Microsoft Technology Licensing, LLC

Google Inc.

Xinying Song

Jianfeng Gao

Jinchao Li

Hui Su

Yu Wang

Haiyuan Cao

Jaideep Sarkar

Yelong Shen

Karan Srivastava

Xiaodong He

Jianshu Chen

Prabhdeep Singh

Li Deng

Ye-Yi Wang

Chenlei Guo

Yunbo Cao

Chin-Yew Lin

Brian D Remick

Divya Jetley

Byungki Byun

Yu Hu

Edward Thiele

Mohammed Aatif Ali

Marcus Gois

Stephen Friesen

Yang Song

Johnson T Apacible

Michael Gamon

Patrick Pantel

Lin Xiao

Zhiyuan Chen

Ji He

Mariana Stepp

Hamid Palangi

Andreea Bianca Spataru

Yang Zou

Tae Yano

Shahina Ferdous

Ah Young Kim

Systems and methods for performing inference for word or wordpiece tokenization are disclosed using a left-to-right longest-match-first greedy process. In some examples, the vocabulary may be organized into a trie structure in which each node includes a precomputed token or token_ID and a fail link, so that the tokenizer can parse the trie in a single pass to generate a list of only those tokens or token_IDs that correspond to the longest matching vocabulary entries in the sample string, without the need for backtracking. In some examples, the vocabulary may be organized into a trie in which each node has a fail link, and any node that would share token(s) or token_ID(s) of a preceding node is instead given a prev_match link that points back to a chain of nodes with those token(s) or token_ID(s).

Inference methods for word or wordpiece tokenization

Systems and methods for performing inference for word or wordpiece tokenization are disclosed using a left-to-right longest-match-first greedy process. In some examples, the vocabulary may be organized into a trie structure in which each node includes a precomputed token or token ID and a fail link, so that the tokenizer can parse the trie in a single pass to generate a list of only those tokens or token IDs that correspond to the longest matching vocabulary entries in the sample string, without the need for backtracking. In some examples, the vocabulary may be organized into a trie in which each node has a fail link, and any node that would share token(s) or token_ID(s) of a preceding node is instead given a prev_match link that points back to a chain of nodes with those token(s) or token_ID(s).

Generally discussed herein are devices, systems, and methods for scheduling tasks to be completed by resources. A method can include identifying features of the task, the features including a time-dependent feature and a time-independent feature, the time-dependent feature indicating a time the task is more likely to be successfully completed by the resource, converting the features to feature values based on a predefined mapping of features to feature values in a first memory device, determining, by a gradient boost tree model and based on a first current time and the feature values, a likelihood the resource will successfully complete the task, and scheduling the task to be performed by the resource based on the determined likelihood.

Resource scheduling using machine learning

A workflow engine tool is disclosed that enables scientists and engineers to programmatically author workflows (e.g., a directed acyclic graph, 'DAG') with nearly no overhead, using a simpler script that needs almost no modifications for portability among multiple different workflow engines. This permits users to focus on the business logic of the project, avoiding the distracting tedious overhead related to workflow management (such as uploading modules, drawing edges, setting parameters, and other tasks). The workflow engine tool provides an abstraction layer on top of workflow engines, introducing a binding function that converts a programming language function (e.g., a normal python function) into a workflow module definition. The workflow engine tool infers module instances and induces edge dependencies automatically by inferring from a programming language script to build a DAG.

Workflow engine tool

Systems and methods are disclosed for intelligent scheduling of calls to sales leads, leveraging machine learning (ML) to optimize expected results. One exemplary method includes determining, using a connectivity prediction model, call connectivity rate predictions; determining timeslot resources; allocating, based at least on the call connectivity rate predictions and timeslot resources, leads to timeslots in a first time period; determining, within a timeslot and using a lead scoring model, lead prioritization among leads within the timeslot; configuring, based at least on the lead prioritization, the telephone unit with lead information for placing a phone call; and applying a contextual bandit (ML) process to update the connectivity prediction model, the lead scoring model, or both. During subsequent time periods, the updated connectivity prediction and lead scoring models are used, thereby improving expected results over time.

Intelligent scheduling tool

Generally discussed herein are devices, systems, and methods for task routing. A method can include receiving, from a resource, a request for a task, in response to receiving the request, determining whether to retrieve a new task of new tasks stored in a first queue or a backlog task of backlog tasks stored in a second queue based on a combined amount of backlog tasks and new tasks relative to a capacity of the resource or the resources, retrieving the new task or the backlog task from the determined first queue or second queue, respectively, based on the determination, and providing the retrieved task to the resource.

Dynamic, load balanced task routing

A processing unit can successively operate layers of a multilayer computational graph (MCG) according to a forward computational order to determine a topic value associated with a document based at least in part on content values associated with the document. The processing unit can successively determine, according to a reverse computational order, layer-specific deviation values associated with the layers based at least in part on the topic value, the content values, and a characteristic value associated with the document. The processing unit can determine a model adjustment value based at least in part on the layer-specific deviation values. The processing unit can modify at least one parameter associated with the MCG based at least in part on the model adjustment value. The MCG can be operated to provide a result characteristic value associated with test content values of a test document.

Training and operating multi-layer computational models

Techniques for providing a people recommendation system for predicting and recommending relevant people (or other entities) to include in a conversation based on contextual indicators. In an exemplary embodiment, email recipient recommendations may be suggested based on contextual signals, e.g., project names, body text, existing recipients, current date and time, etc. In an aspect, a plurality of properties including ranked key phrases are associated with profiles corresponding to personal entities. Aggregated profiles are analyzed using first- and second-layer processing techniques. The recommendations may be provided to the user reactively, e.g., in response to a specific query by the user to the people recommendation system, or proactively, e.g., based on the context of what the user is currently working on, in the absence of a specific query by the user.

Contextual people recommendations

Systems, methods, and computer-readable media for providing semantically-relevant discovery of solutions are described herein. In some examples, a computing device can receive an input, such as a query. The computing device can process each word of the input sequentially to determine a semantic representation of the input. Techniques and technologies described herein determine a response to the input, such as an answer, based on the semantic representation of the input matching a semantic representation of the response. An output including one or more relevant responses to the request can then be provided to the requestor. Example techniques described herein can apply machine learning to train a model with click-through data to provide semantically-relevant discovery of solutions. Example techniques described herein can apply recurrent neural networks (RNN) and/or long short term memory (LSTM) cells in the machine learning model.

Semantically-relevant discovery of solutions

Techniques for providing a people recommendation system for predicting and recommending relevant people (or other entities) to include in a conversation. In an exemplary embodiment, a plurality of conversation boxes associated with communications between a user and target recipients, or between other users and recipients, are collected and stored as user history. During a training phase, the user history is used to train encoder and decoder blocks in a de-noising auto-encoder model. During a prediction phase, the trained encoder and decoder are used to predict one or more recipients for a current conversation box composed by the user, based on contextual and other signals extracted from the current conversation box. The predicted recipients are ranked using a scoring function, and the top-ranked individuals or entities may be recommended to the user.

Relevance group suggestions

A set of representations of item-page pairs of items and respective web pages that include the respective items is obtained, each representation including feature function values indicating weights associated with features of associated web pages, the features including page classification features. An annotated set of labeled training data that is annotated with salience annotation values of items for respective web pages that include the items is obtained. The salience annotation values are determined based on a soft function, by determining a first count of a total number of user queries associated with corresponding visits to the respective web pages, and determining a ratio of a second count to the first count, the second count determined as a cardinality of a subset of the corresponding visits that are associated with user queries that include the item, the subset included in the corresponding visits. Models are trained using the annotated set.

Identifying salient items in documents

Described herein are techniques for extracting data records containing user-generated content from documents. The documents may be processed into document trees in which sub-trees represent the data records of the document. Domain constraints may be used to locate structured portions of the document tree. For example, anchor trees may be located as being sets of sibling sub-trees with similar tag paths that contain the domain constraints. The anchor trees may then be used to determine a record boundary (e.g., the start offset and length) of the data records. Finally, the data records may be extracted based on the anchor trees and the record boundaries.

Domain constraint path based data record extraction

Embodiments for a Mining Data Records based on Anchor Trees (MiBAT) process are disclosed. In accordance with at least one embodiment, the MiBAT process extracts data records containing user-generated content from web documents. The web document is processed into a Document Object Model (DOM) tree in which sub-trees of the DOM tree represent the data records of the web document. Domain constraints are used to locate structured portions of the DOM tree. Anchor trees are then located as being sets of sibling sub-trees which contain the domain constraints. The anchor trees are then used to determine a record boundary (i.e. the start offset and length) of the data records. Finally, the data records are extracted based on the anchor trees and the record boundaries.

Domain constraint based data record extraction

Growing community of inventors

Bellevue, WA, United States of America

Xinying Song