Call graph image classification

[attilamester]

• this is an epic containing all work related to BagNet development:

  • code
  • experiments
  • deployment

• [x] #76

• [x] #77

• [x] #80

• [x] #94

• [x] #95

• [x] #96

• [x] #100

• [x] #101

• [x] #103

• [x] #106

• [x] #107

• [x] #113

• [x] #118

• [x] #97

• [x] #121

• [x] #125

• [x] #5

• [x] #139

• [ ] #146

• [x] #152

• [ ] #155

• [x] #156

[attilamester]

Train results from March 15-18

https://github.com/attilamester/malflow/assets/38431467/cdcb3606-72fc-44c7-8ad8-2a223351b3de

Conclusions

C1. Best results are obtained by resnet models over bagnet models

C2. Best results are obtained by larger images / smaller batch size

C3. Larger images matter more than smaller batch size

• smaller batch contributes to better metrics (this is expected)
• larger images with small batch contribute even more

[attilamester]