Programmable Agents

2017/06/20 by Misha Denil, Sergio Gómez Colmenarejo, Denil, Misha +7 · 2 voices
Computer Science · Mathematics · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (stat.ML) #Multimodal Machine Learning Applications #Neural and Evolutionary Computing (cs.NE) #Reinforcement Learning in Robotics #Topic Modeling #cs.AI #cs.NE #stat.ML

paper · pdf · doi:10.48550/arxiv.1706.06383

arxiv created 2017/06/20 · openalex publication_date 2017/06/20 · arxiv published 2017/06/20 · arxiv updated 2017/06/21 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

We build deep RL agents that execute declarative programs expressed in formal language. The agents learn to ground the terms in this language in their environment, and can generalize their behavior at test time to execute new programs that refer to objects that were not referenced during training. The agents develop disentangled interpretable representations that allow them to generalize to a wide variety of zero-shot semantic tasks.

Citations

• Learning to Transduce with Unbounded Memory
• Distributed Representations of Words and Phrases and their Compositionality
• DeepCoder: Learning to Write Programs
• Learning a Natural Language Interface with Neural Programmer
• A Compositional Object-Based Approach to Learning Physical Dynamics

Discussions

• Programmable Agents [hn, 2 points, 0 comments]
• Programmable Agents [pdf] [hn, 1 points, 0 comments]

Related

• Deep Sets for Generalization in RL
• Learning Abstractions for Hierarchical Planning in Program-Synthesis Agents
• Generalizable End-to-End Tool-Use RL with Synthetic CodeGym
• Game On: Towards Language Models as RL Experimenters
• Ground-Compose-Reinforce: Grounding Language in Agentic Behaviours using Limited Data