Augmenting supervised neural networks with unsupervised objectives for large-scale image classification
왠지 중요한것 같아서라기보다 CVPR2017 best paper가 인용한 논문이 한국사람의 논문이라서 봄
학습할 때 gradient vanishing, explosion등 문제를 해결하는 방법의 일종인듯.. 괜히읽었다. Augmented autoencoder의 variation 세가지를 제안함.
목차
Abstract
we investigate joint supervised and unsupervised learning in a large-scale setting by augmenting existing neural networks with decoding pathways for reconstruction.
1. Introduction
Unsupervised로 pre-training 한참 하다가 요즘엔 아무도 안한다. There are many (supervised-learning) technics which were not limited by the modeling assumptions of unsupervised methods.
Supervised랑 unsupervised를 엮으려는 시도들이 조금 있었으나 대량의 데이터에 관한 것이 아니었다. Zhao et al.[1]이 Zeiler et al.[2]의 “unpooling” operator를 사용해 “what-where” autoencoder(SWWAE)를 제안했고 CIFAR with extended unlabeled data에서 괜찮은 결과를 얻었으나, 그보다 더 큰 데이터에서는 그렇지 못했다.
Classification network의 일부를 떼어내고 그것을 그대로 뒤집(은 후 augmentation을 더해 얻)은 것을 (reconstructive) decoding pathway로 쓰겠다. 이 autoencoder를 finetuning하면, 앞단(원래 classification에 쓰이던 부분)의 성능도 향상된다.
We will provide insight on the importance of the pooling switches and the layer-wise reconstruction loss.
2. Related work
- sparse coding
- dictionary learning
- autoencoder
3.Methods
3.1. Unsupervised loss for intermediate representations
원래 보통의 loss는 $$\frac{1}{N}\sum^N_{i=1} C(x_i, y_i), C(x,y) = l(a_{L+1}, y)$$ where
- \(a_L\) :\(L\)th layer’s output,
- \(y\) : ground truth,
- \(l\) : cross-entropy loss,
- \(C\) : classification loss.
이지만, net 중간중간에 loss를 만들도록 적당히 뭔가를 넣으면(3.2에 나온다)
$$\frac{1}{N}\sum^N_{i=1} (C(x_i, y_i)+ \lambda U(x_i))$$
이렇게 할 수 있다.그냥 뻔한 이야기
3.2. Network augmentation with autoencoders
아래가 저자들이 제안한 것.
아래는 ladder net[3]
4. Experiments
- ImageNet ILSVRC 2012 dataset
- mainly based on the 16-layer VGGNet
- partially used AlexNet
4.1. Training procedure
Steps
- 초기화 : encoding pathway는 pretrained, decoding은 Gaussian
- encoding을 고정시킨 후 SAE-layerwise net을 학습시킴
- SAE-first/all은 SAE-layerwise를 pretrained net으로 해서 학습시킴
- encoding/decoding 모두 reduce learning rate으로 finetuning
SGD momentum 0.9 꽤나 크게 줬네? (We found the learning rate annealing not critical for SAE-layerwise pretraining. Proper base learning rates could make it sufficiently converged within 1 epoch.)
4.2. Image reconstruction via decoding pathways
Conclusion
- the pooling switch connections between the encoding and decoding pathways were helpful, but not critical for improving the performance of the classification network in large- scale settings
- the decoding pathways mainly helped the supervised objective reach a better optimum
- the layer-wise reconstruction loss could effectively regularize the solution to the joint objective.
References
- ↑ Zhao, J., Mathieu, M., Goroshin, R., and Lecun, Y. Stacked what-where auto-encoders. arXiv:1506.02351, 2015.
- ↑ Zeiler, M., Taylor, G., and Fergus, R. Adaptive deconvolu- tional networks for mid and high level feature learning. In ICCV, 2011.
- ↑ Rasmus, A., Valpola, H., Honkala, M., Berglund, M., and Raiko, T. Semi-supervised learning with ladder network. In NIPS, 2015.

