QELM

Tradi­tio­nal approa­ches of super­vi­sed learning equip­ped with Neural Networks (NN) archi­tec­tures require itera­tive adjus­t­ment of many parame­ters (often throug­hout deeply inter­con­nec­ted layers) while training the algorithm, which can be compu­ta­tio­nally expen­sive. Extreme Learning Machi­nes (ELMs), a form of reser­voir compu­ting, was created to offer a faster alter­na­tive: ELMs use a randomly initia­li­zed “reser­voir” layer that projects inputs into a high-dimen­sio­nal space, with only the final readout layer trained. Quantum Extreme Learning Machi­nes (QELMs) adapted this idea to the quantum hardware. A quantum reservoir—qubits evolving under fixed random initia­li­zed quantum dynamics—processes input data through super­po­si­tion and entan­gle­ment, creating a more complex and high-dimen­sio­nal feature space. Only classi­cal readout weights are trained, avoiding controll­a­bi­lity issues on the quantum hardware, making it suited for timese­ries forecas­ting applications.