Physics-Informed vs Raw MLP Classifier
Controls
Pause
Clear
Auto-launch random
Speed ×
1.0
Manual launch
v (m/s)
30
θ (deg)
45
Launch current v,θ
—
Live classifier
Two models train in parallel on the
same samples
(seeded).
Steps / tick
120
Learning rate
0.25
Batch size
64
Tick (ms)
50
Seed
Reseed & Rebuild
physics: acc — | loss —
raw MLP: acc — | loss —
Trajectories — Physics (top) & Raw MLP (bottom)
Same shots; colors from each model
Short (R < 100 m)
Medium (100–200 m)
Long (≥ 200 m)
Physics‑informed predictions
Raw MLP predictions
Decision fields — side by side
Dots are the same training samples
Physics‑informed Softmax (feature: R̂ = v² sin(2θ)/g)
Raw MLP ([1, v/60, sinθ, cosθ], hidden=16)
Physics confusion
Raw-MLP confusion
Validation Accuracy over Time
Live comparison of model performance
━━
Physics-informed
━━
Raw MLP