🧲 2D Matching Networks: Attention-Weighted Few-Shot Classification, Softmax Temperature and the Nearest-Neighbour Limit
Embedding-plane view of a matching network: support points, a draggable query, attention-weighted class probabilities, an accuracy-versus-temperature curve and the Bayes-optimal ceiling, verified by the 1-NN and uniform limits, a finite-difference gradient and permutation invariance.
⚙ Under the hood
Embedding-plane view of a matching network: support points, a draggable query, attention-weighted class probabilities, an accuracy-versus-temperature curve and the Bayes-optimal ceiling, verified by the 1-NN and uniform limits, a finite-difference gradient and permutation invariance.
Matching NetworksFew-Shot LearningAttentionSoftmax TemperatureNearest NeighbourEpisodic TrainingMetric LearningBayes Optimal Classifier
2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install