Fraunhofer AISEC’s QML Playground puts a re-uploading quantum classifier in a browser tab

The new element in Fraunhofer AISEC’s QML Playground is the interface, not the quantum model. The project packages the data re-uploading variational classifier introduced by Pérez-Salinas et al. in 2020 into a browser application modeled on TensorFlow Playground, aimed at learners, educators, and researchers who want to inspect training behavior without setting up a simulator or quantum hardware. In that model, each layer applies trainable rotations, Rot(φ, θ, ω) = RZ(ω)·RY(θ)·RZ(φ), with angles set to w·x + b so the classical input is re-uploaded at every layer. The playground supports 1- and 2-qubit circuits with 1–15 layers, an optional entangling gate after each layer for the 2-qubit case, and Adam training with mini-batches and an adjustable learning rate. Architecture choices include CNOT, CZ, or no entangler, uploading data on all rotation angles or only on RY, and trainable or fixed data weights at w = π. Tasks are limited to 2-D problems. Classification covers seven synthetic datasets — circle, 3 circles, square, 4 squares, crown, tricrown, and wavy lines. Regression fits Fourier series of degree 1–5 with gaps in the training data. L1/L2 regularization is exposed to show overfitting and smoothness effects, and regression uncertainty is estimated by Monte Carlo sampling of noisy parameters, displayed as a ±2σ band. The README links the work to an IEEE DOI and an arXiv preprint, but the repository itself does not present new theoretical results. What you can inspect The circuit view is the core of the tool. Each gate shows its angle as a