As an undergraduate researcher working with Dr. Fan Chen, I explored quantum convolutional neural networks (QCNNs). I started with a literature review to help the team decide which QML architectures were worth pursuing, then presented my findings, walking through the models, the core concepts, and where they apply across industries.

For the hands-on work, I built a hybrid quantum-classical QCNN in TensorFlow Quantum and Cirq to classify downsampled MNIST digits (8×8 pixels, 8 qubits). It uses angle encoding with layered, parameterized convolution and pooling circuits. I benchmarked it against a classical CNN with a matched parameter count across 5 random seeds, comparing accuracy, convergence speed, and training time. The QCNN reached comparable accuracy with roughly ~40% fewer trainable parameters.

I also ran ablation studies on encoding strategy (angle vs. amplitude), circuit depth, and simulated depolarizing noise to see how well the model holds up under hardware-like error.

Along the way I synthesized 10+ papers on QCNNs and variational quantum circuits to guide the architecture design, and shared the results in internal briefings and a research poster session