To say AI is the running theme in tech for 2023 would be a cliche at this stage of the year – but it’s undeniably true.

From ChatGPT’s whirlwind popularity among a curious and experimental public to the accelerated development among vendors of their own generative AI solutions and features, in less than a year, that tiny two-letter acronym has become an industry colossus. In many ways, it has grown into the epicentre of the tech space.

However, AI hasn’t emerged out of a vacuum, fully formed and ready to reform our civilisation ever. There are several companies and organisations who have pioneered its development, honed its progress, and contributed meaningfully to where we are today.

One of the most influential of those businesses is Qualcomm, the semiconductor and software giant that has been exploring AI for a decade, and one of that company’s leading AI experts is Megha Daga, Sr. Director, Product Management, AI/ML Lead for IoT at Qualcomm.

“AI has taken on a drastic development,” Daga told UC Today. “You can see the pace it has taken over.”

Things which used to be just generic artificial intelligence very quickly moved into machine learning and deep learning and getting closer to how you can enact the brain, which is where sensors are.”

AI’s sensor-like response to data used to be very much the premium tier of AI only a couple of years ago — and the exclusive purview of science fiction before that — but thanks to the work of Daga and her contemporaries and colleagues, that’s no longer the case.

“For example, on audio, automatic speech recognition is like a bare-minimum feature now,” she said. “Whatever form of AI I pick, I will have that enablement. That NLP (natural language processing) in the background has become the backend system. If we think about it, any kind of image enhancement, image modification, clarity, super-resolution, everything is moving towards AI.”

Daga leads Qualcomm’s AI portfolio and the charter for the whole IoT (Internet of Things) business unit. She oversees the different generations of hardware Qualcomm brings to market and the AI use cases embedded in them, as well as the software, including creating different SDKs for engineers to seamlessly utilise Qualcomm’s AI products for their own development.

“But all those different frameworks, from an AI perspective, have their own ecosystem,” she explained. “Ecosystem is a huge space for us, and we have been exponentially growing in enabling more and more third parties across that whole segment of connected smart systems.”

While Qualcomm’s AI products can be leveraged in every industry from retail to factory manufacturing, unless you’ve been living under a rock for the past year, you will have noticed it has become a prominent concept in UC and collaboration.

Smart AI capabilities that are customisable and personalisable to suit business needs have transformed hybrid meeting spaces and products, for example. But the technology behind that is incredibly complex.

“A few things are very critical,” Daga outlined, “all this enablement used to seamlessly happen into the cloud, but as these new use cases are coming and people are becoming more aware of their personalisation space, privacy is critical. When it comes to responsiveness, latency is critical. From a business perspective, cost and reliability are critical. With all those coming into existence, it has become clear that this AI enablement has to happen at the edge. It has to happen where the sensors are.”

“We were the first to do AI on the edge with a mobile phone,” Daga continued, “but we very quickly moved on to the other sensor devices. Camera and audio have been the most critical of those, as we see in those connected smart spaces now. We are looking at an environment where it’s a complete, distributed AI computing so we can start with hybrid work culture.”

When you think about (hybrid meeting) culture, it used to be just a video bar in the front or one box sitting in the centre of the room. Now, we’re looking at an ecosystem. Hybrid AI.”

“It’s not just one bar or hub sitting on the device. Now you have multiple cameras around you, and there could be a back panel gateway, all of these devices in a connected space. We’re looking at how these AIs talk to each other and, at the same time, keep the capabilities with respect to AI enablement and latency in check.”

Qualcomm’s AI-powered video bars and hubs can be enabled with the vendor’s partnered suites of platforms, such as fully certified Teams and Zoom services. The ecosystem extends to the smart cameras dotted about the room, each with use cases including audio or noise cancellation or intelligent framing of environments. All that is possible via the video bar.

In June, Qualcomm launched its next-gen Video Collaboration Suite, a solution stacked with AI-powered platforms and devices designed to accelerate the smart deployment and design of video conferencing products. It offered a cavalcade of cutting-edge hybrid meeting technology, including all-in-one video collaboration systems, huddle camera bars, smart controllers, and personal conference devices.

“The video collaboration solution has been a very cross-collaborative effort, a (mission) which has been going on for a few years now, especially after the pandemic video conferencing bang,” Daga added. “‘How do we make this hybrid culture work seamlessly no matter where you’re sitting and where you’re working from?’ Those were some of the pushes we had, and we obviously collaborated with the likes of Teams and Zoom to get certification taken care of.”

“We picked the right platforms so that our end users can get the right grade for compute and checkpoints. Then there was smart enablement, which is where AI comes in.”

“When it comes to AI, let me split it into two parts — audio and vision, the two key components. When we talk about audio, we have our own homegrown, well-proven solutions. Things like noise cancellation and suppression, that’s a key unit to it. That’s very critical. I don’t think there’s anybody in this space who has not had an issue with background noise while having these kinds of calls, so the way it works has to be seamless.

“The same thing applies to vision. With our software stack, we have homegrown solutions in detection technologies, face detection, body pose detection, and point detection on your bodies.

With that, we have built the technology of framing — framing with respect to group framing, a group of people are sitting down and focusing the camera towards them. Or speaker framing, as it is important to focus on somebody who is the major speaker in that particular scene.”

There are other complexities to honing smart vision in hybrid environments; creases that need ironing out that the layperson might not notice while deploying a hybrid space but can become a significant impediment to successful meetings.

“These technologies can seem very seamless, but there are many corner cases,” Daga explained. “Awkward room dimensions, or having a mirror in the room or glass window. All those things are very critical, so there’s a lot of fine-tuning that has gone behind it.”

As an AI pioneer for a decade, Qualcomm isn’t patting itself on the back for innovations well done and resting on its laurels. The AI boom of the last 12 months or so has excitingly brought the business’s research and products into the limelight but has also presented opportunities of its own. Generative AI, for example, is an area that Daga and her colleagues have worked on in the past but are exploring even further now and in the future.

“Generative AI has taken a huge foundational base. We have showcased gen AI applications on devices. That’s not just on mobile, it’s across the business unit, be it on mobile or computer, and the same thing you will see soon on smart systems. It’s across the portfolio.”

“This isn’t the end,” Daga said. “There is so much more that we’re looking at, especially generative AI. It changes the whole space of consumer behaviour.”

While AI’s explosion into mainstream consciousness has seen Qualcomm adapt on the fly, there has been no seismic change in strategic direction. Hardware, for example, is changing in design, as massive memory-centric models like LLMs or ChatGPT might soon be outmoded by AIs that aren’t memory-bound. Qualcomm will adapt to that kind of transition but will persist with the best practices it has refined for a decade as one of the most experienced businesses in the AI game.

“Because we’ve been here since the start, we know what we need from both hardware and software,” Daga said. “There are design evolutions that will happen, and you will change some directional aspects of things. But it’s not like we were working towards a plan, and it all went apart. We are revolutionising and adapting to how things will change while, at the same time, not breaking what we have.”



from UC Today https://ift.tt/wY2sGAn