EFFICIENT RECOVERY FROM FAILURES DURING DISTRIBUTED TRAINING OF MACHINE LEARNING MODELS

    公开(公告)号:US20240428082A1

    公开(公告)日:2024-12-26

    申请号:US18491604

    申请日:2023-10-20

    Abstract: A placement plan for training state checkpoints of a machine learning model is generated based at least in part on a number of training servers of a distributed training environment. The plan indicates, with respect to an individual server, one or more other servers at which replicas of training state checkpoints of the individual server are to be stored. During selected periods of one or more training iterations of the model, respective portions of a replica of a training state checkpoint of a first server are transmitted to a second server selected based on the placement plan. After an event causes disruption of the training iterations, one of the checkpoints generated at the first server is retrieved from the second server and used to resume the training iterations.

    Analyzing web service frontends to extract security-relevant behavior information

    公开(公告)号:US11509730B1

    公开(公告)日:2022-11-22

    申请号:US17119238

    申请日:2020-12-11

    Abstract: Techniques are described for generating a specification of security-relevant behavior associated with web services of a cloud provider network. Source code or software development artifacts associated with an implementation of a web service is obtained, where the source code of software development artifacts include an implementation of a request handler for an action of the service. The request handler includes a request authorization component, e.g., which may involve interaction with an identity and access management service of the cloud provider network to authenticate and authorize requests and may further rely upon one or more authorization contexts included in the requests received by the request handler. An interprocedural data flow analyzer is used to analyze a model representation of the bytecode to identify and generate specifications of authorization patterns associated with the request handler.

Patent Agency Ranking