Israeli startup Cortica says it has developed an AI-based technology that allows computers to see and understand their surroundings by mimicking the function of mammalian brains.
The technology, according to co-founder Igal Raichelgauz, outshines that of Mobileye, Israel’s leading maker of self-driving car technologies, which was acquired by Intel Corp. in 2017 for a whopping $15.3 billion.
“There is today nothing close to Mobileye in the market,” Raichelgauz said at the startup’s offices in Tel Aviv, but “we believe we can be a strong and better alternative to Mobileye.”
Cortica made headlines last year when local press, including The Times of Israel, reported that during his visit to Israel, Elon Musk, the CEO of electric-car maker Tesla, would be meeting with Cortica. Musk later denied meeting with the company, saying the reports were completely false and that he had never even heard of the startup.
In the interview with The Times of Israel, Raichelgauz declined to comment on the matter.
Cortica, founded in 2007, employs some 120 people in Tel Aviv and New York. This year, 2019, will be a “breakthrough year” for the firm, which will see its software and chips transition from the lab and into the automotive industry, leading to a “massive deployment” of Cortica-enabled solutions on the road by 2021, predicted Raichelgauz.
In January, the firm said it would start collaborating with Renesas Electronics, a Japanese maker of chips and processors for the automobile industry, to put Cortica’s AI-based autonomous technology into the chips for the front cameras developed by the Japanese firm, which will be installed into new cars.
In a statement, Cortica said the collaboration was a “milestone” for the startup, and will pave the way for the mass deployment of the company’s advanced driver assistance systems (ADAS) and automating driving (AD) applications in the automobile industry. Together with Renesas, “we will provide a technology that competes with Mobileye,” Raichelgauz said.
In an emailed comment to The Times of Israel regarding Cortica’s claim that its technology will compete with that of Mobileye, a Mobileye spokeswoman said: “Aspirations are a worthy goal. When a technological actor supplies technology to production cars in volume we could begin to relate and consider this as competition. This is not the case here.”
“What we are bringing to the market is not just a new product but a new approach,” said the 38-year old Raichelgauz, a graduate of electrical engineering from Haifa’s Technion – Israel Institute of Technology and of the Israeli army’s elite 8200 tech unit. “We have developed a mathematical model that mimics what the brain of mammals can do.”
With machine learning — the technology used by companies like Mobileye to teach computers to recognize objects — computers are manually fed with millions of images, a process that teaches them to recognize and distinguish objects. The process is labor intensive and needs very high computational power.
Cortica’s approach is different, said Raichelgauz. The technology is modeled on the neuron activity and learning mechanisms occurring in the brain. The technology is based on research done at the Technion by Cortica co-founders Raichelgauz, Karina Odinaev, a specialist in brain sciences, and their computer vision and neuroscience professor Josh Zeevi.
Together they studied the cortex of a rat, to which they connected electrodes and sent electric signals of audio and images. For each signal they sent they got a specific neural reaction, or what is called a “digital signature.”
“Every object has a signature,” explained Raichelgauz.
The researchers then transformed these signatures into mathematical algorithms and stored them in a database for the computers or chips to access. When exposed to stimuli, computers can refer back to the signatures they have stored, recognizing the object or noise. If, however they are exposed to new stimuli, then, just like newborns, they learn organically. They take the new input and compare it to the data they have stored in their memories and find similar signatures. They are thus able to teach themselves new concepts by finding similar signatures, he said.
Cortica’s AI technology learns in real time, said Raichelgauz, reinforcing the intelligence it already has with new concepts and situations it faces.
“Our system can learn in an unsupervised way by finding similar signatures,” said Raichelgauz. “If this object looks like a pedestrian, then it is likely to be a pedestrian.” The system develops the concept of a pedestrian, not just of a particular shape, he said.
The brains of mammals, including that of humans, are good at recognizing objects and situations and at predicting what can happen next. They can also make quick decisions regarding the best course of action to minimize risk, said Raichelgauz.
“Our system is very good at recognizing objects and at recommending what should be done next,” he said.
A predictive algorithm developed by the entrepreneurs also mimics how living mammals, including humans, learn to predict what is going to happen next based on their previous experiences. The algorithm knows what the typical reaction to certain kinds of situations can be and this enables it to create a number of reaction scenarios to events in a few seconds, and to try to avoid the negative ones. “And that is the core of autonomous driving,” which will be a key focus of the new technology, Raichelgauz said.
Medical imaging applications, such as reading CT scans, and drone applications are further markets the company will be looking at for its products, Raichelgauz said.
The computational power required for Cortica’s technology, he said, is far lower than that of other machine learning applications — just 0.5 watt vs 5-10 watts of computer power necessary for computers to learn, he said.
Additional partners for the startup’s technology are in the pipeline, Raichelgauz said. These are leading chipmakers or original equipment manufacturers in the automotive industry, but the exposure to date has been limited, he said, to ensure the company’s advantage.
The firm is already in talks with two other firms — a chip maker and an automotive chip maker — to make use of Cortica’s technology. “We want to stay an independent and neutral player to any chip and work with any sensor and any automotive company. This will make better use of our technology and not just limit it to one segment,” he said.