R-SQAIR: Relational Sequential Attend, Infer, Repeat
arXiv:1910.05231
Abstract
Traditional sequential multi-object attention models rely on a recurrent mechanism to infer object relations. We propose a relational extension (R-SQAIR) of one such attention model (SQAIR) by endowing it with a module with strong relational inductive bias that computes in parallel pairwise interactions between inferred objects. Two recently proposed relational modules are studied on tasks of unsupervised learning from videos. We demonstrate gains over sequential relational mechanisms, also in terms of combinatorial generalization.
4 page workshop paper accepted at the NeurIPS 2019 Workshop on Perception as Generative Reasoning: Structure, Causality, Probability
References in corpus (7)
- Semi-Supervised Learning with Ladder Networks
- Recurrent World Models Facilitate Policy Evolution
- Attend, Infer, Repeat: Fast Scene Understanding with Generative Models
- Multi-Object Representation Learning with Iterative Variational Inference
- MONet: Unsupervised Scene Decomposition and Representation
- COBRA: Data-Efficient Model-Based RL through Unsupervised Object Discovery and Curiosity-Driven Exploration
- A Perspective on Objects and Systematic Generalization in Model-Based RL
Cited by in corpus (8)
- Object-Centric Learning with Slot Attention
- Learning Physical Graph Representations from Visual Scenes
- Unsupervised Learning of Compositional Energy Concepts
- Unsupervised Discovery of 3D Physical Objects from Video
- Towards Unsupervised Learning of Generative Models for 3D Controllable Image Synthesis
- Unsupervised object-centric video generation and decomposition in 3D
- Learning Disentangled Representations of Video with Missing Data
- Conditional Object-Centric Learning from Video