main concepts scientists sections
SimplePopularAdvanced

How to read an enemy in the sky in seconds ✎ express

Original: "An Evidential Reasoning Approach for Aerial Target Classification and Intent Prediction"
· Tenzing Thiley Bhutia, Subash Kumaraguru, Devaprakash Muniraj
arXiv:2607.14606 · 2026-07-16 · arXiv non-exclusive · independent analysis · 1 min
eess.SY cs.SY
Combat algorithms learn to make decisions from scraps of data, honestly acknowledging uncertainty.

In combat, a pilot needs to instantly understand what an object is and what its intentions are. Conventional systems wait for complete radar data — like a goalkeeper who only reacts after the shot. The new algorithm works like an experienced goalkeeper: from the attacker's first moves, it predicts the target's maneuver.

The algorithm collects fragmentary signals and gradually refines its conclusion. It doesn't hide its doubts: when data is scarce, entropy serves as a measure of uncertainty. The system honestly reports, 'I'm not sure' — like a goalkeeper who doesn't dive where the ball might not even be.

In air combat, as in soccer, false confidence is deadlier than doubt.

Intentions — attack or evasion — are inferred by summing signals over time (just as in photometry faint flashes are combined into one). Tests on eight training targets yielded 88% accuracy for type and 93% for intentions, and it's faster than waiting for full data collection. Notably, the system was trained on artificial examples — real combat records are classified.

Whole graph →

To train the system, engineers had to generate artificial combat scenarios — real data is classified.

Scientists
Ludwig BoltzmannRudolf ClausiusJacob BekensteinStephen HawkingJames Clerk MaxwellEdward Witten
Concepts
evidential neural networks information criterion forecast skill prediction training data training probabilistic forecast prediction accuracy Machine Learning Explainable AI predictive learning neural networks random forest feature importance prior distribution speech recognition decision making labeled data confidence interval lead time classification