轉載自:阿里雲node
https://yq.aliyun.com/articles/149583?utm_content=m_27089git
代碼實現:C:\Users\Josie\AppData\Local\Programs\Python\Python35\Scripts\1\1.Backpropagation.ipynbgithub
參考論文:1.Synthesizing Robust Adversarial Examples.pdf數據庫
1.2 Synthesizing Robust Adversarial Examples 這篇和第一篇應該是同樣的,有些許的不一樣網絡
是第一篇論文的拓展復現:post
對應github上下載的代碼:https://github.com/prabhant/synthesizing-robust-adversarial-examples測試
下載到本地存放在:C:\Users\Josie\AppData\Local\Programs\Python\Python35\Scripts\1\synthesizing-robust-adversarial-examples-masterui
2.反向傳播:Calculus on Computational Graphs Backpropagation 阿里雲
http://colah.github.io/posts/2015-08-Backprop/?spm=a2c4e.11153940.blogcont149583.20.4ab360c05me4Uv.net
3.對抗樣本:Intriguing properties of neural networks
4.轉移到物質世界:Adversarial examples in the physical world
19.介紹Inception-v3的論文:Rethinking the Inception Architecture for Computer Vision
註釋:
一、ImageNet is an image database organized according to the WordNet hierarchy (currently only the nouns), in which each node of the hierarchy is depicted by hundreds and thousands of images. Currently we have an average of over five hundred images per node. We hope ImageNet will become a useful resource for researchers, educators, students and all of you who share our passion for pictures.
ImageNet是根據WordNet層次結構(當前只是名詞)組織的圖像數據庫,其中層次結構的每一個節點由數百和數千個圖像描繪。 目前,咱們每一個節點平均有超過500個圖像。 咱們但願ImageNet將成爲研究人員,教育工做者,學生以及全部分享咱們對圖片的熱情的全部人的有用資源。
二、TensorFlow-slim圖像分類庫:
https://blog.csdn.net/chaipp0607/article/details/74139895
https://blog.csdn.net/yifen4234/article/details/80252189
https://blog.csdn.net/u010197508/article/details/72909446
TF-slim是用於定義,訓練和評估複雜模型的TensorFlow(tensorflow.contrib.slim)的新型輕量級高級API。 該目錄包含了幾種普遍使用的卷積神經網絡(CNN)圖像分類模型的訓練和測試代碼。它包含腳本,容許您從頭開始訓練模型或從預訓練(pre-train)的模型進行fine-tune。 它還包含用於下載標準圖像數據集的代碼,將其轉換爲TensorFlow的TFRecord格式,並能夠使用TF-Slim的數據讀取和隊列程序進行讀取。您能夠輕鬆地使用這些數據集進行任意模型的訓練,以下所示。