4 ms·

At GrAI Matter Labs we are working on next gen non-von Neumann brain inspired computing assuring post-Moore performance scalability. This technolog is only mov

At GrAI Matter Labs we are working on next gen non-von Neumann brain inspired computing assuring post-Moore performance scalability. This technolog is only moving now from research to production and has been labeled as transformational by Gartner. Our technology fuses neuro science and computer science in one architecture that is both, trainable and fully programmable. And yes, we are hiring. https://www.graimatterlabs.ai https://www.graimatterlabs.ai

relaunched 7y ago
"labeled as transformational by Gartner" According to wikipedia, there are 64 magic quadrants with roughly 10-25 companies per quadrant. This type of marketing is more of a detractor than anything else, especially on HN. It's much more helpful to provide technical / product related information to substantiate your claim. If you are working on a hard problem, describe it. The more detail, the better.
- cg94301 7y ago
  The hard problem is to provide processor performance scalability in a post-Moore world. We are tackling this by event based, near-memory, sparse computing. Traits that can be just as well attributed to the brain. The challenge here is that a whole new set of algorithms is required. A good illustration for that are DVS (Dynamic Vision Sensors). They produce a continuous stream of asynchronous events, just like our eyes, and algorithms like object detection have to work on this stream of events rather than a simple sequence of frames. The advantages are ultra-low latency, high energy efficient computing. Only what needs to be computed is computed, asynchronously. Since DVS is not a very widespread technology (yet) we do frame-to-event conversion to tap into the energy efficiency of sparse, near-memory compute. All of this processing can be done on the same architecture, since it can be explicitly programmed to do e.g. frame-to-event conversion, and trained to do inference on objects in the resulting stream.

relaunched 7y ago

"labeled as transformational by Gartner" According to wikipedia, there are 64 magic quadrants with roughly 10-25 companies per quadrant. This type of marketing is more of a detractor than anything else, especially on HN. It's much more helpful to provide technical / product related information to substantiate your claim. If you are working on a hard problem, describe it. The more detail, the better.

cg94301 7y ago
The hard problem is to provide processor performance scalability in a post-Moore world. We are tackling this by event based, near-memory, sparse computing. Traits that can be just as well attributed to the brain. The challenge here is that a whole new set of algorithms is required. A good illustration for that are DVS (Dynamic Vision Sensors). They produce a continuous stream of asynchronous events, just like our eyes, and algorithms like object detection have to work on this stream of events rather than a simple sequence of frames. The advantages are ultra-low latency, high energy efficient computing. Only what needs to be computed is computed, asynchronously. Since DVS is not a very widespread technology (yet) we do frame-to-event conversion to tap into the energy efficiency of sparse, near-memory compute. All of this processing can be done on the same architecture, since it can be explicitly programmed to do e.g. frame-to-event conversion, and trained to do inference on objects in the resulting stream.