• linkedin
  • twitter
  • youtube
  • View Cart
  • Contact Us

Applications

  • Identification and Classification
  • Novelty and Anomaly Detection
  • Prediction
  • Nearest Neighbor
  • Clustering
  • Supervised and Unsupervised Learning

From Cognitive Sensing to Cognitive Computing

The CogniMem chip can be used as companion chip to sensors enabling real-time data recognition and transmission only when the information is of interest. Leaving the arena of the low-cost, low-power embedded systems, multiple CogniMem chips can be daisy-chained to build massively parallel data mining systems with unlimited capacity and a recognition time independent from the size of the knowledge base.

Real-time Trainable Neural Network

RBF networks have attracted a high degree of interest in research communities. They have been applied to nonlinear mapping & classification, function approximation and data clustering. The CogniMem RBF network is highly adaptive and capable of real-time reinforced learning. Its hardware implementation enables the tracing of the examples which actually commit new neurons. Thanks to the parallel architecture of CogniMem, the recognition time is independent of the number of models retained by the neurons, making this powerful non-linear classifier suitable for real-time embedded systems as well as large data mining systems.

KNN in Microseconds

K Nearest Neighbor is an algorithm that is very simple and works incredibly well for closest match and pattern classification. A clever use of KNN can make things very simple for applications ranging from vision to proteins to computational geometry to data mining and so on. The downside of the algorithm is that it is highly computational, but thanks to the CogniMem parallel architecture, its execution time becomes independent of the number of trained examples.