0514 presentation
- 2. 覈: 蠍一 覦郁鍵 襾碁Μ 襾語 蠍磯 煙 覿襯
- 3. 螳 Name: 焔 Grade: 4 Name: ′綾 Grade: 4 Name: 蟾 Grade: 2
- 4. 覈 襾語 (Machine Learning) 蠍磯 譴, Deep Neural Network(DNN) 蟲譟磯ゼ 伎 煙襯 覿襯 煙襯 class襦 覿襯, 觜訣蟆 malwareる 螳 class襦 覿襯蟆 旧れ
- 5. 煙(malware): 貉危 覦企, 貉危 蟆 襯 殊 覈 貊 豐豺 覿(static analysis) 煙 hexa dump file, assembly file 轟(feature) 豢豢
- 6. Kaggle 螻牛 煙 AI 蟯 global contest 郁規襯 讌 貉るる https://www.kaggle.com/c/malware-classification
- 7. 瑚概讌(AI, Artificial Intelligence) レ 譯殊伎覃 豢レ 企慨碁.
- 11. 螻給 朱襖 一危
- 12. Deep Neural Network Hidden layer螳 2螳 伎 NN
- 13. Used feature 煙 レ
- 14. Used feature 煙 伎觚襴
- 15. Used feature 伎觚襴 -> 螳 instruction 豢豢 unigram 蠍磯 feature レ -> 2byte 伎 counting vector レ -> 覦危 れ FF襯 search -> FF 覦 碁ゼ 覓語企 伎 -> counting vector
- 16. ろ Train file #: 10868, Test file #: 10873 Input layer node # : 10000 Hidden layer #: 7 Layer 1: 10000 / 2 Layer 2: 10000 / 4 Layer 3: 10000 / 8 Layer 4: 10000 / 4 Layer 5: 10000 / 2 Layer 6: 10000 Layer 7: 10000 * 2 Learning rate: 0.00001, dropout prob = 0.5 Epoch #: 7000 Mini-batch size: 128 螳
- 17. 蟆郁骸 Kaggle site 豈 企骸 = 98.83
- 18. 螳矧
Editor's Notes
- #5: 煙襯 覿襯 伎 る