- 3D ML compilations
- Pointnet
- Pointcloud annotation
- Papers
- A Frontier-Based Approach for Autonomous Exploration (Foundational Resource)
- Frontier Based Exploration for Autonomous Robot (Implementation)
-
Machine learning Fundementals
-
Deep Learning
- Deep Learning Roadmap
- Mit Deep Learning
- Deep Learning Basics
- 46:16 - Convolutional Neural Network: image classification
- 47:55 - R-CNN (Region) - object detection/ localization
- Explanation
- 49:06 - single-shot method - performance vs accuracy
- 50:04 - semantic segentation - pixel level boundaries instead of boxes
- 51:26 - transfer learning
- 52:26 - autoencoders - training to reproduce the output from the input
- 55:03 - Generative Adversarial Networks (GANs) - are a way to make a generative model by having two neural networks compete with each other.
- 57:04 - Word Embeddings - natural language processing
- 58:55 - Recurrent Neural Networks
- long-term dependency - depend on sequence sata and depend on sequenced data, challenge here is how to understand it in longer contexts.
- long short-term memory networks - pick what to forget and what to remember
- decide (1.) what to forget (state) (2.) what to remember (state) (3.) what to output (if any)
- 1:00:15 - bidirectional RNN
- 1:00:52 - encoder-decoder architecture
- 1:01:34 - Attention
- 1:02:13 - AutoML and Neural Architecture Search (NASNet)
- Deep Learning Basics
-
Robotics