Opening your workspace…
Opening your workspace…
AN INTERACTIVE MACHINE LEARNING LAB
Explore neural networks, image filters and learning algorithms through small, real experiments. No account. No experiment database.
Start with a Neuron →Artificial Neuron → Activation Functions → Forward Propagation → Backpropagation → Gradient Descent
Image Tensors → Convolution → Pooling → CNN Explorer → ResNet
Vanilla RNN → Backpropagation Through Time → LSTM → GRU → Encoder–Decoder
Change inputs, weights and bias to see a neuron respond.
BeginnerCompare activation curves and their local derivatives.
BeginnerFollow weighted sums and activations through a 2–2–1 network.
BeginnerInspect the chain rule and apply a real gradient update.
BeginnerCompare the cost of an incorrect prediction.
BeginnerTurn logits into normalized class probabilities.
Follow actual optimization steps across a loss landscape.
IntermediateCompare SGD, momentum, Nesterov and adaptive updates.
IntermediateSee how repeated derivatives shrink or amplify a signal.
IntermediateCompare zero, small, large, Xavier and He initialization.
Train a small network on nonlinear data in your browser.
IntermediateExplore why nonlinear problems need hidden representations.
IntermediateChange L1, L2 and early stopping while watching validation loss.
IntermediateTrain with seeded neuron masks and compare generalization.
Update a linear classifier one example at a time.
IntermediateFit a line and measure the residual error.
IntermediateTrain a probabilistic classifier and inspect its boundary.
IntermediateMove a query point and inspect its neighbors and votes.
IntermediateCombine class priors with feature likelihoods.
IntermediateInspect splits chosen by Gini impurity or entropy.
IntermediateCombine bootstrap trees and inspect their votes.
IntermediateExplore soft margins and linear or RBF kernels.
IntermediateAlternate nearest-center assignment with centroid updates.
IntermediateDiscover density clusters, core points and noise.
IntermediateProject data onto its greatest-variance axis.
IntermediateChange a threshold to explore confusion matrices and ROC.
IntermediateCompare training and validation error as complexity changes.
Inspect pixels, spatial dimensions and channel counts.
IntermediateMultiply an editable kernel with each image window.
IntermediateTry edge detection, blur, sharpening and embossing.
IntermediateChange the movement of the kernel and output dimensions.
IntermediateKeep a visible zero border and calculate the resulting shape.
IntermediateCompare max, average and min pooling on real values.
IntermediateFollow a small image through convolution, ReLU, pooling and scores.
IntermediateInspect intermediate spatial activations.
IntermediateCompare shared filters with independent dense weights.
Inspect the early convolutional architecture for digits.
IntermediateExplore the model’s convolutional and dense stages.
IntermediateInspect repeated small convolution blocks.
IntermediateExplore parallel filters and channel concatenation.
IntermediateFollow a residual block and its identity shortcut.
IntermediateCompare depthwise and pointwise parameter counts.
Watch shared weights update hidden memory through time.
IntermediateInspect exact recurrent state derivatives backward through time.
IntermediateInspect forget, input and output gates along the cell-state highway.
IntermediateChange reset and update behavior in a gated recurrent unit.
IntermediateCompare forward and backward sequence context.
IntermediatePass an encoder context into a small recurrent decoder.
Train a bottleneck to reconstruct simple patterns.
IntermediateReconstruct clean patterns from locally generated noisy inputs.
IntermediateTrain with an actual activation sparsity penalty.
IntermediateExplore a spatial bottleneck and reconstruction.
IntermediateExplore mean, variance, reparameterization and the Gaussian KL term.
No account required. Local images and model files stay in your browser. Educational models demonstrate mechanisms; they are not production predictors.