CyberNeuro-RT

S01 Cutting-Edge Unsupervised ML

• Scalable Unsupervised Outlier Detection (SUOD)
• 6 ML algorithm ensemble
• Model approximation for complex models
• Variational Autoencoder (VAE), trained to minimize reconstruction error of initial input and reconstructed output

S02 Proprietary Pipeline Adapts to Any Dataset

75x dataset growth in under 2 months.

• Existing dataset ingestion: proprietary system enables ingestion of any existing network capture dataset with flexible support for any labelling system
• From-the-wild zero day sampling: system enables capturing and simulation of novel threats for additional data sampling
• Data generation via simulation: ThreatATI database and proprietary ingestion system enable sampling and augmentation for cataloged threats from proprietary and public threat databases
• Follow threats home with dark web tracking

S03 At-the-edge Neuromorphic Processing

Two offerings from the leading neuromorphic developers: Intel and Brainchip.

• Small form factor, magnitudes less power consumption than GPU
• On-chip learning for deployment network specific attack detection
• Intel Loihi
• Brainchip Akida

S04 Dashboards Minimize Operator Fatigue

A robust, multi-faceted, user-friendly cyber analyst dashboard prevents operator fatigue that allows cyber attacks to happen. Large numbers of false alarms cause real threats to be missed, and false alarms fatigue the cyber analyst, further increasing the risk of missed threats.

• AI based false alarms are minimized, trained for minimal false positive rate
• Possible threats are ranked by importance and confidence
• Only the most relevant and likely alarms are actioned upon