AnalyzingClassifiers: Fisher Vectors and Deep Neural Networks
fv与dnn特征进行比较,得出的结论是fv比较偏向上下文信息,而dnn只比较关注物体本身信息。
TowardsDropout Training for Convolutional Neural Networks
除了dropout神经元之外,是否还可以dropoutpool, fully connection, convolution ?
该文章做了一些对比。
这个是比较好的一个工程点。
Rethinkingthe Inception Architecture for Computer Vision
5x5conv->3x3 conv+3x3 fc :we end up with a net 18 25× reduction of computation,resulting in a relative gain of 28% by this factorization
Instead,we argue that the auxiliary classifiers act as regularizer
Thisis supported by the fact that the main classifier of the network performsbetter if the side branch is batch-normalized [7] or has a dropout layer. Thisalso gives a weak supporting evidence for the conjecture that batchnormalization acts as a regularizer
Inaddition, gradient clipping [14] was found to be useful to stabilize thetraining
Googlenet特点: Much of the original gains of theGoogLeNet network [20] arise from a very generous use of dimension reduction.
网络结构:
结果对比:
MXNet:A Flexible and Efficient Machine Learning Library for Heterogeneous DistributedSystems
深度学习并行化训练相关代码
Attribute2Image:Conditional Image Generation from Visual Attributes
比较有意思的限定场景下的,attribute2image实验以及应用方向
1. search
2. image generation
Actions∼Transformations
Inthis paper, we propose a novel representation for actions by modeling action asa transformation which changes the state of the environment before the actionhappens (precondition) to the state after the action (effect)
蛮优雅的一个网络结构