Abstract: Systems and methods for training a neural network or an ensemble of neural networks are described. A hyper-parameter that controls the variance of the ensemble predictors is used to address overfitting. For larger values of the hyper-parameter, the predictions from the ensemble have more variance, so there is less overfitting. This technique can be applied to ensemble learning with various cost functions, structures and parameter sharing. A cost function is provided and a set of techniques for learning are described.

Abstract: Systems and methods for scoring and visualizing the effects of variants in biological sequences. Variants may include substitutions, insertions and deletions. The method comprises encoding biological sequences as vector sequences and then operating a neural network in the forward-propagation mode and possibly in the back-propagation mode to compute variant scores. Variant scores are determined by normalizing the gradients. Variant scores may be used to select a subset of variants, which are then used to produce modified vector sequences which are analyzed by the neural network operating in forward-propagation mode, to determine improved variant scores. The variant scores may be visualized using black and white, greyscale or colored elements that are arranged in blocks with dimensions corresponding to different possible symbols and the length of the sequence. These blocks are aligned with the biological sequence, which is illustrated by a symbol sequence arranged in a line.

Abstract: We describe a system and a method that ascertains the strengths of links between pairs of biological sequence variants, by determining numerical link distances that measure the similarity of the molecular phenotypes of the variants. The link distances may be used to associate knowledge about labeled variants to other variants and to prioritize the other variants for subsequent analysis or interpretation. The molecular phenotypes are determined using a neural network, called a molecular phenotype neural network, and may include numerical or descriptive attributes, such as those describing protein-DNA interactions, protein-RNA interactions, protein-protein interactions, splicing patterns, polyadenylation patterns, and microRNA-RNA interactions. Linked genetic variants may be used to ascertain pathogenicity in genetic testing, to identify drug targets, to identify patients that respond similarly to a drug, to ascertain health risks, or to connect patients that have similar molecular phenotypes.

Abstract: We describe a system and a method that ascertains the strengths of links between pairs of biological sequence variants, by determining numerical link distances that measure the similarity of the molecular phenotypes of the variants. The link distances may be used to associate knowledge about labeled variants to other variants and to prioritize the other variants for subsequent analysis or interpretation. The molecular phenotypes are determined using a neural network, called a molecular phenotype neural network, and may include numerical or descriptive attributes, such as those describing protein-DNA interactions, protein-RNA interactions, protein-protein interactions, splicing patterns, polyadenylation patterns, and microRNA-RNA interactions. Linked genetic variants may be used to ascertain pathogenicity in genetic testing, to identify drug targets, to identify patients that respond similarly to a drug, to ascertain health risks, or to connect patients that have similar molecular phenotypes.

Abstract: Systems and methods for training a neural network or an ensemble of neural networks are described. A hyper-parameter that controls the variance of the ensemble predictors is used to address overfitting. For larger values of the hyper-parameter, the predictions from the ensemble have more variance, so there is less overfitting. This technique can be applied to ensemble learning with various cost functions, structures and parameter sharing. A cost function is provided and a set of techniques for learning are described.

Abstract: Systems and methods for training a neural network or an ensemble of neural networks are described. A hyper-parameter that controls the variance of the ensemble predictors is used to address overfitting. For larger values of the hyper-parameter, the predictions from the ensemble have more variance, so there is less overfitting. This technique can be applied to ensemble learning with various cost functions, structures and parameter sharing. A cost function is provided and a set of techniques for learning are described.

Abstract: Systems and methods for training a neural network or an ensemble of neural networks are described. A hyper-parameter that controls the variance of the ensemble predictors is used to address overfitting. For larger values of the hyper-parameter, the predictions from the ensemble have more variance, so there is less overfitting. This technique can be applied to ensemble learning with various cost functions, structures and parameter sharing. A cost function is provided and a set of techniques for learning are described.

Abstract: Systems and methods for training a neural network or an ensemble of neural networks are described. A hyper-parameter that controls the variance of the ensemble predictors is used to address overfitting. For larger values of the hyper-parameter, the predictions from the ensemble have more variance, so there is less overfitting. This technique can be applied to ensemble learning with various cost functions, structures and parameter sharing. A cost function is provided and a set of techniques for learning are described.

Abstract: Systems and methods for scoring and visualizing the effects of variants in biological sequences. Variants may include substitutions, insertions and deletions. The method comprises encoding biological sequences as vector sequences and then operating a neural network in the forward-propagation mode and possibly in the back-propagation mode to compute variant scores. Variant scores are determined by normalizing the gradients. Variant scores may be used to select a subset of variants, which are then used to produce modified vector sequences which are analyzed by the neural network operating in forward-propagation mode, to determine improved variant scores. The variant scores may be visualized using black and white, greyscale or colored elements that are arranged in blocks with dimensions corresponding to different possible symbols and the length of the sequence. These blocks are aligned with the biological sequence, which is illustrated by a symbol sequence arranged in a line.

Abstract: We describe a system and a method that ascertains the strengths of links between pairs of biological sequence variants, by determining numerical link distances that measure the similarity of the molecular phenotypes of the variants. The link distances may be used to associate knowledge about labeled variants to other variants and to prioritize the other variants for subsequent analysis or interpretation. The molecular phenotypes are determined using a neural network, called a molecular phenotype neural network, and may include numerical or descriptive attributes, such as those describing protein-DNA interactions, protein-RNA interactions, protein-protein interactions, splicing patterns, polyadenylation patterns, and microRNA-RNA interactions. Linked genetic variants may be used to ascertain pathogenicity in genetic testing, to identify drug targets, to identify patients that respond similarly to a drug, to ascertain health risks, or to connect patients that have similar molecular phenotypes.

Abstract: Systems and methods for training a neural network or an ensemble of neural networks are described. A hyper-parameter that controls the variance of the ensemble predictors is used to address overfitting. For larger values of the hyper-parameter, the predictions from the ensemble have more variance, so there is less overfitting. This technique can be applied to ensemble learning with various cost functions, structures and parameter sharing. A cost function is provided and a set of techniques for learning are described.

Abstract: A region-based push-relabel formulation is disclosed that removes the requirement that the entire graph should fit into the computer memory and yields an implementation that can reduce the required size and redundancy of accesses to the data memory, thus improving speed performance, while allowing for an efficient parallel processing implementation. The algorithm assigns all vertices that are not part of the sources or sinks with a value of 1. Sinks are assigned with zeros and sources are assigned a label equal to the number of their vertices. The preflow is then pushed from the sources to their neighbors, if any. When the preflow has all reached the boundaries, an adjacent region of the neighboring set is selected and preflow is pushed within this region. When the values of the preflow have been exhausted, region relabeling is done to update the label values. This is repeated within the region until all preflow has exited to the boundary of this region.

Abstract: A region-based push-relabel formulation is disclosed that removes the requirement that the entire graph should fit into the computer memory and yields an implementation that can reduce the required size and redundancy of accesses to the data memory, thus improving speed performance, while allowing for an efficient parallel processing implementation. The algorithm assigns all vertices that are not part of the sources or sinks with a value of 1. Sinks are assigned with zeros and sources are assigned a label equal to the number of their vertices. The preflow is then pushed from the sources to their neighbors, if any. When the preflow has all reached the boundaries, an adjacent region of the neighboring set is selected and preflow is pushed within this region. When the values of the preflow have been exhausted, region relabeling is done to update the label values. This is repeated within the region until all preflow has exited to the boundary of this region.

Abstract: A high throughput, compact image display apparatus comprises a source unit to generate a spatially uniform light beam that propagates along a first beam path in a first direction. The image display apparatus also comprises an imaging unit to collect and focus the light beam. The imaging unit includes a first refractive optical element disposed in the first beam path, a second refractive optical element disposed in the first beam path, a first reflecting mirror disposed in the first beam path, and a second reflecting mirror, having a concave reflecting surface, disposed in a second beam path defined by the first reflecting mirror and the second reflecting mirror. The second beam path is oriented in a second direction different from the first direction. The image display apparatus also comprises a digital micromirror device (“DMD”) to receive the light beam reflected by the second reflecting mirror. The DMD is disposed in a third beam path defined by the second reflecting mirror and the DMD.

Abstract: A high throughput, compact image display apparatus comprises a source unit to generate a spatially uniform light beam that propagates along a first beam path in a first direction. The image display apparatus also comprises an imaging unit to collect and focus the light beam. The imaging unit includes a first refractive optical element disposed in the first beam path, a second refractive optical element disposed in the first beam path, a first reflecting mirror disposed in the first beam path, and a second reflecting mirror, having a concave reflecting surface, disposed in a second beam path defined by the first reflecting mirror and the second reflecting mirror. The second beam path is oriented in a second direction different from the first direction. The image display apparatus also comprises a digital micromirror device (“DMD”) to receive the light beam reflected by the second reflecting mirror. The DMD is disposed in a third beam path defined by the second reflecting mirror and the DMD.