Self-Organizing Discrete Recurrent Network Digital Image Codec
AssigneeEMERSYS Co., LTD
InventorsChristopher J. Hillar, Kilian Koepsell, Ram Mehta, Jascha Sol-Dickstein
Life of this patent
Estimating the term…
Abstract
An invention based on learning a discrete recurrent neural network for a given signal domain is described. In one implementation to the domain of visual images, the method can be used to efficiently compress digital photographs and to devise a new perceptual distortion measure between images that well-matches data collected from a human psychophysics experiment. Other applications of the invention include unsupervised detection of recurrent patterns in high-dimensional data and Shannon-optimal error-correcting coding from few training examples.
Claims8
A method for processing data in signal domains comprising the steps of:
- training a fully-visible DRNN on training data which is appropriately discretized if necessary with a user-specified probabilistic model,
- using a learning rule, such as Minimum Probability Flow, to do said training, and
- using the memories of the DRNN as features of the data for various tasks of practical import, including but not limited to lossy compression and object classification.
A system for implementing the method of claim on comprising:
- an open-source software package (“HDNet”) to carry out method on signals, such as neural recordings and digital photographic images
- and a validation method for determining coding performance relative to theoretical bounds placed on its rate-distortion theory.
Dependencies are inferred from the claim text.