A Perspective on Objects and Systematic Generalization in Model-Based RL
arXiv:1906.01035
Abstract
In order to meet the diverse challenges in solving many real-world problems, an intelligent agent has to be able to dynamically construct a model of its environment. Objects facilitate the modular reuse of prior knowledge and the combinatorial construction of such models. In this work, we argue that dynamically bound features (objects) do not simply emerge in connectionist models of the world. We identify several requirements that need to be fulfilled in overcoming this limitation and highlight corresponding inductive biases.
Accepted to the ICML 2019 workshop on Workshop on Generative Modeling and Model-Based Reasoning for Robotics and AI
References in corpus (9)
- An Introduction to Deep Reinforcement Learning
- Recurrent Models of Visual Attention
- Interaction Networks for Learning about Objects, Relations and Physics
- A simple neural network module for relational reasoning
- Recurrent World Models Facilitate Policy Evolution
- Poincaré Embeddings for Learning Hierarchical Representations
- AI Safety Gridworlds
- Sequential Attend, Infer, Repeat: Generative Modelling of Moving Objects
- Learning to Reason with Third-Order Tensor Products
Cited by in corpus (10)
- Remix: Rebalanced Mixup
- Enhancing the Transformer with Explicit Relational Encoding for Math Problem Solving
- R-SQAIR: Relational Sequential Attend, Infer, Repeat
- Generative Neurosymbolic Machines
- Levels of Analysis for Machine Learning
- Unsupervised Object Keypoint Learning using Local Spatial Predictability
- Self-supervised Visual Reinforcement Learning with Object-centric Representations
- Pretrained Encoders are All You Need
- Modular Object-Oriented Games: A Task Framework for Reinforcement Learning, Psychology, and Neuroscience
- The Impact of Negative Sampling on Contrastive Structured World Models