Inception-v4, Inception-ResNet and the Impact of Residual Connections

Authors: Szegedy, C., Ioffe, S., Vanhoucke, V., & Alemi, A. A. Year: 2016 Venue: AAAI ArXiv: https://arxiv.org/abs/1602.07261

Summary

Inception-v4 combines the Inception architecture with residual connections. It achieves state-of-the-art on ImageNet and demonstrates the power of combining multiple architectural innovations.

Key Concepts

• Inception modules with multiple pathways
• Residual connections for gradient flow
• Large-scale architecture design
• Multi-scale feature extraction
• Efficient computation

Impact

Inception-v4 showed that combining residual connections with Inception modules improves both accuracy and training stability, influencing subsequent architecture designs.

Related Papers

• Inception: Going Deeper with Convolutions
• ResNet: Deep Residual Learning
• MobileNet: Efficient Architectures

---

Citation:

@inproceedings{szegedy2016inception,
  title={Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning},
  author={Szegedy, Christian and Ioffe, Sergey and Vanhoucke, Vincent and Alemi, Alexander A},
  booktitle={AAAI Conference on Artificial Intelligence},
  year={2016}
}