US Patent Application 17719335. OPTIMIZATION-BASED PARAMETRIC MODEL FITTING VIA DEEP LEARNING simplified abstract
Contents
OPTIMIZATION-BASED PARAMETRIC MODEL FITTING VIA DEEP LEARNING
Organization Name
Microsoft Technology Licensing, LLC
Inventor(s)
Julien Pascal Christophe Valentin of Zurich (CH)
Federica Bogo of Kilchberg (CH)
Vasileios Choutas of Zurich (CH)
Jingjing Shen of Cambridge (GB)
OPTIMIZATION-BASED PARAMETRIC MODEL FITTING VIA DEEP LEARNING - A simplified explanation of the abstract
- This abstract for appeared for US patent application number 17719335 Titled 'OPTIMIZATION-BASED PARAMETRIC MODEL FITTING VIA DEEP LEARNING'
Simplified Explanation
The abstract describes a neural optimizer that can be used for various fitting problems. It can run quickly and efficiently without needing manual input, and it incorporates information from previous iterations. The optimizer controls the learning rate of each parameter separately for better results and combines gradient descent with a method that rapidly reduces the fitting energy. A neural fitter uses this optimizer to estimate parameter values by iteratively updating an initial estimate.
Original Abstract Submitted
A neural optimizer is disclosed that is easily applicable to different fitting problems, can run at interactive rates without requiring significant efforts, does not require hand crafted priors, carries over information about previous iterations of the solve, controls the learning rate of each parameter independently for robustness and convergence speed, and combines updates from gradient descent and from a method capable of very quickly reducing the fitting energy. A neural fitter estimates the values of the parameters Θ by iteratively updating an initial estimate Θ.
