publicly available models, including Apple’s. MobileViT and by Google DeepLab.

As Geifman explained, the AutoNAC engine looks at a large search space of neural architectures. While searching this space, it takes into account parameters such as baseline accuracy, performance target, inference hardware, compiler, and quantization. AutoNAC attempts to solve a constrained optimization problem while accomplishing several goals at once – that is, maintaining baseline accuracy with a model with a given memory.

These models deliver 2x lower latency and 3 to 7% higher accuracy, says Geifman. This allows companies to develop new use cases and applications on advanced AI devices, reducing the cost of inference (as AI practitioners will no longer need to run tasks in the environment). expensive cloud), opens up new markets and shortens development time, Geifman said. AI teams can solve deployment challenges while achieving the accuracy, speed, and model size desired.

“DeciSeg models enable semantic segmentation tasks that were previously impossible on edge applications because they are so resource-intensive,” says Geifman. The new set of models “has the potential to transform industries at large.”

VentureBeat’s mission is a digital city square for technical decision-makers to gain knowledge of transformative enterprise technology and transactions. Explore our summary report.



Source link

goznews

Goz News: Update the world's latest breaking news online of the day, breaking news, politics, society today, international mainstream news .Updated news 24/7: Entertainment, Sports...at the World everyday world. Hot news, images, video clips that are updated quickly and reliably.

Related Articles

Back to top button