Data encoding into qubits is straightforward

Data encoding is one of the most critical and underappreciated bottlenecks in QML. There is no single correct encoding — amplitude encoding, angle encoding, basis encoding, and IQP-style encoding each make different trade-offs between expressiveness, circuit depth, and the number of qubits required. Amplitude encoding can store 2^n values in n qubits but requires O(2^n) gates to prepare. Angle encoding uses fewer gates but wastes the exponential state space. The choice of encoding fundamentally shapes what the model can learn and whether any quantum advantage is possible.

Data encoding requires parameterized circuits with real dataset inputs — beyond what a browser simulator can demonstrate fairly. See the Schuld and Petruccione (2021) reference for a rigorous treatment.