FEED Issue 25

62 SITE VISIT InterDigital

had a dedicated group working on imaging research, like video compression, but since Oisel’s arrival the lab has grown to more than 80 researchers and now includes research into HDR. The lab’s work in prototyping, demonstration and research also takes in some of InterDigital’s other partnerships in the imaging field, which include Technicolor, Phillips and Sony, as well as academic institutions. When Oisel’s team was at Technicolor, its remit was split between researchers who were contracted out to InterDigital and Technicolor’s production services business, which largely served the visual effects industry. The move into InterDigital’s research department has completely changed the way team members think about their work. “The DNA of InterDigital was completely different from the DNA of Technicolor,” said Oisel. “At InterDigital, most people are researchers, with PhDs and research is their main motivation. At Technicolor, research was short term – it claimed to be long term, but long term for Technicolor is one year. That’s really long term for the VFX industry. “Now that we’ve moved to InterDigital, short term is five years and long term – I don’t even know what long term is. We know we won’t have a direct impact today. We’re preparing for the future. That’s good for researchers and it’s a good way to attract people. InterDigital is all about research and I don’t know too many other companies like that.” FINDING FRUGAL AI Laurent Depersin heads up InterDigital’s Home Lab, which looks at delivering new technologies to the home, including a whole basket of technologies aimed at

IT TAKES TWO A digital duplicate of the FEED editor

benefiting the end customer. These are everything from wireless networking, including 5G, video and VR streaming, cloud computing and gaming to in-home use of AI and edge computing. Depersin sees a lot of promise in machine learning in the future of video – and not in the unsettling robots vs humans sense that’s producing so much anxiety. WE SEE A TIME WHEN YOU WILL HAVE JUST A SINGLE AI IN THE HOME THAT WILL INTERFACE WITH EVERY ROOM

His lab is looking at the possibilities of using AI much closer to edge devices and in the home. These local AI’s will use less energy and be better at preserving privacy. The ubiquitous connectivity promised in the future, driven by AI, also promises to consume an unsustainable amount of energy. Rather than AI and machine learning operating in every device in every home and workplace, Depersin sees a single, local AI being used for many purposes. “Now if you have several rooms and you want to use an Amazon Alexa, you need several Alexas,” he says. “But we see a time when you will have just a single AI in the home that will interface with every room and none of that data would have to leave the building.” InterDigital’s Palo Alto AI lab are working on deep neural model compression, which tries to minimise the amount of memory and processing an AI is using. Meanwhile, Depersin’s lab in Rennes is also working on what is called “organic AI” or “frugal AI”. “When you show a kid a picture of a cat, you don’t need to show him several thousand cats so that he can recognise it,” explains Depersin. “This is something we’re still missing with the deep learning approach. It still requires a lot of data for one task. This is the idea of creating an AI that can learn on the go and doesn’t need a huge amount of processing power.” The time of an all-encompassing virtual reality, where we all interact with each other’s digital avatars in a zero-latency global network driven by personal AIs might seems a few years off, but at InterDigital, they’re right around the corner. Get ready, the future is coming.

feedzinesocial feedmagazine.tv

Powered by