Microsoft Technology Licensing, LLC

Frank Torsten Bernd Seide

Li Deng

Gang Li

Dong Yu

Albert Joseph Kishan Thambiratnam

Alejandro Acero

Xuedong D Huang

Lie Lu

Gabriel J White

Hagai Thomas Attias

Asela J R Gunawardana

Jian-Iai Zhou

James G Droppo

Qifa Ke

David Andrew Howell

Jian-Lai Zhou

Adam Clyde Eversole

John C Langford

Sridhar Sundararaman

Ray A Bittner, Jr

David T Fong

Roger Peng Yu

Ryota Tomioka

Xie Chen

Bruno S Bozza

Albert J K Thambiratnam

Qi Liu

Rajanala Sai Krishna Sravanthi

Timo Pascal Mertens

Weiwu Zhu

Wilhelm Richert

Sha Meng

John Langford

James Droppo

Examples of the present disclosure describe systems and methods relating to generating relevance scores for one or more words of a passage which is an answer to a natural language query. For instance, a passage extracted from a highly relevant electronic file along with the query may encoded and augmented to generate a multi-dimensional, augmented semantic vectors using recurring neural networks. The augmented semantic vectors along with a multi-dimensional vector that represent words of the passage may be decoded to generate relevance scores for one or more words of the passage, based on levels of relevance to the query.

Method and system for highlighting answer phrases

Techniques and constructs can reduce the time required to determine solutions to optimization problems such as training of neural networks. Modifications to a computational model can be determined by a plurality of nodes operating in parallel. Quantized modification values can be transmitted between the nodes to reduce the volume of data to be transferred. The quantized values can be as small as one bit each. Quantization-error values can be stored and used in quantizing subsequent modifications. The nodes can operate in parallel and overlap computation and data transfer to further reduce the time required to determine solutions. The quantized values can be partitioned and each node can aggregate values for a corresponding partition.

Computing system for training neural networks

Disclosed in some examples are methods, systems, machine-readable media, and devices which operate a neural network defined by user code. A method includes identifying, operations from user code that are integral in operating the neural network, combining a subset of the identified operations into a single processing sequence to be transmitted to an array of hardware processors, performing operations that are not integral in operation of the neural network in a separate thread of execution from the operations that are integral in operating the neural network; and mapping results to the combined operations that were included in the single processing sequence.

Operating a neural network defined by user code

Discriminative pretraining technique embodiments are presented that pretrain the hidden layers of a Deep Neural Network (DNN). In general, a one-hidden-layer neural network is trained first using labels discriminatively with error back-propagation (BP). Then, after discarding an output layer in the previous one-hidden-layer neural network, another randomly initialized hidden layer is added on top of the previously trained hidden layer along with a new output layer that represents the targets for classification or recognition. The resulting multiple-hidden-layer DNN is then discriminatively trained using the same strategy, and so on until the desired number of hidden layers is reached. This produces a pretrained DNN. The discriminative pretraining technique embodiments have the advantage of bringing the DNN layer weights close to a good local optimum, while still leaving them in a range with a high gradient so that they can be fine-tuned effectively.

Discriminative pretraining of deep neural networks

In a data center, neural network evaluations can be included for services involving image or speech recognition by using a field programmable gate array (FPGA) or other parallel processor. The memory bandwidth limitations of providing weighted data sets from an external memory to the FPGA (or other parallel processor) can be managed by queuing up input data from the plurality of cores executing the services at the FPGA (or other parallel processor) in batches of at least two feature vectors. The at least two feature vectors can be at least two observation vectors from a same data stream or from different data streams. The FPGA (or other parallel processor) can then act on the batch of data for each loading of the weighted datasets.

Memory bandwidth management for deep learning applications

Content-based analysis is performed on multimedia content prior to encoding the multimedia content in the rendering chain of processing. A content-based index stream is generated based on the content-based analysis and the content-based index stream is embedded in the multimedia file during rendering. The content-based index stream can be used to generate a content-based searchable index when necessary.

Embedding content-based searchable indexes in multimedia files

Described is a technology by which a playback list comprising similar songs is automatically built based on automatically detected/generated song attributes, such as by extracting numeric features of each song. The attributes may be downloaded from a remote connection, and/or may be locally generated on the playback device. To build a playlist, a seed song's attributes may be compared against attributes of other songs to determine which other songs are similar to the seed song and thus included in the playlist. Another way to build a playlist is based on similarity of songs to a set of user provided-attributes, such as corresponding to moods or usage modes such as 'resting' 'reading' &#x201c;jogging&#x201d; or &#x201c;driving&#x201d; moods/modes. The playlist may be dynamically adjusted based on user interaction with the device, such as when a user skips a song, queues a song, or dequeues a song.

Music steering with automatically detected musical attributes

In various embodiments, a transcript that represents a media file is created. Keyword candidates that may represent topics and/or content associated with the media content are then be extracted from the transcript. Furthermore, a keyword set may be generated for the media content utilizing a mutual information criteria. In other embodiments, one or more queries may be generated based at least in part on the transcript, and a plurality of web documents may be retrieved based at least in part on the one or more queries. Additional keyword candidates may be extracted from each web document and then ranked. A subset of the keyword candidates may then be selected to form a keyword set associated with the media content.

Keyword generation for media content

The use of a pipelined algorithm that performs parallelized computations to train deep neural networks (DNNs) for performing data analysis may reduce training time. The DNNs may be one of context-independent DNNs or context-dependent DNNs. The training may include partitioning training data into sample batches of a specific batch size. The partitioning may be performed based on rates of data transfers between processors that execute the pipelined algorithm, considerations of accuracy and convergence, and the execution speed of each processor. Other techniques for training may include grouping layers of the DNNs for processing on a single processor, distributing a layer of the DNNs to multiple processors for processing, or modifying an execution order of steps in the pipelined algorithm.

Deep neural networks training for speech and pattern recognition

Techniques are described for training a speech recognition model for accented speech. A subword parse table is employed that models mispronunciations at multiple subword levels, such as the syllable, position-specific cluster, and/or phone levels. Mispronunciation probability data is then generated at each level based on inputted training data, such as phone-level annotated transcripts of accented speech. Data from different levels of the subword parse table may then be combined to determine the accented speech model. Mispronunciation probability data at each subword level is based at least in part on context at that level. In some embodiments, phone-level annotated transcripts are generated using a semi-supervised method.

Subword-based multi-level pronunciation adaptation for recognizing accented speech

Deep Neural Network (DNN) training technique embodiments are presented that train a DNN while exploiting the sparseness of non-zero hidden layer interconnection weight values. Generally, a fully connected DNN is initially trained by sweeping through a full training set a number of times. Then, for the most part, only the interconnections whose weight magnitudes exceed a minimum weight threshold are considered in further training. This minimum weight threshold can be established as a value that results in only a prescribed maximum number of interconnections being considered when setting interconnection weight values via an error back-propagation procedure during the training. It is noted that the continued DNN training tends to converge much faster than the initial training.

Exploiting sparseness in training deep neural networks

A voice activity detection (VAD) module analyzes a media file, such as an audio file or a video file, to determine whether one or more frames of the media file include speech. A speech recognizer generates feedback relating to an accuracy of the VAD determination. The VAD module leverages the feedback to improve subsequent VAD determinations. The VAD module also utilizes a look-ahead window associated with the media file to adjust estimated probabilities or VAD decisions for previously processed frames.

Leveraging speech recognizer feedback for voice activity detection

Improved systems and methods are provided for transcribing audio files of voice mails sent over a unified messaging system. Customized grammars specific to a voice mail recipient are created and utilized to transcribe a received voice mail by comparing the audio file to commonly utilized words, names, acronyms, and phrases used by the recipient. Key elements are identified from the resulting text transcription to aid the recipient in processing received voice mails based on the significant content contained in the voice mail.

Growing community of inventors

Beijing, China

Frank Torsten Bernd Seide