Microsoft Technology Licensing, LLC

Ravi Chandru Shahani

Michael V Calcagno

Su Chin Chang

Domenic J Cipollone

David Parkinson

Mari J B Olsen

Mark Eugene Russinovich

William Michael Zintel

George Joy

Brian Robert Crawford

Mahmood Gulam Qadir

Vito J Sabella

Wojtek Kozaczynski

Brian P Ellis

James O Todd

Ken Ming-Kin Yip

Vitaliy Titov

Christopher M Lang

Jonathan K Johnson

Amy M Lewis

Kevin Roland Powell

Amar Dinesh Patel

Douglas John Hines

Siarhei Alonichau

Yizheng Cai

Lei Wang

Katherine J Brainard

Nicholas Van Caldwell

Tae Hyung Kim

Allan Folting

Charles David Ellis

Nicholas Caldwell

Benjamin Edward Childs

Zhenghao Wang

Abhijeet Surendra Hatekar

Qing Chang

Robert Stumberger

Juilee Rege

Roman Holenstein

Revanth Rameshkumar

Ethan Jacob Holland

A method of operating a telemetry system includes automatically populating a base field of a schema in an event definition using a logging library of the telemetry system for an event of an instrumented application, and automatically populating a conditional field of the schema in the event definition using the logging library in response to a selected condition for the event.

Telemetry system extension

IPRID reputation assessment enhances cybersecurity. IPRIDs include IP addresses, domain names, and other network resource identities. A convolutional neural network or other machine learning model is trained with data including aggregate features or rollup features or both. Aggregate features may include aggregated submission counts, classification counts, HTTP code counts, detonation statistics, and redirect counts, for instance. Rollup features reflect hierarchical rollups of data using &#x3c;unknown&#x3e; value placeholders specified in IPRID templates. The trained model can predictively infer a label, or produce a rapid lookup table of IPRIDs and maliciousness probabilities. Training data may be organized in grids with rows, columns, planes, branches, and slots. Training data may include whois data, geolocation data, and tenant data. Training data tuple sets may be expanded by date or by original IPRID. Trained models can predict domain labels accurately at scale, even when most of the domains encountered have never been classified before.

Predictive internet resource reputation assessment

A method of operating a telemetry system includes automatically populating a first set of fields in a schema of an event definition using a logging library of the telemetry system, and receiving the set of fields via a request message in an application protocol.

Telemetry request system

Cybersecurity enhancements help avoid malicious Uniform Resource Locators (URLs). Embodiments may reduce or eliminate reliance on subjective analysis or detonation virtual machines. URL substrings are automatically analyzed for maliciousness using malice patterns. Patterns may test counts, lengths, rarity, encodings, and other inherent aspects of URLs. URLs may be analyzed individually, or in groups to detect shared portions, or both. URL analysis may use or avoid machine learning, and may use or avoid lookups. Malice patterns may be used individually or in combinations to detect malicious URLs. Analysis results may enhance security through blocking use of suspect URLs, flagging them for further analysis, or allowing their validated use, for instance. Analysis results may also be fed back to further train a machine learning model or a statistical model.

Uniform resource locator security analysis using malice patterns

A method of operating a telemetry system includes automatically populating a first set of fields in a schema of an event definition using a logging library of the telemetry system, and populating a second set of fields in the schema selected by an event author.

Telemetry definition system

Potential linguistic errors within a sequence of words of a sentence are identified based on analysis of a configurable sliding window. The analysis is performed based on an assumption that if a sequence of words occurs frequently enough within a large, well-formed corpus, its joint probability for occurring in a sentence is very likely to be greater than the same words randomly ordered.

Linguistic error detection

The present disclosure is directed to a method of verifying a compound word. The method includes receiving an input signal indicative of a textual input and accessing a rule and a lexical data structure from data stores. The rule is applied to the textual input to determine whether the textual input is a valid compound word. An output signal is provided that is indicative of whether the textual input is a compound word.

Regular expression word verification

Tools and techniques are described for providing multi-lingual word hyphenation using inductive machine learning on training data. Methods provided by these techniques may receive training data that includes hyphenated words, and may inductively generate hyphenation patterns that represent substrings of these words. The hyphenation patterns may include the substrings and hyphenation codes associated with characters occurring in the substrings. The methods may receive induction parameters applicable to generating the hyphenation patterns, and may store the hyphenation patterns into a language-specific lexicon file. These methods may also receive requests to hyphenate input words that occur in a human language, and may evaluate how to process the request based on the language. The methods may search for hyphenation patterns occurring in the input words, with the hyphenation patterns being stored in the lexicon file. Finally, the methods may respond to the request, indicating whether the hyphenation patterns occurred in the input words.

Multi-lingual word hyphenation using inductive machine learning on training data

Computer-readable media, computer systems, and computing methods are provided for recommending websites that are relevant to a current website to which a user has navigated. A search engine is used to track a set of websites the user has visited immediately prior to the current website, while predictive model(s) are used to generate a sequence of websites that include the current website and the tracked websites. The sequence is compared against strings of websites within a browser-history log to identify matching strings, where the matching strings include the sequence and a respective candidate website. A probability of relevance is computed from a frequency that each of the matching strings has been visited within a predefined time frame. The probability of relevance for each of the matching strings is ranked against one another to distill the highest-ranked matching strings, which are parsed to extract and present the candidate websites included therein.

Predictively suggesting websites

A framework for generating a semantic interpretation of natural language input includes an interpreter, a first set of types, and a second set of types. The interpreter is adapted to mediate between a client application and one or more analysis engines to produce interpretations of the natural language input that are valid for the client application. The first set of types is adapted to define interactions between the interpreter and the one or more analysis engines. The second set of types is adapted to define interactions between the interpreter and the client application.

Semantic framework for natural language programming

Architecture that employs adaptive learning algorithms to adapt a data correction tool to user-specific behavior during runtime. The architecture includes a framework for training and measuring adaptive learning algorithms, adapting the current text correction tool codebase, and one or more different adaptive learning algorithms. This enables a text correction system to adapt the behavior of the text correction system to an individual user based on the user's interaction with the data correction system. This also facilitates the testing and improvements in an adaptive learning algorithm at the vendor before shipping in a product to the end-user. This reduces the risk of shipping a feature the precise behavior of which is different for each user.

Adaptive learning framework for data correction

A software development tool for programming natural language software applications is described. The software development tool includes a programming language and a compiler. The programming language has a set of programming constructs for facilitating natural language programming. The compiler is adapted to take a software program containing instances of the set of programming constructs and to generate a software application.

Semantic programming language

Growing community of inventors

Redmond, WA, United States of America

Ravi Chandru Shahani