Patents by Inventor Balaji Lakshminarayanan
Balaji Lakshminarayanan has filed for patents to protect the following inventions. This listing includes patent applications that are pending as well as patents that have already been granted by the United States Patent and Trademark Office (USPTO).
-
Patent number: 11954902Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a final classification output for an image of eye tissue. The image is provided as input to each of one or more segmentation neural networks to obtain one or more segmentation maps of the eye tissue in the image. A respective classification input is generated from each of the segmentation maps. For each of the segmentation maps, the classification input for the segmentation map is provided as input to each of one or more classification neural networks to obtain, for each segmentation map, a respective classification output from each classification neural network. A final classification output for the image is generated from the respective classification outputs for each of the segmentation maps.Type: GrantFiled: December 8, 2020Date of Patent: April 9, 2024Assignee: Google LLCInventors: Jeffrey De Fauw, Joseph R. Ledsam, Bernardino Romera-Paredes, Stanislav Nikolov, Nenad Tomasev, Samuel Blackwell, Harry Askham, Xavier Glorot, Balaji Lakshminarayanan, Trevor Back, Mustafa Suleyman, Pearse A. Keane, Olaf Ronneberger, Julien Robert Michel Cornebise
-
Publication number: 20230244912Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for learning from delayed outcomes using neural networks. One of the methods includes receiving an input observation; generating, from the input observation, an output label distribution over possible labels for the input observation at a final time, comprising: processing the input observation using a first neural network configured to process the input observation to generate a distribution over possible values for an intermediate indicator at a first time earlier than the final time; generating, from the distribution, an input value for the intermediate indicator; and processing the input value for the intermediate indicator using a second neural network configured to process the input value for the intermediate indicator to determine the output label distribution over possible values for the input observation at the final time; and providing an output derived from the output label distribution.Type: ApplicationFiled: April 6, 2023Publication date: August 3, 2023Inventors: Huiyi Hu, Ray Jiang, Timothy Arthur Mann, Sven Adrian Gowal, Balaji Lakshminarayanan, András György
-
Patent number: 11714994Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for learning from delayed outcomes using neural networks. One of the methods includes receiving an input observation; generating, from the input observation, an output label distribution over possible labels for the input observation at a final time, comprising: processing the input observation using a first neural network configured to process the input observation to generate a distribution over possible values for an intermediate indicator at a first time earlier than the final time; generating, from the distribution, an input value for the intermediate indicator; and processing the input value for the intermediate indicator using a second neural network configured to process the input value for the intermediate indicator to determine the output label distribution over possible values for the input observation at the final time; and providing an output derived from the output label distribution.Type: GrantFiled: March 11, 2019Date of Patent: August 1, 2023Assignee: DeepMind Technologies LimitedInventors: Huiyi Hu, Ray Jiang, Timothy Arthur Mann, Sven Adrian Gowal, Balaji Lakshminarayanan, András György
-
Publication number: 20220253747Abstract: The present disclosure is directed to systems and method to perform improved detection of out-of-distribution (OOD) inputs. In particular, current deep generative model-based approaches for OOD detection are significantly negatively affected by and struggle to distinguish population level background statistics from semantic content relevant to the in-distribution examples. In fact, such approaches have even been experimentally observed to assign higher likelihood to OOD inputs, which is opposite to the desired behavior. To resolve this problem, the present disclosure proposes a likelihood ratio method for deep generative models which effectively corrects for these confounding background statistics.Type: ApplicationFiled: May 26, 2020Publication date: August 11, 2022Inventors: Jie Ren, Balaji Lakshminarayanan, Peter Junteng Liu, Joshua Vincent Dillon, Roland Jasper Snoek, Ryan Poplin, Mark Andrew DePristo, Emily Amanda Fertig
-
Publication number: 20220108220Abstract: Example aspects of the present disclosure are directed to systems and methods for performing automatic label smoothing of augmented training data. In particular, some example implementations of the present disclosure which in some instances can be referred to “AutoLabel” can automatically learn the labels for augmented data based on the distance between the clean distribution and augmented distribution. AutoLabel is built on label smoothing and is guided by the calibration-performance over a hold-out validation set. AutoLabel is a generic framework that can be easily applied to existing data augmentation methods, including AugMix, mixup, and adversarial training, among others. AutoLabel can further improve clean accuracy, as well as the accuracy and calibration over corrupted datasets. Additionally, AutoLabel can help adversarial training by bridging the gap between clean accuracy and adversarial robustness.Type: ApplicationFiled: October 4, 2021Publication date: April 7, 2022Inventors: Yao Qin, Alex Beutel, Ed Huai-Hsin Chi, Xuezhi Wang, Balaji Lakshminarayanan
-
Publication number: 20210118198Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a final classification output for an image of eye tissue. The image is provided as input to each of one or more segmentation neural networks to obtain one or more segmentation maps of the eye tissue in the image. A respective classification input is generated from each of the segmentation maps. For each of the segmentation maps, the classification input for the segmentation map is provided as input to each of one or more classification neural networks to obtain, for each segmentation map, a respective classification output from each classification neural network. A final classification output for the image is generated from the respective classification outputs for each of the segmentation maps.Type: ApplicationFiled: December 8, 2020Publication date: April 22, 2021Inventors: Jeffrey De Fauw, Joseph R. Ledsam, Bernardino Romera-Paredes, Stanislav Nikolov, Nenad Tomasev, Samuel Blackwell, Harry Askham, Xavier Glorot, Balaji Lakshminarayanan, Trevor Back, Mustafa Suleyman, Pearse A. Keane, Olaf Ronneberger, Julien Robert Michel Cornebise
-
Patent number: 10878601Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a final classification output for an image of eye tissue. The image is provided as input to each of one or more segmentation neural networks to obtain one or more segmentation maps of the eye tissue in the image. A respective classification input is generated from each of the segmentation maps. For each of the segmentation maps, the classification input for the segmentation map is provided as input to each of one or more classification neural networks to obtain, for each segmentation map, a respective classification output from each classification neural network. A final classification output for the image is generated from the respective classification outputs for each of the segmentation maps.Type: GrantFiled: December 28, 2018Date of Patent: December 29, 2020Assignee: Google LLCInventors: Jeffrey De Fauw, Joseph R. Ledsam, Bernardino Romera-Paredes, Stanislav Nikolov, Nenad Tomasev, Samuel Blackwell, Harry Askham, Xavier Glorot, Balaji Lakshminarayanan, Trevor Back, Mustafa Suleyman, Pearse A. Keane, Olaf Ronneberger, Julien Robert Michel Cornebise
-
Publication number: 20190279076Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for learning from delayed outcomes using neural networks. One of the methods includes receiving an input observation; generating, from the input observation, an output label distribution over possible labels for the input observation at a final time, comprising: processing the input observation using a first neural network configured to process the input observation to generate a distribution over possible values for an intermediate indicator at a first time earlier than the final time; generating, from the distribution, an input value for the intermediate indicator; and processing the input value for the intermediate indicator using a second neural network configured to process the input value for the intermediate indicator to determine the output label distribution over possible values for the input observation at the final time; and providing an output derived from the output label distribution.Type: ApplicationFiled: March 11, 2019Publication date: September 12, 2019Inventors: Huiyi Hu, Ray Jiang, Timothy Arthur Mann, Sven Adrian Gowal, Balaji Lakshminarayanan, Andras Gyorgy
-
Publication number: 20190139270Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a final classification output for an image of eye tissue. The image is provided as input to each of one or more segmentation neural networks to obtain one or more segmentation maps of the eye tissue in the image. A respective classification input is generated from each of the segmentation maps. For each of the segmentation maps, the classification input for the segmentation map is provided as input to each of one or more classification neural networks to obtain, for each segmentation map, a respective classification output from each classification neural network. A final classification output for the image is generated from the respective classification outputs for each of the segmentation maps.Type: ApplicationFiled: December 28, 2018Publication date: May 9, 2019Inventors: Jeffrey De Fauw, Joseph R. Ledsam, Bernardino Romera-Paredes, Stanislav Nikolov, Nenad Tomasev, Samuel Blackwell, Harry Askham, Xavier Glorot, Balaji Lakshminarayanan, Trevor Back, Mustafa Suleyman, Pearse A. Keane, Olaf Ronneberger, Julien Robert Michel Cornebise
-
Patent number: 10198832Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a final classification output for an image of eye tissue. The image is provided as input to each of one or more segmentation neural networks to obtain one or more segmentation maps of the eye tissue in the image. A respective classification input is generated from each of the segmentation maps. For each of the segmentation maps, the classification input for the segmentation map is provided as input to each of one or more classification neural networks to obtain, for each segmentation map, a respective classification output from each classification neural network. A final classification output for the image is generated from the respective classification outputs for each of the segmentation maps.Type: GrantFiled: June 28, 2018Date of Patent: February 5, 2019Inventors: Jeffrey De Fauw, Joseph R. Ledsam, Bernardino Romera-Paredes, Stanislav Nikolov, Nenad Tomasev, Samuel Blackwell, Harry Askham, Xavier Glorot, Balaji Lakshminarayanan, Trevor Back, Mustafa Suleyman, Pearse A. Keane, Olaf Ronneberger, Julien Robert Michel Cornebise
-
Publication number: 20190005684Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a final classification output for an image of eye tissue. The image is provided as input to each of one or more segmentation neural networks to obtain one or more segmentation maps of the eye tissue in the image. A respective classification input is generated from each of the segmentation maps. For each of the segmentation maps, the classification input for the segmentation map is provided as input to each of one or more classification neural networks to obtain, for each segmentation map, a respective classification output from each classification neural network. A final classification output for the image is generated from the respective classification outputs for each of the segmentation maps.Type: ApplicationFiled: June 28, 2018Publication date: January 3, 2019Inventors: Jeffrey De Fauw, Joseph R. Ledsam, Bernardino Romera-Paredes, Stanislav Nikolov, Nenad Tomasev, Samuel Blackwell, Harry Askham, Xavier Glorot, Balaji Lakshminarayanan, Trevor Back, Mustafa Suleyman, Pearse A. Keane, Olaf Ronneberger, Julien Robert Michel Cornebise
-
Patent number: 9768731Abstract: Described embodiments provide a radio frequency (RF) amplifier system having at least one amplifier. The at least one amplifier includes an RF input port, an RF output port and a drain bias port. At least one voltage modulator is coupled to the bias port of the least one amplifier to provide a bias voltage. The bias voltage is selected by switching among a plurality of discrete voltages. At least one filter circuit is coupled between the at least one voltage modulator and the at least one amplifier. The at least one filter circuit controls spectral components resultant from transitions in the bias voltage when switching among the plurality of discrete voltages. A controller dynamically adapts at least one setting of the at least one voltage modulator by using multi-pulse transitions when switching among the plurality of discrete voltages for a first operating condition of the RF amplifier.Type: GrantFiled: December 14, 2015Date of Patent: September 19, 2017Assignee: Eta Devices, Inc.Inventors: David J. Perreault, Joel L. Dawson, Wei Li, Yevgeniy A. Tkachenko, Balaji Lakshminarayanan, John Hoversten
-
Publication number: 20160099686Abstract: Described embodiments provide a radio frequency (RF) amplifier system having at least one amplifier. The at least one amplifier includes an RF input port, an RF output port and a drain bias port. At least one voltage modulator is coupled to the bias port of the least one amplifier to provide a bias voltage. The bias voltage is selected by switching among a plurality of discrete voltages. At least one filter circuit is coupled between the at least one voltage modulator and the at least one amplifier. The at least one filter circuit controls spectral components resultant from transitions in the bias voltage when switching among the plurality of discrete voltages. A controller dynamically adapts at least one setting of the at least one voltage modulator by using multi-pulse transitions when switching among the plurality of discrete voltages for a first operating condition of the RF amplifier.Type: ApplicationFiled: December 14, 2015Publication date: April 7, 2016Applicant: Eta Devices, Inc.Inventors: David J. Perreault, Joel L. Dawson, Wei Li, Yevgeniy A. Tkachenko, Balaji Lakshminarayanan, John Hoversten
-
Patent number: 8880439Abstract: In a recommender method, Bayesian Matrix Factorization (BMF) is performed on a matrix having user and item dimensions and matrix elements containing user ratings for items made by users in order to train a probabilistic collaborative filtering model. A recommendation is generated for a user using the probabilistic collaborative filtering model. The recommendation may comprise a predicted item rating, or an identification of one or more recommended items. The recommender method is suitably performed by an electronic data processing device. The BMF may employ non-Gaussian priors, such as Student-t priors. The BMF may additionally or alternatively employ a heteroscedastic noise model comprising priors that include (1) a row dependent variance component that depends upon the matrix row and (2) a column dependent variance component that depends upon the matrix column.Type: GrantFiled: February 27, 2012Date of Patent: November 4, 2014Assignee: Xerox CorporationInventors: Cedric Archambeau, Guillaume Bouchard, Balaji Lakshminarayanan
-
Publication number: 20130226839Abstract: In a recommender method, Bayesian Matrix Factorization (BMF) is performed on a matrix having user and item dimensions and matrix elements containing user ratings for items made by users in order to train a probabilistic collaborative filtering model. A recommendation is generated for a user using the probabilistic collaborative filtering model. The recommendation may comprise a predicted item rating, or an identification of one or more recommended items. The recommender method is suitably performed by an electronic data processing device. The BMF may employ non-Gaussian priors, such as Student-t priors. The BMF may additionally or alternatively employ a heteroscedastic noise model comprising priors that include (1) a row dependent variance component that depends upon the matrix row and (2) a column dependent variance component that depends upon the matrix column.Type: ApplicationFiled: February 27, 2012Publication date: August 29, 2013Applicant: Xerox CorporationInventors: Cedric Archambeau, Guillaume Bouchard, Balaji Lakshminarayanan
-
Patent number: 7741936Abstract: The present invention provides a monolithic inductor developed using radio frequency micro electromechanical (RF MEMS) techniques. In a particular embodiment of the present invention, a tunable radio frequency microelectromechanical inductor includes a coplanar waveguide and a direct current actuatable contact switch positioned to vary the effective width of a narrow inductive section of the center conductor of the CPW line upon actuation the DC contact switch. In a specific embodiment of the present invention, the direct current actuatable contact switch is a diamond air-bridge integrated on an alumina substrate to realize an RF switch in the CPW and microstrip topology.Type: GrantFiled: September 4, 2007Date of Patent: June 22, 2010Assignee: University of South FloridaInventors: Thomas Weller, Balaji Lakshminarayanan, Srinath Balachandran
-
Patent number: 7676903Abstract: /The present invention provides a method of use for a monolithic device utilizing cascaded, switchable slow-wave CPW sections that are integrated along the length of a planar transmission line. The purpose of the switchable slow-wave CPW sections element is to enable control of the propagation constant along the transmission line while maintaining a quasi-constant characteristic impedance. The method can be used to produce true time delay phase shifting components in which large amounts of time delay can be achieved without significant variation in the effective characteristic impedance of the transmission line, and thus also the input/output return loss of the component. Additionally, for a particular value of return loss, greater time delay per unit length can be achieved in comparison to tunable capacitance-only delay components.Type: GrantFiled: July 25, 2007Date of Patent: March 16, 2010Assignee: University of South FloridaInventors: Thomas Weller, Balaji Lakshminarayanan
-
Patent number: 7274278Abstract: The present invention provides a monolithic inductor developed using radio frequency micro electromechanical (RF MEMS) techniques. In a particular embodiment of the present invention, a tunable radio frequency microelectromechanical inductor includes a coplanar waveguide and at least one direct current actuatable contact switch positioned to vary the effective width of a narrow inductive section of the center conductor of the CPW line upon actuation the DC contact switch.Type: GrantFiled: September 9, 2005Date of Patent: September 25, 2007Assignee: University of South FloridaInventors: Thomas Weller, Balaji Lakshminarayanan, Srinath Balachandran
-
Patent number: 7269810Abstract: The present invention is a substrate dependent circuit modeling system for substrate-mounted components. The height and dielectric constant of a substrate have a significant impact on the frequency response of such components, and these effects cannot be treated independently from the circuit model. The equivalent circuit parameters in the model must be made to vary in accordance with changes in the substrate.Type: GrantFiled: October 18, 2005Date of Patent: September 11, 2007Assignee: University of South FloridaInventors: Thomas Weller, John Capwell, Horace Gordon, Balaji Lakshminarayanan
-
Patent number: 7259641Abstract: The present invention provides a method and apparatus for a monolithic device utilizing cascaded, switchable slow-wave CPW sections that are integrated along the length of a planar transmission line. The purpose of the switchable slow-wave CPW sections elements is to enable control of the propagation constant along the transmission line while maintaining a quasi-constant characteristic impedance. The device can be used to produce true time delay phase shifting components in which large amounts of time delay can be achieved without significant variation in the effective characteristic impedance of the transmission line, and thus also the input/output return loss of the component. Additionally, for a particular value of return loss, greater time delay per unit length can be achieved in comparison to tunable capacitance-only delay components.Type: GrantFiled: February 28, 2005Date of Patent: August 21, 2007Assignee: University of South FloridaInventors: Thomas Weller, Balaji Lakshminarayanan