Classification Based on Missing Features in Deep Convolutional Neural Networks

Explore a novel research approach for enhancing neural network accuracy and robustness in challenging or adversarial scenarios through a talk from the EuroPython 2019 conference. Delve into the workings of convolutional layers and learn how to modify them to classify based on missing features in images. Gain insights into improving model performance on a variation of the MNIST dataset using PyTorch 1.1. Discover the potential implications for critical applications like self-driving cars. Follow along as the speaker covers transfer learning, activation functions, Python code implementation, testing, and results. Conclude with a discussion on future work and different architectures in this cutting-edge exploration of deep convolutional neural networks.

Introduction
Presentation Overview
Missing Features Classification
Step 1 Transfer Learning
Step 2 Activation Functions
Real Activation Functions
Python Code
Code
Testing
Remarks
Results
Future work
Different architectures
Conclusion