Toward Equation of Motion for Deep Neural Networks: Continuous-time Gradient Descent and Discretization Error Analysis.
Taiki MiyagawaPublished in: NeurIPS (2022)
Keyphrases
- error analysis
- neural network
- motion estimates
- least squares
- cross ratio
- optical flow
- cost function
- space time
- euclidean reconstruction
- motion estimation
- motion model
- motion parameters
- velocity field
- error correction
- image sequences
- motion field
- projective geometry
- moving objects
- objective function
- learning rules
- evolution equation
- motion analysis
- camera motion
- feature points
- human motion
- optic flow
- artificial neural networks
- humanoid robot
- activation function
- vector field
- d scene
- back propagation
- level set
- image velocity