Dell Products Inc.

Emc IP Holding Company LLC

Dell & Products L.p.

Amihai Savir

Ofir Ezrielev

Jehuda Shemer

Tomer Kushnir

Avitan Gefen

Nicole Reineke

Oshry Ben Harush

Stav Sapir

Naor Radami

Oshry Ben-Harush

Arieh Don

Elie Antoun Jreij

Vinay Sawal

Or Herman Saffar

Tsehsin Jason Liu

Romulo Teixeira De Abreu Pinho

Deepak Krishna

Jason Liu

Anat Parush Tzur

Oshry BenHarush

Roberto Stelling

Alice Jiang

Aviv Zeilig

Peter Ziesmann

Rachel Shalom

Jordan Leventis

Methods and systems for managing an artificial intelligence (AI) model are disclosed. An AI model may be part of an evolving AI model pipeline, the processes of which may include obtaining training data from data sources used to update the AI model. An attacker may introduce poisoned training data via one or more of the data sources as a form of attack on the AI model. When the poisoned training data is identified, the poisoned training data may be compared to existing training data to determine the attacker's goal. Based on the attacker's goal, remedial actions may be performed that may update operation of pipeline. The updated operation of the pipeline may reduce the computational expense for remediating impact of the poisoned training data, and may reduce the likelihood of obtaining poisoned training data in the future.

Managing artificial intelligence models to identify goals of malicious attackers

Methods and systems for identifying poisoned training data used for training artificial intelligence (AI) models are disclosed. To identify poisoned training data in a proposed training dataset, a causal model may be obtained. The causal model may include relationships relating data elements. The proposed training dataset may be identified as poisoned when data elements within the proposed training dataset do not satisfy the relationships set forth by the causal model. When the identification of poisoned training data is made, the AI model may not be updated using the proposed training dataset and the proposed training dataset may be discarded. If poisoned training data is not identified prior to training an AI model, methods and systems are disclosed for the remediation of the poisoned training dataset and subsequent tainted AI models. By doing so, the effect of poisoned training data may be prevented and/or efficiently computationally mitigated.

System and method for proactively identifying poisoned training data used to train artificial intelligence models

Methods and systems for providing customized services to users of data processing systems are disclosed. To do so, a query may be obtained from a user. Contextual information for the query may be attempted to be obtained using a first data source associated with the user. If the portion of the contextual information is stored as a static data chunk in the first data source, the static data chunk may be obtained. If the portion of the contextual information is stored as a reference in the first data source, the reference may be obtained. The reference may indicate that the portion of the contextual information is available as a dynamic data chunk from another data source. The portion of the contextual information may be obtained using the static data chunk and/or the dynamic data chunk and the query may be serviced using at least the portion of the contextual information.

Managing contextual information for queries using data sources managed by personal agents

Methods and systems for managing artificial intelligence (AI) models are disclosed. To manage AI models, AI models may be updated over time to obtain updated AI model instances. Following each update process, the updated instance of the AI model may be analyzed to determine whether poisoned training data was used to update the AI model. To perform the analysis, characteristics associated with the updated instance of the AI model may be compared to characteristics of the previous instance of the AI model. If the characteristics of the updated instance of the AI model differ from the characteristics of the previous instance of the AI model by an amount dictated by a threshold, the training data used to obtain the updated instance of the AI model may be treated as including poisoned training data.

System and method for detecting poisoned training data based on characteristics of updated artificial intelligence models

Methods, systems, and devices for providing computer-implemented services are disclosed. To provide the computer-implemented services, inference models used by data processing systems may be deployed to locations to generate inferences. Prior to deploying an inference model to a location, it may be determined whether the location is trustworthy. If the location is determined to not be trustworthy, an input data attack resistant inference model may be selected and deployed to perform inference generation. The input data attack resistant inference model may be trained to decrease a likelihood of the inferences generated by the input data attack resistant inference model being usable to reconstruct input data used to generate the inferences. By doing so, sensitive input data may be obfuscated during inference generation.

Managing input data attack resistant inference models

Architectures and techniques are described that can generate or receive a dependency graph of a microservices platform. The dependency graph can be constructed based on run time operation of microservices such that the nodes of the graph can represent microservices and the edges can represent the interactions during run time. The dependency graph along with an anomaly pattern can be embedded into an embedding space, and based on an examination of the embedding space, it can be determined whether the anomaly pattern exists in the dependency graph.

Anomaly detection for a microservices platform

An air-gapped computer receives, from a simplex communication input device, an encoded message communicated by a simplex communication output device. The air-gapped computer system can use a private key to decrypt an encrypted message which was encrypted by a corresponding public key. The air-gapped computer system decodes the encoded message. The air-gapped computer system verifies that the decoded message indicates that a required number of approvers have approved of a user executing a command. The air-gapped computer system extracts the approved command from the decoded message. The air-gapped computer system enables execution of the command by executing the command, or by providing the user with an access token which enables the user to physically access the air-gapped computer system and execute the command.

Executing commands on air-gapped computer systems

Disk drive reallocation or replacement is disclosed. When performing a data protection operation, a health score is determined for each of the disk drives associated with the data protection operation. Replacement is recommended for each of the disk drives with an unfavorable health score. The recommendation may also identify a drive class based in part on the write or wear pattern. This allows unhealthy drives to be replaced prior to performing the data protection operation.

Re-allocation of disks based on disk health prior to restore

A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system based on interactions of known users interacting with data assets in the FBDL system. Predicted data usage for a population of possible users is derived by training a generative model that uses reconstructive self-supervised learning (SSL) techniques to generate possible values for missing data usage features of the possible users. The predicted usage is then used to generate similarity scores that are combined with those of the known users to return relevant results to a target user. The predicted usage for possible users is processed in a consensus mechanism to produce integrated recommendations that are iteratively optimized using a chosen usage by the target user an a optimizer using a genetic algorithm to change consensus parameters to derive a final recommendation to the target user.

Search engine using predictive models based on partial information for a population of possible users through an optimized consensus mechanism

Methods and systems for identifying areas of interest in an image are disclosed. To manage identification of areas of interest in an image, a combination of automated and subject matter expert driven processes may be used to identify the areas of interest. Some areas of interest may be highly relevant for a user of an image and may be classified as primary areas of interest. The image and subsequent use of the image may be enhanced through the areas of interest by improving storage and downstream usability of the image.

System and method for identifying auxiliary areas of interest for image based on focus indicators

A system, method, and computer-readable medium for performing a data center management and monitoring operation to predict event failures, comprising taking time stamped telemetry metrics of data asset clusters; sampling the time stamped telemetry metrics; separating the sampled time stamped telemetry metrics into groups based on metric functionality; providing data of the groups into a residual neural network to learn patterns of the three groups; and concatenating output of the residual neural network of the groups, wherein concatenated output is feed back to the residual neural network to learn relationships of the three groups.

Data center monitoring and management operation for data unavailable/data loss (DU/DL) predictions, including service level agreement (SLA) failure prediction

Methods and systems for managing storage of data are disclosed. To manage storage of data, images may be stored across a number of storages that provide varying levels of storage performance and have correspondingly varying costs for storing data. To store the images across the storages, the images may be segmented into image segments and a likelihood of each of the image segments being used in the future may be identified. The image segments that are more likely to be used in the future may be stored in higher performance storages while the image segments that are less likely to be used in the future may be stored in lower performance storages. To identify the likelihood of each of the image segments being used in the future, the image segments may be classified based on their membership in one or more areas of interest within the images.

System and method for storage management of images

Methods and systems for managing artificial intelligence (AI) models are disclosed. To manage AI models, an instance of an AI model may not be re-trained using training data determined to be potentially poisoned. By doing so, malicious attacks intending to influence the AI model using poisoned training data may be prevented. To do so, a first level of strength of a first causal relationship present in historical training data may be compared to a second level of strength of a second causal relationship present in a candidate training data set. The first level of strength and the second level of strength may be expected to be similar within a threshold. If a difference between the first level of strength and the second level of strength is not within the threshold, the candidate training data may be treated as including poisoned training data.

Detecting possible attacks on artificial intelligence models using strengths of causal relationships in training data

A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system. The search engine receives a search query from an unprivileged user or a user not having sufficient privileges to access the FBDL. It returns initial results to the unprivileged user including a first data asset recommendation responsive to the search query. It then determines a causal reason that the first data asset was recommended, and uses a similarity engine conditioned on the causal reason to return a replacement data asset in response to the search query.

Search engine using causal replacement of search results for unprivileged access rights

A search engine responding to a user query to find relevant data assets in a federation business data lake (FBDL) system by monitoring and recording interactions of known users interacting with data assets in the FBDL system. Predicted data usage for unknown or new users is derived by training a generative model that uses reconstructive self-supervised learning (SSL) techniques to generate possible values for missing data usage features of the unknown users. The predicted usage is then used to generate similarity scores that are combined for those of the known users to help inform the search engine processing to return relevant results to a target user.

Growing community of inventors

Newton, MA, United States of America

Amihai Savir