Industry Problem

Artificial intelligence (AI) is “hitting a wall.” The complexity of neural networks continues to grow at an exponential rate, rapidly increasing compute requirements. This trend is not sustainable with today‘s hardware. A new solution is necessary.

Model complexity
“My view is throw it all away and start again.” Geoff Hinton – The godfather of deep learning

Many AI experts believe the industry needs to start over, acknowledging that simply building ever-larger neural networks is not the right path to generalized intelligence and predicting that, from first principles, the industry needs to push itself to discover more elegant, efficient ways to model intelligence in machines.

The Solution is ALiX™

ALiX tackles these problems head-on by addressing the root cause. Current AI training uses random sampling and other statistical methods which require many runs and a lot of training data.

AI training
ALiX reduces the complexity of the neural network and finds the optimal training solution in a single run.

By comparison, ALiX is the first AI training system to use a non-statistical training method. Proven mathematical theorems allow ALiX to find the optimal training solution in a single run and compress the size and complexity of the neural network for faster training and inferencing.

Powered by Custom RISC-V Processors

RISC AI’s custom processors power ALiX. The processors are CPUs based on the RISC-V standard that feature vector processing, for GPU-like capabilities, and nonstandard mathematical extensions. Together, these features of the RISC AI hardware form a general purpose computing platform that runs Linux operating system and provides the best acceleration of ALiX and the AI training process.