ARMORER Navigation, spectrum and airborne systems v@armorer.org

Practice 06

Battlefield AI.

Machine learning on RF and sensor data, and the evidence to trust it.

Request a consultation

Also

FIG 07 Detection performance · ROC
Detection performance, illustrative Receiver operating characteristic curves. A learned classifier reaches a higher probability of detection than a fixed-threshold detector at the same false-alarm rate; the dashed diagonal is chance performance. The marked operating point is a five per cent false-alarm rate at roughly ninety per cent detection. 0 .25 .5 .75 1 0 .25 .5 .75 1 Pfa .05 / Pd .90 chance Pd P(false alarm)
A learned classifier against a fixed‑threshold detector at the same false‑alarm rate. Illustrative, not measured. LearnedThreshold
Classification
Interference, emitter and anomaly classification on RF and navigation data.
Data and labelling
Turning trials and monitoring data into training sets, with provenance recorded so a result can be traced back.
Edge deployment
What fits, and what it costs in power and latency, on the hardware actually carried.
Assurance
Test, evaluation and failure analysis for models whose output informs an operational decision.

Where we stop

Detection, classification and decision support, with a person accountable for the decision. We do not work on autonomous engagement.

A one-hour scoping call, no charge.

We will tell you whether this is a half‑day question or a programme.

v@armorer.org

Send an enquiry