Deep_Learning_For_Beginners
None
view repo
In this paper we investigate the performance of different types of rectified activation functions in convolutional neural network: standard rectified linear unit (ReLU), leaky rectified linear unit (Leaky ReLU), parametric rectified linear unit (PReLU) and a new randomized leaky rectified linear units (RReLU). We evaluate these activation function on standard image classification task. Our experiments suggest that incorporating a nonzero slope for negative part in rectified activation units could consistently improve the results. Thus our findings are negative on the common belief that sparsity is the key of good performance in ReLU. Moreover, on small scale dataset, using deterministic negative slope or learning it are both prone to overfitting. They are not as effective as using their randomized counterpart. By using RReLU, we achieved 75.68% accuracy on CIFAR100 test set without multiple test or ensemble.
READ FULL TEXT VIEW PDF
Rectified linear unit (ReLU) is a widely used activation function for de...
read it
The concept of nonlinearity in a Neural Network is introduced by an
act...
read it
Recently, selfnormalizing neural networks (SNNs) have been proposed wit...
read it
The classification task of sentences is very challenging because of the
...
read it
We study layered neural networks of rectified linear units (ReLU) in a
m...
read it
This work introduces a novel activation unit that can be efficiently emp...
read it
Rectified activation units (rectifiers) are essential for stateofthea...
read it
None
Convolutional neural network (CNN) has made great success in various computer vision tasks, such as image classification
(Krizhevsky et al., 2012; Szegedy et al., 2014), object detection(Girshick et al., 2014) and tracking(Wang et al., 2015). Despite its depth, one of the key characteristics of modern deep learning system is to use nonsaturated activation function (e.g. ReLU) to replace its saturated counterpart (e.g. sigmoid, tanh). The advantage of using nonsaturated activation function lies in two aspects: The first is to solve the so called “exploding/vanishing gradient”. The second is to accelerate the convergence speed.In all of these nonsaturated activation functions, the most notable one is rectified linear unit (ReLU) (Nair & Hinton, 2010; Sun et al., 2014). Briefly speaking, it is a piecewise linear function which prunes the negative part to zero, and retains the positive part. It has a desirable property that the activations are sparse after passing ReLU. It is commonly believed that the superior performance of ReLU comes from the sparsity (Glorot et al., 2011; Sun et al., 2014). In this paper, we want to ask two questions: First, is sparsity the most important factor for a good performance? Second, can we design better nonsaturated activation functions that could beat ReLU?
We consider a broader class of activation functions, namely the rectified unit family. In particular, we are interested in the leaky ReLU and its variants. In contrast to ReLU, in which the negative part is totally dropped, leaky ReLU assigns a noonzero slope to it. The first variant is called parametric rectified linear unit (PReLU) (He et al., 2015)
. In PReLU, the slopes of negative part are learned form data rather than predefined. The authors claimed that PReLU is the key factor of surpassing humanlevel performance on ImageNet classification
(Russakovsky et al., 2015) task. The second variant is called randomized rectified linear unit(RReLU). In RReLU, the slopes of negative parts are randomized in a given range in the training, and then fixed in the testing. In a recent Kaggle National Data Science Bowl (NDSB) competition
^{1}^{1}1Kaggle National Data Science Bowl Competition: https://www.kaggle.com/c/datasciencebowl, it is reported that RReLU could reduce overfitting due to its randomized nature.In this paper, we empirically evaluate these four kinds of activation functions. Based on our experiment, we conclude on small dataset, Leaky ReLU and its variants are consistently better than ReLU in convolutional neural networks. RReLU is favorable due to its randomness in training which reduces the risk of overfitting. While in case of large dataset, more investigation should be done in future.
In this section, we introduce the four kinds of rectified units: rectified linear (ReLU), leaky rectified linear (Leaky ReLU), parametric rectified linear (PReLU) and randomized rectified linear (RReLU). We illustrate them in Fig.1 for comparisons. In the sequel, we use to denote the input of th channel in th example , and to denote the corresponding output after passing the activation function. In the following subsections, we introduce each rectified unit formally.
Rectified Linear is first used in Restricted Boltzmann Machines
(Nair & Hinton, 2010). Formally, rectified linear activation is defined as:(1) 
Leaky Rectified Linear activation is first introduced in acoustic model(Maas et al., 2013). Mathematically, we have
(2) 
where is a fixed parameter in range . In original paper, the authors suggest to set to a large number like 100. In additional to this setting, we also experiment smaller in our paper.
Randomized Leaky Rectified Linear is the randomized version of leaky ReLU. It is first proposed and used in Kaggle NDSB Competition. The highlight of RReLU is that in training process,
is a random number sampled from a uniform distribution
. Formally, we have:(3) 
where
(4) 
In the test phase, we take average of all the in training as in the method of dropout (Srivastava et al., 2014) , and thus set to to get a deterministic result. Suggested by the NDSB competition winner, is sampled from . We use the same configuration in this paper.
In test time, we use:
(5) 
We evaluate classification performance on same convolutional network structure with different activation functions. Due to the large parameter searching space, we use two stateofart convolutional network structure and same hyper parameters for different activation setting. All models are trained by using CXXNET^{2}^{2}2CXXNET: https://github.com/dmlc/cxxnet.
The CIFAR10 and CIFAR100 dataset (Krizhevsky & Hinton, 2009) are tiny nature image dataset. CIFAR10 datasets contains 10 different classes images and CIFAR100 datasets contains 100 different classes. Each image is an RGB image in size 32x32. There are 50,000 training images and 10,000 test images. We use raw images directly without any preprocessing and augmentation. The result is from on single view test without any ensemble.
The network structure is shown in Table 1. It is taken from Network in Network(NIN)(Lin et al., 2013).
Input Size  NIN 

5x5, 192  
1x1, 160  
1x1, 96  
3x3 max pooling, /2 

dropout, 0.5  
5x5, 192  
1x1, 192  
1x1, 192  
3x3,avg pooling, /2  
dropout, 0.5  
3x3, 192  
1x1, 192  
1x1, 10  
8x8, avg pooling, /1  
10 or 100  softmax 
In CIFAR100 experiment, we also tested RReLU on Batch Norm Inception Network (Ioffe & Szegedy, 2015). We use a subset of Inception Network which is started from inception3a module. This network achieved 75.68% test accuracy without any ensemble or multiple view test ^{3}^{3}3CIFAR100 Reproduce code: https://github.com/dmlc/mxnet/blob/master/example/notebooks/cifar100.ipynb.
The task for National Data Science Bowl competition is to classify plankton animals from image with award of $170k. There are 30,336 labeled gray scale images in 121 classes and there are 130,400 test data. Since the test set is private, we divide training set into two parts: 25,000 images for training and 5,336 images for validation. The competition uses multiclass logloss to evaluate classification performance.
We refer the network and augmentation setting from team AuroraXie^{4}^{4}4Winning Doc of AuroraXie: https://github.com/auroraxie/KaggleNDSB, one of competition winners. The network structure is shown in Table 5. We only use single view test in our experiment, which is different to original multiview, multiscale test.
Input Size  NDSB Net 

3x3, 32  
3x3, 32  
3x3, max pooling, /2  
3x3, 64  
3x3, 64  
3x3, 64  
3x3, max pooling, /2  
split: branch1 — branch 2  
3x3, 96 — 3x3, 96  
3x3, 96 — 3x3, 96  
3x3, 96 — 3x3, 96  
3x3, 96  
channel concat, 192  
3x3, max pooling, /2  
3x3, 256  
3x3, 256  
3x3, 256  
3x3, 256  
3x3, 256  
SPP (He et al., 2014) {1, 2, 4}  
flatten  
fc1  
fc2  
121  softmax 
Table 3 and 4 show the results of CIFAR10/CIFAR100 dataset, respectively. Table 5 shows the NDSB result. We use ReLU network as baseline, and compare the convergence curve with other three activations pairwisely in Fig. 2, 3 and 4, respectively. All these three leaky ReLU variants are better than baseline on test set. We have the following observations based on our experiment:
Not surprisingly, we find the performance of normal leaky ReLU () is similar to that of ReLU, but very leaky ReLU with larger is much better.
On training set, the error of PReLU is always the lowest, and the error of Leaky ReLU and RReLU are higher than ReLU. It indicates that PReLU may suffer from severe overfitting issue in small scale dataset.
The superiority of RReLU is more significant than that on CIFAR10/CIFAR100. We conjecture that it is because the in the NDSB dataset, the training set is smaller than that of CIFAR10/CIFAR100, but the network we use is even bigger. This validates the effectiveness of RReLU when combating with overfitting.
For RReLU, we still need to investigate how the randomness influences the network training and testing process.
Activation  Training Error  Test Error 
ReLU  0.00318  0.1245 
Leaky ReLU,  0.0031  0.1266 
Leaky ReLU,  0.00362  0.1120 
PReLU  0.00178  0.1179 
RReLU ()  0.00550  0.1119 
Activation  Training Error  Test Error 
ReLU  0.1356  0.429 
Leaky ReLU,  0.11552  0.4205 
Leaky ReLU,  0.08536  0.4042 
PReLU  0.0633  0.4163 
RReLU ()  0.1141  0.4025 
Activation  Train LogLoss  Val LogLoss 
ReLU  0.8092  0.7727 
Leaky ReLU,  0.7846  0.7601 
Leaky ReLU,  0.7831  0.7391 
PReLU  0.7187  0.7454 
RReLU ()  0.8090  0.7292 
In this paper, we analyzed four rectified activation functions using various network architectures on three datasets. Our findings strongly suggest that the most popular activation function ReLU is not the end of story: Three types of (modified) leaky ReLU all consistently outperform the original ReLU. However, the reasons of their superior performances still lack rigorous justification from theoretic aspect. Also, how the activations perform on large scale data is still need to be investigated. This is an open question worth pursuing in the future.
We would like to thank Jason Rolfe from DWave system for helpful discussion on test network for randomized leaky ReLU.
Proceedings of the 14th International Conference on Artificial Intelligence and Statistics. JMLR W&CP Volume
, volume 15, pp. 315–323, 2011.The Journal of Machine Learning Research
, 15(1):1929–1958, 2014.
Comments
There are no comments yet.