Blackswan Technologies

vision inspired next generation video coding creator of a novel video codec inspired by the human visual system, using a data-driven algorithm borrowing concepts from sparse coding/ica, and neural network autoencoders. our codec achieved compression ratios significantly better than that achievable using the itu/mpeg standard, with a road map of optimizations remaining. a small portion of the compressed features is also directly useful for image/video recognition and search applications using little additional compute and memory overhead. the technology can be also used for compressing the activations within convolutional neural networks. key technologies employed: neural networks, representation learning, sparse coding, ica, entropy coding, gpu computing, video scene categorization