Personal portfolio

Summary of some of my personal work

1. Kaggle Petfinder

• Deep Learning
• Computer Vision
• Regression
• Goal: Given a pet image an addional metadata -> predict the image social score
• Framework: Pytorch on GPU
• Experiments:
  • Xgboost (EfficientNet b0 feature extraction + metadata)
  • ConvNet (EfficientNet b0)
  • MLP based on Backbone (EfficientNet b0 + MLP Metadata)
  • MLP based on Backbone (EfficientNet b1 + MLP Metadata)
• Best model: MLP based on Backbone (EfficientNet b2 + MLP Metadata)
  • Kaggle test set RMSE: ~ 19. (best rank: p50 of the leaderboard)