Build feedforward neural networks layer by layer. Train on simple datasets and watch weights, activations, and loss converge in real time.
Trace a feedforward network from its input layer through hidden units to its predictions while training changes the connections. You can assemble layers, run a simple dataset through them, and observe weights, unit activations, and loss as learning progresses. The display connects normally abstract quantities to the network's behavior: activations show the signals produced for particular inputs, while loss summarizes prediction error. The common surprise is that a falling training loss does not prove that the network discovered the intended rule or will predict unseen examples correctly.
Each example moves through the layers in one direction, from input to output. A unit combines values arriving through weighted connections and applies its activation function to produce the value sent onward. The visualization exposes these intermediate activations and weights, allowing you to relate a prediction to particular paths through the network rather than seeing only the final output.
Training first compares the network's predictions with the dataset's targets to obtain a loss. Backpropagation determines how changes to individual weights would affect that loss, after which the weights are adjusted and another training step can run. Watching successive steps reveals whether loss is converging, oscillating, or barely changing, and whether hidden units are developing distinct activation patterns.
No. It means the current parameters fit the examples according to the loss being displayed. A network may fit a small dataset while behaving poorly between or beyond those examples. Treat the visualization as evidence about optimization and internal activity, not as proof of generalization, causality, or a uniquely correct network structure.
Additional units create more activations and weighted connections, and several units may learn overlapping or interchangeable roles. The network may gain capacity while losing a simple visual explanation. More units also do not guarantee lower loss in every observed training interval, because successful training depends on how the parameters evolve, not only on how many parameters exist.
The behavioral notes were checked against the browser implementation. Standards and primary references below define the relevant format, formula, or platform behavior.
Calculate percentages, percentage change, and what percent one number is of another with step-by-step breakdowns.
Calculate tips and split bills between multiple people with customizable tip percentages and rounding options.
Calculate monthly payments, total interest, and generate amortization schedules for loans and mortgages.
Calculate Body Mass Index from height and weight with category classification and healthy range guidance.
Calculate sale prices, savings amounts, and final costs after single or stacked percentage discounts.
Compute mean, median, mode, standard deviation, variance, and other descriptive statistics from a data set.