Demiurge Raised $9.5 Million to Develop Deep Learning 2.0 for Mobile Robots

Demiurge Technologies AG (www.demiurge.technology) is a Switzerland-based artificial intelligence company developing the next generation of deep neural networks and brain chips for mobile robots. Demiurge has raised $9.5 million funding 6 months after its launch in May 2015, including $1.8 million angel round from Lun Feng at Vantone Holdings among three angel investors; and $7.7 million A-1 round from Hongdao Capital.

True intelligence requires a self-supervised, fully-adaptive, and always-online learning of the world model via dynamic physical agent-environment interactions through a closed sensory-motor feedback loop. Most of the envisaged AI products and applications (e.g. fully autonomous vehicles, home service robots, space exploratory robots, etc.) require realized true intelligence to deliver much cheaper solutions with much better performance under stringent constraints (power, latency, stability, etc.) in real user cases.

Deep Reinforcement Learning cannot realize true intelligence because deep learning and reinforcement learning are two wrong parts that couldn’t make a right whole. On the one hand, the neurons of deep learning are too simple to take advantage of spatiotemporal complexity for modeling objects and understanding scenes. On the other hand, the rewards of reinforcement learning are too simple to take advantage of spatiotemporal perception-action correlations for finding the optimal policy. In summary, the blindness of deep learning and the naiveness of reinforcement learning prohibit deep reinforcement learning from realizing high-performance autonomous learning via the closed sensory-motor feedback loop.

Deep Learning 2.0 with redesigned neuronshas the potential of realizing true intelligence as it is optimal for spatiotemporal pattern recognition and action selection, and it is developed from a rebuilt foundation of mathematics, physics, neuroscience and computer science. Deep Learning 2.0 in this context refers to the general physical mechanism of sensory information processing in biological neural networks.

The new world of artificial intelligence can only be discovered when they sail away from the charted island of deep learning towards the uncharted waters. Understanding directions may be unnecessary for wanderers on the island but critical for sailers in the ocean, as the race of discovery in artificial intelligence ends when it starts with wrong directions. Demiurge has set course with the compass of first principles and has maintained top speed by adopting the synthetic methodology that bridges model testing and product prototyping by iteratively building entire systems and thus could substantially reduce the time-to-market of artificial intelligence products.

It is one thing for islanders to imagine the impact of discoveries from the new world, but it is another thing for explorers to deliver the benefit of those discoveries to the entire mankind. Demiurge’s core value is to benefit all lives with true intelligence, and they are creating children-friendly workspace, pioneering family-based benefits, celebrating a communal culture, and championing meritocracy and candor.

The Demiurge family is the crew of a ship embarking on a journey towards the greatest unknowns. It takes extraordinary courage to elbow through the crowded pier, and it takes extraordinary composure to enjoy the focus from loneliness and the hope from uncertainty. All members of the Demiurge crew lead independently with head and interdependently with heart, because their life trajectories have shown enduring genuine interest in Demiurge’s mission, tailored interdisciplinary background to advance it, and tested fearless mindset to achieve it. You are most welcome to board Demiurge (send cover letter and CV to crew@demiurge.technology) if you are ready for the new Age of Discovery!