Summer research project led by Paul Ruvolo.

Our goal is to combine a convolutional neural network with an efficient lifelong learning algorithm and use the result on a mobile robot. We aim to avoid the problem of catastrophic interference/forgetting that tends to affect incremental task learners through some model manipulation; this repository hosts the results of all of our experiments. You can view our progress presentation here.

The team:

• Joey Maalouf
  • Convolutional Neural Network Configuration
  • ROS Video Integration
  • Catastrophic Interference Demonstration
  • Top-Layer Model Integration/Comparison (ELLA/SVC/etc.)
  • Multi-Net Model Design/Creation/Experimentation
  • General Functions Module Creation
  • Caffe Experimentation/Prediction Extension
  • Real-World Dataset Testing
• Sean Carter
  • Modern Neural Network Configuration
  • Neural Network Generalization/Experimentation
  • Multiple Dataset Analysis
  • Caffe Experimentation
• Zhecan Wang
  • ROS Video Integration
  • Real-World Dataset Gathering/Testing
  • Neural Network Experimentation
  • Collective-Net Model Creation/Experimentation