Cisco Technology, Inc.

Thomas Michel-Ange Feltin

Frank Brockners

Wenqin Shao

Parisa Foroughi

Jerome Henry

Robert E Barton

Indermeet Singh Gandhi

Andre Surcouf

Matthias Falkner

Leo Marche

Techniques and mechanisms are described for enabling a user to run heavy deep learning workloads on standard edge networks without off-loading computation to a cloud, leveraging the available edge computing resources, and efficiently partitioning and distributing a Deep Neural Network (DNN) over a network. The techniques enable the user to split a workload into multiple parts and process the workload on a set of smaller, less capable compute nodes in a distributed manner, without compromising on performance, and while meeting a Service Level Objective (SLO).

Partitioning and placement of models of models at a network edge

Novel techniques and mechanisms enable processing of heavy deep learning workloads on standard edge network devices to optimize the overall inference throughput of the network while meeting Service Level Agreement(s) (SLAs). The techniques can include receiving a deep learning model, determining a graph structure of the deep learning model including neurons organized in layers (the layers including an input layer, a plurality of hidden layers, and an output layer), assigning to a first IP subnet, at least a part of a first hidden layer of the plurality of hidden layers, assigning to a second IP subnet, at least a part of a second hidden layer of the plurality of hidden layers, and deploying the parts of the first and second hidden layers to edge devices as containerized applications with assigned IP addresses, which may be hidden from the user and/or third party application.

Hidden-layer routing for disaggregated artificial neural networks

Techniques and mechanisms for automatically identifying counters/features of a network component that are related to a state change (or event) for the network component or for the network itself. For example, using data obtained from the network component around a time of the state change, delta-averages for the counters/features around the time of the state change may be determined. The delta-averages may be utilized to determine which counters/features are most descriptive for a particular state change. Determining which counters/features are most descriptive may also include determining which counters/features are most relevant, i.e., counters/features that contribute most to preserving the manifold structure of the original data or counters/features with the highest or lowest correlation with the other counters/features in the data set. Thus, the techniques described herein provide for an approach to distill which counters/features contribute the most to a particular state change from a data driven perspective.

Data-driven identification and selection of features related to a state change of a network component

Techniques and mechanisms for automatically identifying counters/features of a network component that are related to a state change (or event) for the network component or for the network itself. For example, using data obtained from the network component around a time of the state change, delta averages for the features around the time of the state change may be determined. The delta averages may be utilized to determine which counters/features are most descriptive for a particular state change. The counter/features that are the most descriptive for a particular state change is as important as the change detection itself. This is especially true since in a case of an event/state change occurring, a large amount of counters/features may react to the state change or event. Thus, the techniques described herein provide for an approach to distill which counters/features contribute the most to a particular state change from a data driven perspective.

Growing community of inventors

Paris, France

Thomas Michel-Ange Feltin