Sri International

Fair Isaac Corporation

Burning Glass Technologies, LLC

Dayne B Freitag

Phillip Andrew Porras

William Scott Mark

Kenneth C Nitz

Necip Fazil Ayan

Michael John Wolverton

Anu Kumar Pathria

John James Byrnes

Matthias Blume

Edgar Kalns

Thomas J Lee

Keith M Skinner

Scott Michael Zoldi

Girish Acharya

Yi Yao

Edmond D Chow

Melinda Gervasio

Theodore James Crooks

Richard Rohwer

Michael Wessel

Krishna Gopinathan

Mark Alan Laffoon

Anirban Roy

Richard Calmbach

Matthew N Dailey

Louise Yarnall

Paul Stern Kalmar

Robert Sasseen

Zhiqiang Wang

Chalaporn Hathaidharm

Zachary Weiler

A method for creating a training playbook to train a potential target of social engineering against an attack by an adversarial actor includes generating a training playbook based on a plurality of conversational patterns from a respective plurality of dialogs between non-human synthetic persona with the adversarial actor. A given conversational pattern of the plurality of conversational patterns comprises one or more dialog interaction elements utilized by the adversarial actor during a respective dialog of the plurality of dialogs. The method also includes configuring a dialog model for a second synthetic persona to engage in a training dialog with a target trainee to train the target trainee to detect, avoid and/or mitigate a future attack by the adversarial actor. The dialog model is configured to emulate the future attack by the adversarial actor by utilizing the training playbook. The method additionally includes outputting the training playbook.

Systems and methods for training against adversarial attacks

A method of determining an adversarial attack playbook includes receiving, from an adversarial actor, an electronic communication intended for a target user. The method includes engaging in a deep dialog with the adversarial actor by deploying a synthetic persona dynamically during the electronic communication. The deep dialog includes multiple rounds of communication exchanges. The method includes determining a length and type of the deep dialog to obtain attributes related to the adversarial actor. The method includes identifying a conversational pattern from the deep dialog. The conversational pattern comprises dialog interaction elements utilized by the adversarial actor. The method includes dynamically producing, based on the conversational pattern, the playbook associated with the adversarial actor. The playbook is indicative of a dialog interaction strategy implemented by the adversarial actor. The method includes providing the playbook to a social engineering attack (SEA) system in order to detect, avoid and/or mitigate future attacks.

Conversation-depth social engineering attack detection using attributes from automated dialog engagement

This disclosure describes machine learning techniques for capturing human knowledge for performing a task. In one example, a video device obtains video data of a first user performing the task and one or more sensors generate sensor data during performance of the task. An audio device obtains audio data describing performance of the task. A computation engine applies a machine learning system to correlate the video data to the audio data and sensor data to identify portions of the video, sensor, and audio data that depict a same step of a plurality of steps for performing the task. The machine learning system further processes the correlated data to update a domain model defining performance of the task. A training unit applies the domain model to generate training information for performing the task. An output device outputs the training information for use in training a second user to perform the task.

Machine learning system for technical knowledge capture

Mathematical technologies for recommending content to a user based on a user's preferences are disclosed. Embodiments of these technologies can generate a probabilistic representation of a data set, and then adjust the probabilistic representation to reflect a user-specific weighting scheme. The user preference-adjusted representation of the data set can be used to recommend content to the user.

Similarity metric relativized to a user's preferences

A platform for developing a virtual personal assistant ('VPA') application includes an ontology that defines a computerized structure for representing knowledge relating to one or more domains. A domain may refer to a category of information and/or activities in relation to which the VPA application may engage in a conversational natural language dialog with a computing device user. Re-usable VPA components may be linked to or included in the ontology. An ontology populating agent may at least partially automate the process of populating the ontology with domain-specific information. The re-usable VPA components may be linked with the domain-specific information through the ontology. A VPA application created with the platform may include domain-adapted re-usable VPA components that may be called upon by an executable VPA engine to determine a likely intended meaning of conversational natural language input of the user and/or initiate an appropriate system response to the input.

Rapid development of virtual personal assistant applications

Methods and apparatus, including computer program products, related to entity name translation. In general, a set of data records including one or more names of one or more entities in a source language is received and candidate translations are generated for the names of the data records. The candidate translations are strings of characters in a target language that has a different orthography than an orthography of the source language. A data repository is queried for records matching the candidate translations. A translation of at least one of the candidate translations is selected based on a comparison of a combination of candidate translations and properties of the source record comprising of the publication date, entity name co-occurrence and their frequency with the result of the querying the data repository.

Translation of entity names based on source document publication date, and frequency and co-occurrence of the entity names

Entity disambiguation resolves which names, words, or phrases in text correspond to distinct persons, organizations, locations, or other entities in the context of an entire corpus. The invention is based largely on language-independent algorithms. Thus, it is applicable not only to unstructured text from arbitrary human languages, but also to semi-structured data, such as citation databases and the disambiguation of named entities mentioned in wire transfer transaction records for the purpose of detecting money-laundering activity. The system uses multiple types of context as evidence for determining whether two mentions correspond to the same entity and it automatically learns the weight of evidence of each context item via corpus statistics. The invention uses multiple search keys to efficiently find pairs of mentions that correspond to the same entity, while skipping billions of unnecessary comparisons, yielding a system with very high throughput that can be applied to truly massive data.

Method and apparatus for automatic entity disambiguation

A Hidden Markov model is used to segment a data sequence. To reduce the potential for error that may result from the Markov assumption, the Viterbi dynamic programming algorithm is modified to apply a multiplicative factor if a particular set of states is re-entered. As a result, structural domain knowledge is incorporated into the algorithm by expanding the state space in the dynamic programming recurrence. In a specific example of segmenting resumes, the factor is used to reward or penalize (even require or prohibit) a segmentation of the resume that results in the re-entry into a section such as Experience or Contact Information. The method may be used to impose global constraints in the processing of an input sequence or to impose constraints to local sub-sequences.

Method for improving results in an HMM-based segmentation system by incorporating external knowledge

A method and apparatus is disclosed for making predictions about entities represented in documents and for information analysis of text documents or the like, from a large number of such documents. Predictive models are executed responsive to variables derived from canonical documents to determine documents containing desired attributes or characteristics. The canonical documents are derived from standardized documents, which, in turn, are derived from original documents.

Method and apparatus for making predictions about entities represented in documents

Growing community of inventors

Pittsburgh, PA, United States of America

Dayne B Freitag