深度神經(jīng)網(wǎng)絡(luò)可解釋性方法匯總，附Tensorflow代碼實(shí)現(xiàn)
理解神經(jīng)網(wǎng)絡(luò)：人們一直覺(jué)得深度學(xué)習(xí)可解釋性較弱。然而，理解神經(jīng)網(wǎng)絡(luò)的研究一直也沒(méi)有停止過(guò)，本文就來(lái)介紹幾種神經(jīng)網(wǎng)絡(luò)的可解釋性方法，并配有能夠在Jupyter下運(yùn)行的代碼連接。
Activation Maximization
通過(guò)激活最化來(lái)解釋深度神經(jīng)網(wǎng)絡(luò)的方法一共有兩種，具體如下：
1.1 Activation Maximization (AM)
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/1.1%20Activation%20Maximization.ipynb
1.2 Performing AM in Code Space
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/1.3%20Performing%20AM%20in%20Code%20Space.ipynb
Layer-wise Relevance Propagation
層方向的關(guān)聯(lián)傳播，一共有5種可解釋方法。Sensitivity Analysis、Simple Taylor Decomposition、Layer-wise Relevance Propagation、Deep Taylor Decomposition、DeepLIFT。它們的處理方法是：先通過(guò)敏感性分析引入關(guān)聯(lián)分?jǐn)?shù)的概念，利用簡(jiǎn)單的Taylor Decomposition探索基本的關(guān)聯(lián)分解，進(jìn)而建立各種分層的關(guān)聯(lián)傳播方法。具體如下：
2.1 Sensitivity Analysis
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/2.1%20Sensitivity%20Analysis.ipynb
2.2 Simple Taylor Decomposition
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/2.2%20Simple%20Taylor%20Decomposition.ipynb
2.3 Layer-wise Relevance Propagation
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/2.3%20Layer-wise%20Relevance%20Propagation%20%281%29.ipynb
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/2.3%20Layer-wise%20Relevance%20Propagation%20%282%29.ipynb
2.4 Deep Taylor Decomposition
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/2.4%20Deep%20Taylor%20Decomposition%20%281%29.ipynb
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/2.4%20Deep%20Taylor%20Decomposition%20%282%29.ipynb
2.5 DeepLIFT
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/2.5%20DeepLIFT.ipynb
Gradient Based Methods
基于梯度的方法有：反卷積、反向傳播， 引導(dǎo)反向傳播，積分梯度和平滑梯度這幾種。具體可以參考如下鏈接：
https://github.com/1202kbs/Understanding-NN/blob/master/models/grad.py
詳細(xì)信息如下：
3.1 Deconvolution
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/3.1%20Deconvolution.ipynb
3.2 Backpropagation
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/3.2%20Backpropagation.ipynb
3.3 Guided Backpropagation
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/3.3%20Guided%20Backpropagation.ipynb
3.4 Integrated Gradients
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/3.4%20Integrated%20Gradients.ipynb
3.5 SmoothGrad
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/3.5%20SmoothGrad.ipynb
Class Activation Map
類激活映射的方法有3種，分別為：Class Activation Map、Grad-CAM、 Grad-CAM++。在MNIST上的代碼可以參考：
https://github.com/deepmind/mnist-cluttered
每種方法的詳細(xì)信息如下：
4.1 Class Activation Map
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/4.1%20CAM.ipynb
4.2 Grad-CAM
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/4.2%20Grad-CAM.ipynb
4.3 Grad-CAM++
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/4.3%20Grad-CAM-PP.ipynb
Quantifying Explanation Quality
雖然每一種解釋技術(shù)都基于其自身的直覺(jué)或數(shù)學(xué)原理，但在更抽象的層次上確定好解釋的特征并能夠定量地測(cè)試這些特征也很重要。這里再推薦兩種基于質(zhì)量和評(píng)價(jià)的可解釋性方法。具體如下：
5.1 Explanation Continuity
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/5.1%20Explanation%20Continuity.ipynb
5.2 Explanation Selectivity
相關(guān)代碼如下：
http://nbviewer.jupyter.org/github/1202kbs/Understanding-NN/blob/master/5.2%20Explanation%20Selectivity.ipynb
