• 목록

딥러닝

<Attention>

1주: Attention

• Why Attention?
• Seq2Seq (Sequence to Sequence)
• Seq2Seq with Attention
• Query, Key, Value

2주: Transformer

• Self-Attention Mechanism
• Structure of Transformer (3 Different Self-Attention Mechanism)
• Similarities and Differences between Seq2Seq & Transformer
• Multi-Head Attention

3주: Transformer in Vision 1

• ViT (Vision Transformer)
• DeiT (Data Efficient Image Transformer)
• MLP-Mixer

4주: Transformer in Vision 2

• CNN (Convolutional Neural Network) vs ViT vs MLP-Mixer
• DETR (Detection Transformer)

<Graph Neural Networks>

1주: Graph Generation

• Radius-graph
• kNN-graph
• Farthest Point Sampling

2주: Graph Convolution Networks (GCN)

• Neural Network for Graph (NN4G)          
• node2vec
  
• Graph Sage
  
• Kipf and Welling’s GCN
  
• Message Passing Neural Network (MPNN) 

3주: Application of GNN

• Feature Learning on Point Cloud (PointNet++, DGCNN)
• 3D Object Detection (Point GNN)

강화학습

<Meta-Learning and Self-Supervision>

1주: Model Predictive Control and Meta-Learning:   

• Concept of MPC, Sampling-based MPC (Random Shooting, MPPI)
• Meta-Learning (Recurrence-based, MAML, FOMAML, Reptile)

2주: Meta-Learning for System Identification in Sampling-based MPC:   

• Meta-Learning + MPC for Fast Adaptation (ReBAL, GrBAL, FAMLE)                 
• Additional Adaptation Methods (PETS, CARL)

3주: Self-Supervision in Reinforcement Learning

• Self-supervised Learning
  
• Hindsight Goal Relabeling
• Representation Learning 

4주: Unsupervised Skill Discovery

• What is Skill?
  
• Unsupervised Skill Discovery methods
  
• How to utilize skills?