Abstract: Keys are arranged on a keyboard as learned during a training stage. During training, a training corpus of input symbol sequence is provided. Each unique symbol in the corpus has an associated key on the keyboard. A cost function that measures a cost of inputting the symbols of the training corpus is globally minimized. Then, the keys are arranged on the keyboard according to the globally minimized cost function. To reduced the distance a pointer must move, the keys can also be arranged in a hexagonal pattern.

Abstract: Keystrokes on a keyboard are predicted by constructing a model from a training corpus. The training corpus includes symbol sequences. The model predicts a set of symbols, where each symbol of the set continues a particular symbol sequence using variable-length subsequences of the particular symbol sequence. A particular length is chosen to maximize a probability that the predicting is correct. Keys on the keyboard are highlighted. The highlighted keys correspond to selected symbols in the set of symbols.

Abstract: A method recovers a 3D model of non-rigid 3D shape and motion of an object directly from an input video of the object by first identifying a set of features on the object in a reference image of the input video. Correspondences are then determined between the set of features in the reference image and corresponding features in each other image of the input video, These correspondences are factored by cascaded singular value decompositions into a motion matrix and a shape matrix. The 3D model can then be extracted from the factored motion matrix and shape matrix. The 3D model includes a linear basis for deformable shape of the object in the input video, and for each image a 3D rotations matrix, deformation coefficients, and translation vectors. A novel video can now be generated from the input video by manipulating the 3D model.

Abstract: A method models a non-rigid three-dimensional object directly from a sequence of images. A shape of the object is represented as a matrix of 3D points, and a basis of possible deformations of the object is represented as a matrix of displacements of the 3D points. The matrices of 3D points and displacements forming a model of the object. Evidence for an optical flow is determined from image intensities in a local region near each 3D point. The evidence is factored into 3D rotation, translation, and deformation coefficients of the model to track the object in the video.

Abstract: A method infers a target path in a target system from a cue path. The method learns a target state machine, target probability density functions and an occupancy matrix of the state machine from training target paths. Cue probability density functions are learned from a training cue path and the target occupancy matrix. A cue path is analyzed using the cue probability density functions and the target state machine to produce hidden states of the cue path. The target path is synthesized from the hidden states of the cue path and the target probability density functions.

Abstract: A method constructs a super-resolution texture from a sequence of images of a non-rigid three-dimensional object. A shape of the object is represented as a matrix of vertices, and a basis of possible deformations of the object is represented as a matrix of displacements of the 3D points, the matrices of 3D points and displacements form a model of the object in the video. A set of correspondences between the points in model and the object in the images is formed. The points in each image are connected using the set of correspondences to form a triangle texture mesh for each image. Each triangle mesh is warped to a common coordinate system while super-sampling texture in each image. The warped and super-sampled triangle meshes are averaged to produce the super-sampled texture of the object in the image.