JAST: quickly find repeated patterns in videos, images, sound, data, with no prior learning phase

DESCRIPTION

• JAST automatically detects multiple repeating sequences of events (whatever their meaning) in complex and noisy data sets
• Based on a new neural network technology inspired by the Spike Time Dependent Plasticity (STDP) concept
• Patterns are learnt by less than 5 unlabeled presentations, e.g. a few milliseconds for a video
• Handle any kind of data: video, image, sound, signal, data, etc.
• Hardware-friendly algorithm (GPU, low-cost FPGA, ASIC)
• Can be used as smart pre-processing step to enhance other learning algorithms (classification, clustering, etc.)

APPLICATIONS

• Video/Audio surveillance
• Autonomous vehicles
• Robotics, IoT
• Finance
• Health…