Environmental drivers of systematicity and generalization in a situated agent

Abstract

The question of whether deep neural networks are good at generalising beyond their immediate training experience is of critical importance for learning-based approaches to AI. Here, we consider tests of out-of-sample generalisation that require an agent to respond to never-seen-before instructions by manipulating and positioning objects in a 3D Unity simulated room. We first describe a comparatively generic agent architecture that exhibits strong performance on these tests. We then identify three aspects of the training regime and environment that make a significant difference to its performance: (a) the number of object/word experiences in the training set; (b) the visual invariances afforded by the agent's perspective, or frame of reference; and (c) the variety of visual input inherent in the perceptual aspect of the agent's perception. Our findings indicate that the degree of generalisation that networks exhibit can depend critically on particulars of the environment in which a given task is instantiated. They further suggest that the propensity for neural networks to generalise in systematic ways may increase if, like human children, those networks have access to many frames of richly varying, multi-modal observations as they learn.

References in corpus (12)

• IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
• Fine-Grained Analysis of Optimization and Generalization for Overparameterized Two-Layer Neural Networks
• Modular Multitask Reinforcement Learning with Policy Sketches
• Learning and Generalization in Overparameterized Neural Networks, Going Beyond Two Layers
• Gated-Attention Architectures for Task-Oriented Language Grounding
• Stronger generalization bounds for deep nets via a compression approach
• An analytic theory of generalization dynamics and transfer learning in deep linear networks
• Compositional generalization through meta sequence-to-sequence learning
• Interactive Grounded Language Acquisition and Generalization in a 2D World
• Human few-shot learning of compositional instructions
• Understanding Early Word Learning in Situated Artificial Agents
• Systematic Generalization: What Is Required and Can It Be Learned?

Cited by in corpus (10)

• Using cognitive psychology to understand GPT-3
• The Next Decade in AI: Four Steps Towards Robust Artificial Intelligence
• A Survey of Zero-shot Generalisation in Deep Reinforcement Learning
• Compositional Generalization in Semantic Parsing: Pre-training vs. Specialized Architectures
• Warmth and competence in human-agent cooperation
• Grounded Language Learning Fast and Slow
• Building Human-like Communicative Intelligence: A Grounded Perspective
• Exploring Exploration: Comparing Children with RL Agents in Unified Environments
• Distinguishing rule- and exemplar-based generalization in learning systems
• Learning Canonical Transformations