/project_SDE-Net_analysis

Analysis of paper research SDE-Net (2020)

Primary LanguagePython

Final Project: Course Industrial Programming in Python language

Student

  • Фамиля: Эспинола Ривера
  • имя: Хольгер Элиас
  • программа: Исскуственный Интеллект и Машинное обучение
  • Группа: 3540201/ 20301

Content

The project have this files:

  1. SDE-Net_analysis
    Analysis of paper research titled: SDE-Net, Equipping Deep Neural Networks with uncertainty estimates (2020)
    using standard structure of scientific method.

    • Report (курсовая работа) of all parts of the research (format word and pdf)
      Files: ++FinalReport_SDENet.docx and ++FinalReport_SDENet.pdf
    • Original paper (format pdf)
      Files: ++SDE-Net_Equipping DNN with uncertainty estimates.pdf
    • Presentation (format ppt)
      File: ++SDE-Net_presentation.pptx
    • Experimental results (format word)
      File: Experimental Results.docx
    • Flux diagram of training process for SDE-Net (format jpg)
      File: fluxdig_sde.jpg

  2. SDE-Net-program
    Original programming solution purposed by the researches in Pytorch. Was developed experiments for 3
    neural network architectures: Resnet, Resnet + MC-dropout and SDE-Net for classification using MNIST
    and SVHN datasets and for regression using YearMSD dataset. All implementations use Pytorch.

    • Implementation of loading data
      Files: MNIST/data_loader.py, SVHN/data_loader.py, YearMSD/data_loader.py
    • Implementation of model architecture
      Files: MNIST/models/* , SVHN/models/* , YearMSD/models/*
    • Implementation training and testing for models
      Files for MNIST: MNIST/resnet_mnist.py, resnet_dropout_mnist.py, sdenet_mnist.py
      Files for SVHN: SVHN/resnet_svhn.py, resnet_dropout_svhn.py, sdenet_svhn.py
      Files for YearMSD: YearMSD/DNN_mc.py, SDE_regression.py