Home▸Physics & Mechanics▸Learning as Gradient Descent: 2D Loss Map

🧠 Learning as Gradient Descent: 2D Loss Map

Contour map of a loss surface with draggable start point and four optimizers.

Physics & Mechanics2DModerate60 FPS📱 Mobile-adapted⇄ 2D version

Details

CategoryPhysics & Mechanics

RenderingCanvas 2D

DifficultyModerate

Target FPS60

ControlsSliders for learning rate, momentum, batch size, steps per frame, class separation, noise, class axis angle, feature scale and L2 strength; choose a single optimizer or race all four; click or drag on the map to set the start weights; presets for ill-conditioned valley, too-high learning rate, momentum overshoot and noisy minibatches.

artificial intelligencemachine learninggradient descentoptimizersloss landscapelogistic regression

Related simulations

SPH FluidFluid simulation via SPH: pressure, viscosity, gravity and wall bouncing. Splash impulse…

Spring-Mass Cloth SimulationExplore the physics of cloth movement with this simulation! Adjust the spring stiffness and…

Car PhysicsRealistic driving with suspension, drift, nitro and particle effects. Navigate rough terrain. ⊞ All 3207 in Physics & Mechanics →

⚙ Under the hood

Interactive 2D contour map of the loss surface of a logistic-regression learner, where SGD, momentum, RMSProp and Adam race from a draggable start point while an inset shows the data and the decision boundary.

artificial intelligencemachine learninggradient descentoptimizersloss landscapelogistic regression

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install