Roland Tian's Projects
SE-ResNeXt 50 vs 101 layer skin lesion classifier without pretraining, for melanoma vs nevus vs seborrheic keratosis
Skin lesion classifiers for melanoma vs nevus vs seborrheic keratosis, comparing Squeeze & Excitation, SE-ResNeXt, Inception-ResNetV2 and Neural Architecture Search convolutional neural networks
Movie review sentiment LSTM RNN on Amazon SageMaker cloud
NasNet, SENet, SEResNext, MobileNet, and VGG CNN dog breed classified with transfer learning in PyTorch. 89% accuracy predicting from 133 dog breed classes after 20 epochs.