Deep Learning

Project: In this course, we delved into our detailed exploration of the paper "Quaternion Convolutional Neural Networks" by Zhu et al. (2019). Through this study, our aim was to understand and replicate the use of QCNNs in image recognition tasks. We assessed their capabilities across various datasets, witnessing their potential firsthand.

Solution: In our pursuit of replicating the paper, we navigated through a series of challenges and solutions. We encountered discrepancies, notably the ambiguity in feature dimension specification - a pivotal aspect of the study. Our journey provided us with a first-hand understanding of the complexities in implementing academic papers. After substantial investigation, we employed tools like PyTorch to faithfully recreate the findings. The illustrations of our results, in alignment with the original claims, can be viewed on the left.