International Business Machines Corporation

Korea Advanced Institute of Science and Technology

Lg Chem, Ltd

Inseok Hwang

Su Liu

Eric Rozner

Chungkuk Yoo

Chin Ngai Sze

Yu Gu

Yaoguang Wei

Jinho Lee

Gelareh Taban

Myung-Chul Kim

Janani Mukundan

Kanak Behari Agarwal

Junehwa Song

Youngki Lee

Jente Benedict Kuang

Christopher McCall Durham

Catherine Helen Crawford

Matthew Harrison Tong

Chulhong Min

Younghyun Ju

John A Gunnels

Archan Misra

Tomasz J Nowicki

Grzegorz Michal Swirszcz

Bumsoo Kang

Miri Moon

Han Qi

Cheng Xu

Minsik Cho

Anne Elizabeth Gattiker

Michael D Kistler

Seokhyun Yoon

Aisha Walcott

Dongwook Lee

Thomas S Hubregtsen

Frank Liu

Reginald Bryant

Quan Wen Zhang

Uichin Lee

Cliff Chin Ngai Sze

Seungwoo Kang

SeungChul Lee

Jihyun Yu

Chanyou Hwang

Jaeung Lee

Yuhwan Kim

Hyunho Ro

Sangki Chun

Jiyoung Hwang

Sangwon Choi

Pillsoon Park

Seungpyo Choi

A method identifies a type of object in a digital image. A user and/or one or more processors selects, from a plurality of partially overlapping sub-images of a digital image, a first sub-image and a second sub-image that overlap one another. The user/processors input the first sub-image into a neural network to create a first inference result that includes an overlapping inference result, for the overlapping area, that recognizes a partial portion of a specific type of object based on the overlapping area. The user/processors infer that the second sub-image creates a second inference result that recognizes a second portion of the specific type of object in the second sub-image based on the second sub-image and the overlapping inference result. The neural network identifies the specific type of object in the digital image based on the first and second sub-images being sub-images of a same type of object.

Identifying a type of object in a digital image based on overlapping areas of sub-images

A method for generating a musical composition based on user input is described. A first set of musical characteristics from a first input musical piece is received as an input vector. The first set of musical characteristics is perturbed to create a perturbed input vector as input in a first set of nodes in a first visible layer of an unsupervised neural net. The unsupervised neural net comprised of a plurality of computing layers, each computing layer composed of a respective set of nodes. The unsupervised neural net is operated to calculate an output vector from a higher level hidden layer in the unsupervised neural net. The output vector is used to create an output musical piece.

Cognitive music engine using unsupervised learning

A set of partial words is received. At least one partial word in the set of partial words is completed. The set of partial words with the at least one completed partial word is run through a trained deep neural network, the trained deep neural network inferring a word embedding associated with an unfinished word in the set of partial words. An inferred word is determined based on the inferred word embedding associated with the unfinished word. A sentence may be output, which includes at least the completed partial word and the inferred word.

Automatic sentence inferencing network

A method, computer program product, and a system where a processor(s), monitors multilingual switches performed on a client on behalf of a given user. Based on the monitoring, the processor(s) identifies switch patterns of the given user to generate a service profile for the user of machine learned multilingual switch patterns for the given user. The processor(s) determines a priority order for languages comprising the voice input streams, for the given user. The processor(s) obtains a new translation request initiated by the client, on behalf of the given user and applies the priority order to identify one or more languages spoken in a voice input stream of the new translation request. The processor(s) transmits indicators of the identified one or more languages to the client, where upon receiving the indicators, the client translates the voice input stream from the identified one or more languages to one or more target languages.

Dynamic multilingual speech recognition

A computer-implemented method comprises: inputting into an autoencoder sets of input samples, each of the sets of input samples comprising: a reference input sample of a reference dataset and one or more target input samples of one or more target datasets, the autoencoder being trained using the reference dataset. The autoencoder generates a respective set of outputs for each set of the input samples to thereby form one or more respective sets of outputs, each of the one or more sets of outputs comprising the reference output and the one or more target outputs for a respective set of input samples; and determining the similarity of each of the one or more target datasets to the reference dataset by comparing each of the one or more target outputs to respective target input samples of each of the sets of input samples.

Method to measure similarity of datasets for given AI task

Methods and systems for opportunistic multi-party reminders based on sensory data are provided. A system for providing opportunistic multi-party reminders based on sensory data may include a coordination module that coordinates user schedules for a plurality of parties. Also, the system may include a time prediction module that characterizes one or more times in a user schedule in response to gathered context information for one or more parties in the plurality of parties. Further, the system may include a task completion module that identifies at least one time in the one or more times in a user schedule for completing a task in response to the characterization of the one or more times.

Opportunistic multi-party reminders based on sensory data

A method optimizes Convolutional Neural Network (CNN) inference time for full resolution images. One or more processors divide a full resolution image into a plurality of partially overlapping sub-images. The processor(s) select, from the plurality of partially overlapping sub-images, a first sub-image and a second sub-image that overlap one another in an overlapping area. The processor(s) feed the first sub-image, including the overlapping area, into a Convolutional Neural Network (CNN) in order to create a first inference result for the first sub-image, where the CNN has been trained at a fine resolution. The processor(s) cache an inference result from the CNN for the overlapping area, and then utilize the cached inference result when inferring the second sub-image in the CNN. The processor(s) then identify a specific object in the full resolution image based on inferring the first sub-image and the second sub-image.

Overlapping CNN cache reuse in high resolution and streaming-based deep learning inference engines

Methods, systems and computer program products for operating a secure speech recognition are provided. Aspects include receiving a speech recognition request from a user, the speech recognition request including a captured audio file and filtering the captured audio file to create an in-band audio file and at least one out-of-band audio file. Aspects also include creating an in-band text file based on the in-band audio file and creating an out-of-band text file based on the out-of-band audio file. Aspects further include calculating an attack risk based at least in part on the in-band text file and the out-of-band text file. Based on a determination that the attack risk is below a threshold value, aspects include executing the speech recognition request.

Secure speech recognition

A movement notification is received from a display. The movement notification corresponds to a movement. A set of movement frames that reflect the movement is rendered by a graphical processing unit. The set of movement frames is transmitted to the display. A notification that the movement has ceased is received from the display. A stop command is transmitted to the display. The stop command instructs the display to not display the first subset of frames in the set of movement frames. The first subset of frames reflects the end of the movement.

Mitigating latency-based display delays

An augmented reality system collects sensory data from one or more user devices in a manner that is transparent to the user's augmented reality experience. A customer of an augmented reality system requests sensory data at a specified location. The augmented reality system modifies the augmented reality experience for one or more users to move someone to the specified location. The user device(s) collect the sensory data at the specified location in a manner that is transparent to the augmented reality experience of the user(s). The sensory data is uploaded from the user device(s) to the augmented reality system, which sends the sensory data to the customer.

Sensory data collection in an augmented reality system

A first method includes identifying a potential emoji annotation for a selected context within a text source, validating the potential emoji, determining whether the validated emoji annotation is within an emoji annotation list for the selected context, and adding the validated emoji annotation to the emoji annotation list for the selected context responsive to determining that the validated emoji annotation is not within the emoji annotation list for the selected context. A second method includes adding an emoji annotation written in a first language to an emoji annotation list corresponding to the first language, translating the emoji annotation written in the first language to a second language to produce an emoji annotation written in the second language, and adding the emoji annotation written in the second language to an emoji annotation list corresponding to the second language. Corresponding apparatii, computer program products and computer systems are also disclosed herein.

Tracking and managing emoji annotations

Interactive augmented reality system is provided. A first geographical location and a first environmental characteristic is monitored by a user device. A viewpoint of the user device is determined based, at least in part, on (i) the first geographical location and (ii) the first environmental characteristic. A digital image request is generated based, at least in part, on (i) the first geographical location and (ii) the first environmental characteristic. A digital image is identified from a database based, at least in part, on (i) a second geographical location and (ii) a second environmental characteristic associated with the digital image request. In response to identifying the digital image based, at least in part, on the digital image request populating the user device with the digital image.

Virtual view-window

In one embodiment, a method includes modifying at least one aspect of a notification to be displayed on a first portion of a touchscreen of a device in response to a determination that a second portion of the touchscreen includes and/or overlaps with the first portion of the touchscreen. The second portion of the touchscreen os defined by a point on the touchscreen expected to be imminently touched by a user and predetermined areas of the touchscreen adjacent to and surrounding the point.

Adaptive notification modifications for touchscreen interfaces

In one embodiment, a computer program product includes a computer readable storage medium having program instructions embodied therewith. The program instructions are executable by a processing circuit to cause the processing circuit to receive collected data from one or more data collection devices, the collected data being aggregated over at least one month and comprising audio data including first voice input of a user of the one or more data collection devices. The program instructions also cause the processing circuit to store the audio data to a computer readable storage medium; determine an identity of the user based on comparing second voice input to the first voice input; and analyze the audio data for indications of hearing loss in the user over the period of time. The analysis includes determining a user's emotion during production of the audio input.

Daily cognitive monitoring of early signs of hearing loss

A processor may initiate a first verification event. The first verification event may be triggered when a mobile device begins generating new data at a first time. The processor may determine, in response to the first verification event, that the mobile device is above a threshold limit of available storage. The processor may analyze one or more applications on the mobile device. The processor may automatically remove a first re-downloadable portion of a first application binary interface associated with a first application from the mobile device. The processor may store the generated new data on the mobile device.

Instant storage reclamation ensuring uninterrupted media recording

Growing community of inventors

Austin, TX, United States of America

Inseok Hwang