Figure 1. Overview of our framework of using a search method
(e.g., Reinforcement Learning) to search for better data augmen-
tation policies. A controller RNN predicts an augmentation policy
from the search space. A child network with a fixed architecture
is trained to convergence achieving accuracy R. The reward R will
be used with the policy gradient method to update the controller
so that it can generate better policies over time.
probability of applying ShearX is 0.9, and when applied,
has a magnitude of 7 out of 10. We then apply Invert with
probability of 0.8. The Invert operation does not use the
magnitude information. We emphasize that these operations
are applied in the specified order.
Figure 2. One of the policies found on SVHN, and how it can be
used to generate augmented data given an original image used to
train a neural network. The policy has 5 sub-policies. For every
image in a mini-batch, we choose a sub-policy uniformly at ran-
dom to generate a transformed image to train the neural network.
Each sub-policy consists of 2 operations, each operation is associ-
ated with two numerical values: the probability of calling the op-
eration, and the magnitude of the operation. There is a probability
of calling an operation, so the operation may not be applied in that
mini-batch. However, if applied, it is applied with the fixed mag-
nitude. We highlight the stochasticity in applying the sub-policies
by showing how one image can be transformed differently in dif-
ferent mini-batches, even with the same sub-policy. As explained
in the text, on SVHN, geometric transformations are picked more
often by AutoAugment. It can be seen why Invert is a commonly
selected operation on SVHN, since the numbers in the image are
invariant to that transformation.
The operations we used in our experiments are from PIL,
a popular Python image library.
1
For generality, we consid-
ered all functions in PIL that accept an image as input and
1
https://pillow.readthedocs.io/en/5.1.x/
output an image. We additionally used two other promis-
ing augmentation techniques: Cutout [12] and SamplePair-
ing [24]. The operations we searched over are ShearX/Y,
TranslateX/Y, Rotate, AutoContrast, Invert, Equalize, So-
larize, Posterize, Contrast, Color, Brightness, Sharpness,
Cutout [12], Sample Pairing [24].
2
In total, we have 16
operations in our search space. Each operation also comes
with a default range of magnitudes, which will be described
in more detail in Section 4. We discretize the range of mag-
nitudes into 10 values (uniform spacing) so that we can use
a discrete search algorithm to find them. Similarly, we also
discretize the probability of applying that operation into 11
values (uniform spacing). Finding each sub-policy becomes
a search problem in a space of (16× 10 × 11)
2
possibilities.
Our goal, however, is to find 5 such sub-policies concur-
rently in order to increase diversity. The search space with 5
sub-policies then has roughly (16× 10×11)
10
≈ 2.9×10
32
possibilities.
The 16 operations we used and their default range of val-
ues are shown in Table 1 in the Appendix. Notice that there
is no explicit “Identity” operation in our search space; this
operation is implicit, and can be achieved by calling an op-
eration with probability set to be 0.
Search algorithm details: The search algorithm that
we used in our experiment uses Reinforcement Learning,
inspired by [71, 4, 72, 5]. The search algorithm has two
components: a controller, which is a recurrent neural net-
work, and the training algorithm, which is the Proximal
Policy Optimization algorithm [53]. At each step, the con-
troller predicts a decision produced by a softmax; the pre-
diction is then fed into the next step as an embedding. In
total the controller has 30 softmax predictions in order to
predict 5 sub-policies, each with 2 operations, and each op-
eration requiring an operation type, magnitude and proba-
bility.
The training of controller RNN: The controller is
trained with a reward signal, which is how good the policy is
in improving the generalization of a “child model” (a neural
network trained as part of the search process). In our exper-
iments, we set aside a validation set to measure the gen-
eralization of a child model. A child model is trained with
augmented data generated by applying the 5 sub-policies on
the training set (that does not contain the validation set). For
each example in the mini-batch, one of the 5 sub-policies is
chosen randomly to augment the image. The child model
is then evaluated on the validation set to measure the accu-
racy, which is used as the reward signal to train the recurrent
network controller. On each dataset, the controller samples
about 15,000 policies.
Architecture of controller RNN and training hyper-
parameters: We follow the training procedure and hyper-
parameters from [72] for training the controller. More con-
2
Details about these operations are listed in Table 1 in the Appendix.
剩余13页未读，继续阅读
IRUIRUI__
- 粉丝: 553
- 资源: 55
我的内容管理
展开
最新资源
- 解决本地连接丢失无法上网的问题
- BIOS报警声音解析：故障原因与解决方法
- 广义均值移动跟踪算法在视频目标跟踪中的应用研究
- C++Builder快捷键大全：高效编程的秘密武器
- 网页制作入门：常用代码详解
- TX2440A开发板网络远程监控系统移植教程：易搭建与通用解决方案
- WebLogic10虚拟内存配置详解与优化技巧
- C#网络编程深度解析：Socket基础与应用
- 掌握Struts1：Java MVC轻量级框架详解
- 20个必备CSS代码段提升Web开发效率
- CSS样式大全：字体、文本、列表样式详解
- Proteus元件库大全：从基础到高级组件
- 74HC08芯片：高速CMOS四输入与门详细资料
- C#获取当前路径的多种方法详解
- 修复MySQL乱码问题：设置字符集为GB2312
- C语言的诞生与演进：从汇编到系统编程的革命
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
