DD-AIM has developed a patented innovative design for a digital circuit (chip) targeted at the next emerging trend in practical applications of artificial intelligence/machine-learning - PREDICTIVE AI at massive scale: thousands of simultaneous models and millions of inference runs.
Our chip is focused on the inference side of the deep learning/big data revolution -- collections of models all working together to monitor, evaluate, and forecast ("nowcast") real-world systems in real-time, generating money-saving, possibly life-saving alerts.
The chip, while packaged in an ASIC ("Application Specific Integrated Circuit" - considered to be a non-modifiable component), can self-learn, self-optimize, and self-correct hardware or model errors through novel memory management and dynamic computational techniques, while all the time communicating with on-prem or cloud-based servers to upload or download AI-driven model improvements and/or respond to data input drift and distribution changes, doing in situ retraining as necessary.
The ability to scale big-time actually unlocks innovative AI/ML techniques that can crack previously unapproachable problems.
In addition, the fundamental architecture of the chip has been dramatically simplified to bring manufacturing and deployment costs way down, focusing especially on very low energy consumption (along with small physical size/weight and low heat generation -- what the experts call SWaP-2C "Size Weight and Power - Cost/Cooling") which is a major concern of the entire AI ecosystem.
Defense: Suites of cognitive sensors have become the mainstay of electronic warfare and signals intelligence, but machine learning hardware has not kept pace. Our chip fixes that.
Using our chip, quant and algo trading firms can dramatically improve, flex, and scale up their current AI/ML models to profit from overlooked alpha.
Our innovation lets us generate more accurate and useful real-time monitors that can pinpoint manufacturing problems as well as warn of imminent failures.
Continual patient status monitoring is the new standard-of-care, but too few institutions have this fully automated. Our chip solves this, requiring human intervention only to respond to alerts. Integration with Epic and other EHR systems will be straightforward.
In a typical application, 200,000 SKUs and 2,000 stores must be analyzed, but these volumes can only be handled by today’s ML inference systems at aggregate levels. The problem is that aggregation negatively impacts accuracy. Our technology allows each SKU and store cluster to be modeled independently.