International Business Machines Corporation

Olivia Choudhury

Issa Sylla

Aris Gkoulalas-Divanis

Amarendra Das

Panagiotis Karampourniotis

Daniel M Gruen

Rohit Ranchal

Morgan Anna Foreman

Nicola Palmarini

Harigovind Venkatraj Ramasamy

Theodoros Salonidis

Senthil K Bakthavachalam

Yoonyoung Park

By intercepting a natural language communication of a protected party, the communication is monitored, wherein the protected party is a human being. Within the monitored communication using a natural language processing engine, a natural language interaction between the protected party and a second party is detected. To determine an interaction pattern, the natural language interaction is analyzed. The interaction pattern includes data derived from the monitored communication, metadata of the protected party, and metadata of the second party. Using the interaction pattern and an interaction behavior model, an adverse result of the natural language interaction is predicted, wherein the adverse result comprises an economic loss to the protected party. By notifying the protected party, the predicted adverse result is intercepted.

Intercepting inadvertent conversational disclosure of personal information

A method, apparatus, system, and computer program product for training a global machine learning model. A hierarchical structure for nodes in which the global machine learning model is located at a primary node in the hierarchical structure is identified. Authorized nodes in which local data is authorized for use in training in the authorized nodes for a local training of local machine learning models are determined. The machine learning models in the authorized nodes are trained using the local data in the authorized nodes to generate local model updates to weights in the local machine learning models. The local model updates to the weights are propagated upward in the hierarchical structure to the global machine learning model, wherein a node receiving local model updates to the weights from nodes from a lower level aggregates the weights in the local model updates received from the nodes in the lower level.

Hierarchical federated learning using access permissions

In an approach for anonymizing data, a processor receives a mixed-type dataset with at least two relational attributes and at least one textual attribute. A processor runs the mixed-type dataset through a text annotator to discover a set of personally identifiable information (PII). A processor creates a set of ghost attributes to add to the mixed-type dataset. A processor anonymizes data of the at least two relational attributes and the set of ghost attributes. A processor replaces each PII in the textual attribute with the corresponding anonymized data in the at least two relational attributes or the set of ghost attributes to create an anonymized mixed-type dataset. A processor removes the set of ghost attributes from the anonymized mixed-type dataset. A processor shuffles records of the anonymized mixed-type dataset to create a shuffled anonymized mixed-type dataset. A processor outputs the shuffled anonymized mixed-type dataset.

Anonymizing relational and textual data

Systems, computer-implemented methods, and computer program products that can facilitate anonymizing a network based on factors including network attributes, node attributes, and edge attributes describing connections between nodes are described. According to an embodiment, a system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise an anonymizing component that can anonymize network information of the network based on a network attribute for a network and a node attribute of a first node of the network, resulting in an anonymized network.

Anonymizing a network using network attributes and entity based access rights

A method, system, and computer program product for detecting data tampering with resilient watermarking is provided. The method accesses a first relational data set on a data repository. The first relational data set includes a plurality of data elements. The first relational data set is sorted to generate a first sorted list and a second sorted list of the plurality of data elements. The method generates a watermark from the first sorted list and the second sorted list. The watermark contains a hash corresponding to the first sorted list and the second sorted list of the plurality of data elements. In response to an access request for the first relational data set, the method verifies an integrity of the first relational data set based on the watermark.

Resilient watermarking

A computer-implemented method for training a global federated learning model using an aggregator server includes training multiple local models at respective local nodes. Each local node selects a set of attributes from its training dataset for training its local model. Each local node generates an anonymized training dataset by using a syntactic anonymization method, and by selecting quasi-identifying attributes from training attributes, and generalizing the quasi-identifying attributes using a syntactic algorithm. Further, each local node computes a syntactic mapping based on equivalence classes produced in the anonymized training dataset. The aggregator server computes a union of mappings received from all the local nodes. Further, federated learning includes training the global federated learning model by iteratively sending, by the local nodes to the aggregator server, parameter updates computed over the local models. The aggregator server aggregates the received parameter updates, and sends the aggregated parameters to the local nodes.

Anonymizing data for preserving privacy during use for federated machine learning

A method, computer system, and a computer program product for SLA management is provided. The method may include collecting metrics from services within a composite service. The method may include determining, by a first smart contract, a first violation occurred between a first pair of services, whereby the first smart contract and the first pair of services are associated with a first private channel within a blockchain network. The method may include determining, by a second smart contract, a second SLA violation occurred between a second pair of services, whereby the second smart contract and the second pair of services are associated with a second private channel. The method may include determining that the first SLA violation and second SLA violation are related and the second SLA violation occurred before the first SLA violation. The method may include identifying a violating service within the second pair of services.

Growing community of inventors

Cambridge, MA, United States of America

Olivia Choudhury