Modular meta-learning in abstract graph networks for combinatorial generalization
arXiv:1812.07768
Abstract
Modular meta-learning is a new framework that generalizes to unseen datasets by combining a small set of neural modules in different ways. In this work we propose abstract graph networks: using graphs as abstractions of a system's subparts without a fixed assignment of nodes to system subparts, for which we would need supervision. We combine this idea with modular meta-learning to get a flexible framework with combinatorial generalization to new tasks built in. We then use it to model the pushing of arbitrarily shaped objects from little or no training data.
Presented at NeurIPS meta-learning workshop 2018
References in corpus (17)
- Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
- Theoretical Models of Learning to Learn
- Matching Networks for One Shot Learning
- PathNet: Evolution Channels Gradient Descent in Super Neural Networks
- Deep Learning: A Critical Appraisal
- Graph networks as learnable physics engines for inference and control
- Unsupervised Learning for Physical Interaction through Video Prediction
- Neural Relational Inference for Interacting Systems
- Flexible Neural Representation for Physics Prediction
- Relational Neural Expectation Maximization: Unsupervised Discovery of Objects and their Interactions
- Visual Interaction Networks
- Routing Networks: Adaptive Selection of Non-linear Functions for Multi-Task Learning
- Modular meta-learning
- Beyond Shared Hierarchies: Deep Multitask Learning through Soft Layer Ordering
- Neural Task Programming: Learning to Generalize Across Hierarchical Tasks
- Augmenting Physical Simulators with Stochastic Neural Networks: Case Study of Planar Pushing and Bouncing
- Evolutionary Architecture Search For Deep Multitask Networks