Platform for Development of AI based Network Traffic Analytics

Enable fast and focused development/deployment of ML/DL-based solutions. Advantages are: (a) Time savings in Development phase – it includes building blocks (depicted in the diagram) that are specific for network traffic and essential for a pipeline to build high quality model development. (b) Benchmark accuracy of developed models – It provides standard ways to compute and evaluate efficacy of models. It becomes easier to compare newly developed ML/DL model performance with existing models. Experimentation with different datasets is easily reproducible. (c) Solution is highly scalable.

Some of these are: (i) Network flow feature extraction module from pcap files; (ii) Ability to create datasets in a flexible manner; (iii) Perform hyper-parameter optimization, (iv) Having available various pre-configured templates for ML/DL pipeline creation etc.

Supports ML/DL life cycle for data processing, model creation and deployment with necessary tools to develop AI models: Integrated various s/w components with Kubeflow and these can run on Kubernetes; ENTA SDK enables implementation of use cases; Pipeline components to define ML/DL workflows; Feature extraction – Tranalyzer, NFStream, CIC Flowmeter and User defined features; Dataset creation – MinIO and FEAST; Model training and optimization - Training Operators supports Tensorflow, PyTorch, XGBoost, Scikit-learn etc.; Evaluation; Off-the-shelf tools can be integrated with the framework; The platform can be deployed locally or in a cloud.

Environment needs to support for Kubernetes on a Linux platform so that Kubeflow can be run.

Solution developers for cyber security tools, Deep Packet Inspection vendors

Solana Networks will provide the service to enable reuse of this ENTA platform solution in the customer environment during trial period. Licensing agreement for the solution can be arranged after the initial trial period.