Parallel Processing 기반 Transformer 블록의 계층적 구조 설계
Transformers From Scratch: Assembling the Block Behind GPT
Residual Connection 도입으로 Loss 1.70 달성 및 층 심화 시 학습 안정성 확보
Time When More Layers Meant Worse Model ... Birth Of Residual
Spatial 구조 보존과 Weight Sharing을 통한 이미지 인식 효율 최적화
How CNNs Work — From Convolution Kernels to ResNet
Residual Connection을 통한 Encoder-Decoder Attention 최적화
Understanding Transformers – Part 16: Preparing for Output Prediction with Residual Connections
Residual Connection을 통한 Transformer Encoder의 학습 효율 최적화
Understanding Transformers Part 10: Final Step in Encoding