Amazon Technologies, Inc.

Oath Inc.

Baris Coskun

MohamadAli Torkamani

Jeffrey Earl Bickford

Luca Melis

Wei Ding

Yifan Hu

Shane Anil Pereira

Steven S Skiena

Pranav Garg

Meizhu Liu

Sergul Aydore

Zhilu Zhang

Abhinandan Dubey

Christopher Stinson

Jiahao Ding

Qian Cui

Oleg Yurievich Polyakov

Bhavna Soman

Junting Ye

Disclosed are systems and methods that monitor for malicious and unauthorized behaviors, determine categories for detected malicious behaviors, determine why a domain is determined to be malicious, and provide information to users that identifies the categories and reasons as to why a domain is determined to be malicious. In some implementations, the disclosed systems and methods may be utilized to provide monitoring security to customers of a cloud service. For example, customers of a cloud service may maintain an account with the cloud service and the disclosed implementations may be utilized to protect those accounts from malicious attacks and cybercrimes such as, but not limited to, spam, phishing, malware, botnets, etc.

Detection of malicious domains

Systems and methods for performing anomalous activity monitoring for a service provider network are disclosed. In response to receiving an activity log, a machine learning-based activity monitor may parse the activity log into segments, generate event objects from a segment of the activity log, encode the event objects, and then reconstruct the event objects based on decoding the encoded event objects. The encoding and decoding may be performed based on a model that was trained using training data with no known malicious activity. The event objects may comprise at least two or more event defining characteristics and an event count. By comparing the reconstructed event objects to corresponding initial versions of the event objects, the machine learning-activity monitor may determine an anomaly score and may provide an indication of events determined to be anomalous based on the score.

Deep sequential anomalous events detection

Disclosed are systems and methods to compare two or more machine learning models to determine the comparative performance of those models. Markers may be assigned to data items and data item marker scores generated for those data items, independent of the machine learning models. Each of the machine learning models to be compared may then process the data items and generate respective model scores for those data items. A sub-set of the data items may then be generated for each machine learning model based on the model scores assigned to the data items by the respective model. A model marker score may then be computed for each machine learning model based on the marker scores assigned to each of the data items of the sub-set of data items determined for each model. Finally, the model marker scores may be compared to determine which machine learning model has the highest performance.

Machine learning model evaluation and comparison

Techniques for enabling the identification of anomalous events associated with an object storage service of a cloud provider network using a variational autoencoder model including a pre-trained embedding for selected features of events are described. A variational autoencoder, for example, encodes data into a latent space and reconstructs approximations of the data from an encoding in the latent space. In this context, for example, anomalous events of interest might represent unauthorized or abusive behavior associated with storage resources provided by an object storage service (or in association with other types of computing resources provided by other services of a cloud provider network). Legitimate (or benign) access patterns to an object storage service can be modeled by utilizing observed data plane events stored by an account activity monitoring service. Once trained, the model can be used to identify anomalous events.

Detecting anomalous storage service events using autoencoders

Devices, systems, and methods are provided for cloud-based entity reputation scoring. A method may include determining, based on domain name service (DNS) data associated with entities of the cloud-based environment, a k-partite graph with nodes and edges, a node including a first elastic computing instance. The method may include generating features associated with the first elastic computing instance. The method may include determining, based on the features, a minimum value, a maximum value, and an average value, and generating a feature vector comprising the minimum value, the maximum value, and the average value. The method may include determining, based on the feature vector, a reputation score associated with the first elastic computing instance. The method may include communicating based on the reputation score.

Enhanced cloud-based entity reputation scoring

Techniques for monitoring a computing environment for anomalous activity are presented. An example method includes receiving a request to invoke an action within the computing environment. An anomaly score is generated for the received request by applying a probabilistic model to properties of the request. The anomaly score generally indicates a likelihood that the properties of the request correspond to historical activity within the computing environment for a user associated with the request. The probabilistic model generally comprises a model having been trained using historical activity within the computing environment for a plurality of users, the historical activity including information identifying an action performed in the computing environment and contextual information about a historical request. Based on the generated anomaly score, one or more actions are taken to process the request such that execution of requests having anomaly scores indicative of unexpected activity may be blocked pending confirmation.

Scoring events using noise-contrastive estimation for anomaly detection

Systems, devices, and methods are provided for detecting anomalous events from categorical data using autoencoders. A system may receive a data set associated with actions requested within the computing environment, wherein the data set includes first categorical data indicative of anomalous activity in the computing environment. The system may train an autoencoder to reconstruct approximations of requests associated with the computing environment based on the received data set, wherein training the autoencoder includes using a beta divergence and a maximum mean discrepancy divergence. The trained system may receive a request to invoke an action within the computing environment, may generate a reconstruction of the request to invoke the action using the trained autoencoder, may determine a normalcy score based on a probability that the reconstruction of the request exists in the training data set, and, based on the calculated normalcy score, may determine whether requests indicate anomalous data.

Detecting anomalous events from categorical data using autoencoders

Techniques for monitoring a computing environment for anomalous activity are presented. An example method includes receiving a request to invoke an action within a computing environment, with the request including a plurality of request attributes and a plurality of contextual attributes. A normalcy score is generated for the received request by encoding the received request into a code in latent space of an autoencoder, reconstructing the request from the code, and generating a probability distribution indicating a likelihood that the reconstructed request attributes exist in a data set of non-anomalous activity. Based on the calculated normalcy score, one or more actions are taken to process the request such that execution of non-anomalous requests is allowed, and execution of potentially anomalous requests may be blocked pending confirmation.

Detecting anomalous events using autoencoders

Disclosed are systems and methods for improving interactions with and between computers in content searching, generating, hosting and/or providing systems supported by or configured with personal computing devices, servers and/or platforms. The systems interact to identify and retrieve data within or across platforms, which can be used to improve the quality of data used in processing interactions between or among processors in such systems. The disclosed systems and methods provide a novel clustering framework applied on datasets of network interactions to automatically identify IP clusters carrying out a specific task(s) based on an IP blacklist. The disclosed systems and methods can analyze network activity of devices associated with the IP addresses, and/or the IP addresses themselves, and perform an automatic, on-the-spot analysis that results in a determination whether the activity is permitted on or over a network.

Computerized system and method for automatically determining malicious IP clusters using network activity data

Communication accounts may contain information such as account holder names, contact lists, and communication logs. Such information may be processed for generating features that may be used as corpus for a machine learning algorithm for developing classifiers of names. Specifically, names and contact names of the accounts may be arranged in to a document according to the manner in which account holders communicate with the contacts. The document may be used for generating word embedding of the names. Names prelabeled with ethnicity together with their word embedding may be used as training data for developing ethnicity classifiers based on machine learning algorithms.

Growing community of inventors

Glen Rock, NJ, United States of America

Baris Coskun