Fractional Max-Pooling

Benjamin Graham

2014

21 references

Abstract

Convolutional networks almost always incorporate some form of spatial pooling, and very often it is alpha times alpha max-pooling with alpha=2. Max-pooling act on the hidden layers of the network, reducing their size by an integer multiplicative factor alpha. The amazing by-product of discarding 75% of your data is that you build into the network a degree of invariance with respect to translations and elastic distortions. However, if you simply alternate convolutional layers with max-pooling layers, performance is limited due to the rapid reduction in spatial size, and the disjoint nature of the pooling regions. We have formulated a fractional version of max-pooling where alpha is allowed to take non-integer values. Our version of max-pooling is stochastic as there are lots of different ways of constructing suitable pooling regions. We find that our form of fractional max-pooling reduces overfitting on a variety of datasets: for instance, we improve on the state-of-the art for CIFAR-100 without even using dropout.

View Paper PDF

2 repositories

17 references

Code References

▶ pytorch/pytorch

2 files

▶ torch/nn/functional.py

L477 http://arxiv.org/abs/1412.6071

L596 http://arxiv.org/abs/1412.6071

▶ torch/nn/modules/pooling.py

L954 https://arxiv.org/abs/1412.6071

L1041 https://arxiv.org/abs/1412.6071

▶ tensorflow/tensorflow

4 files

▶ tensorflow/core/kernels/fractional_pool_common.cc

L58 // [1]: https://arxiv.org/abs/1412.6071

▶ tensorflow/core/kernels/fractional_pool_common.h

L51 // http://arxiv.org/abs/1412.6071

▶ tensorflow/go/op/wrappers.go

L18279 // Graham, Fractional Max-Pooling](http://arxiv.org/abs/1412.6071) for

L18453 // Graham, Fractional Max-Pooling](http://arxiv.org/abs/1412.6071) for

L18543 // [Benjamin Graham, Fractional Max-Pooling](http://arxiv.org/abs/1412.6071)

▶ tensorflow/python/ops/nn_ops.py

L6130 [Graham, 2015](https://arxiv.org/abs/1412.6071)

L6131 ([pdf](https://arxiv.org/pdf/1412.6071.pdf))

L6219 [Graham, 2015](https://arxiv.org/abs/1412.6071)

L6220 ([pdf](https://arxiv.org/pdf/1412.6071.pdf))

L6318 [Graham, 2015](https://arxiv.org/abs/1412.6071)

L6319 ([pdf](https://arxiv.org/pdf/1412.6071.pdf))

L6375 [Graham, 2015](https://arxiv.org/abs/1412.6071)

L6376 ([pdf](https://arxiv.org/pdf/1412.6071.pdf))

Link copied to clipboard!