Probing Emergent Semantics in Predictive Agents via Question Answering
arXiv:2006.01016
Abstract
Recent work has shown how predictive modeling can endow agents with rich knowledge of their surroundings, improving their ability to act in complex environments. We propose question-answering as a general paradigm to decode and understand the representations that such agents develop, applying our method to two recent approaches to predictive modeling -action-conditional CPC (Guo et al., 2018) and SimCore (Gregor et al., 2019). After training agents with these predictive objectives in a visually-rich, 3D environment with an assortment of objects, colors, shapes, and spatial configurations, we probe their internal state representations with synthetic (English) questions, without backpropagating gradients from the question-answering decoder into the agent. The performance of different agents when probed this way reveals that they learn to encode factual, and seemingly compositional, information about objects, properties and spatial relations from their physical environment. Our approach is intuitive, i.e. humans can easily interpret responses of the model as opposed to inspecting continuous vectors, and model-agnostic, i.e. applicable to any modeling approach. By revealing the implicit knowledge of objects, quantities, properties and relations acquired by agents as they learn, question-conditional agent probing can stimulate the design and development of stronger predictive learning objectives.
ICML 2020
References in corpus (20)
- VQA: Visual Question Answering
- IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
- Unsupervised Visual Representation Learning by Context Prediction
- Modular Multitask Reinforcement Learning with Policy Sketches
- Know What You Don't Know: Unanswerable Questions for SQuAD
- Unsupervised Learning of Visual Representations using Videos
- Target-driven Visual Navigation in Indoor Scenes using Deep Reinforcement Learning
- The Predictron: End-To-End Learning and Planning
- Grounded Language Learning in a Simulated 3D World
- Unsupervised Predictive Memory in a Goal-Directed Agent
- Language Understanding for Text-based Games Using Deep Reinforcement Learning
- Towards Conceptual Compression
- Human Instruction-Following with Deep Reinforcement Learning via Transfer-Learning from Text
- Interactive Grounded Language Acquisition and Generalization in a 2D World
- Visual Semantic Planning using Deep Successor Representations
- VirtualHome: Simulating Household Activities via Programs
- Embodied Question Answering in Photorealistic Environments with Point Cloud Perception
- World Discovery Models
- VideoNavQA: Bridging the Gap between Visual and Embodied Question Answering
- Multi-Target Embodied Question Answering