Skip to main content

> ML_LITERATURE // HE-2015-DELVING-DEEP-INTO-RECTIFIERS-KAIMING-INIT_v1.0

Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification (PReLU / Kaiming Init)

Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun · IEEE International Conference on Computer Vision (ICCV) (2015)

foundational2015foundationalthirdPartyReproduced
DOI: 10.1109/ICCV.2015.123Back to Literature

Principal Contribution

Derived He/Kaiming initialization tailored specifically for non-symmetric rectifier nonlinearities (ReLU) by accounting for zero-clipping (Var(w) = 2/fan-in), surpassing human performance on ImageNet.

Operational Relevance

Serves as qualified theoretical and empirical reference for task-image-classification.

Assumptions

  • Mathematical convexity, regularity, and empirical consistency hold across problem configurations

Limitations

  • Theoretical bounds and benchmark saturation characteristics vary across model architectures and training scales

Connected Algorithms, Architectures & Tools

Related Algorithms:
Related Architectures:
Implementing Libraries: