Differences

This shows you the differences between two versions of the page.

Link to this comparison view

learning_to_communicate [2017/08/29 01:19] (current)
Line 1: Line 1:
 +https://​openai.com/​blog/​learning-to-communicate/​
 +
 +https://​arxiv.org/​abs/​1602.02672v1 Learning to Communicate to Solve Riddles with Deep Distributed Recurrent Q-Networks
 +
 +https://​arxiv.org/​abs/​1605.06676v2 Learning to Communicate with Deep Multi-Agent Reinforcement Learning
 +
 +
 +
 +https://​arxiv.org/​abs/​1608.06409v1 Learning to Communicate:​ Channel Auto-encoders,​ Domain Specific Regularizers,​ and Attention
 +
 +https://​arxiv.org/​abs/​1611.01796v1 Modular Multitask Reinforcement Learning with Policy Sketches
 +
 +
 +https://​arxiv.org/​abs/​1606.02447v1 Learning Language Games through Interaction
 +
 +https://​arxiv.org/​abs/​1703.04908 ​ Emergence of Grounded Compositional Language in Multi-Agent Populations
 +
 +
 +https://​arxiv.org/​abs/​1610.03585 A Paradigm for Situated and Goal-Driven Language Learning
 +
 +https://​arxiv.org/​pdf/​1703.10069v1.pdf Multiagent Bidirectionally-Coordinated Nets
 +for Learning to Play StarCraft Combat Games
 +
 +https://​arxiv.org/​abs/​1704.06960v1 Translating Neuralese ​ https://​github.com/​jacobandreas/​neuralese
 +
 +https://​arxiv.org/​abs/​1705.11192v1 Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols
 +
 +https://​arxiv.org/​pdf/​1705.10369.pdf Emergent Language in a
 +Multi-Modal,​ Multi-Step Referential Game
 +
 +Inspired by previous work on emergent language in referential games, we propose
 +a novel multi-modal,​ multi-step referential game, where the sender and receiver
 +have access to distinct modalities of an object, and their information exchange is
 +bidirectional and of arbitrary duration. The multi-modal multi-step setting allows
 +agents to develop an internal language significantly closer to natural language, in
 +that they share a single set of messages, and that the length of the conversation may
 +vary according to the difficulty of the task. We examine these properties empirically
 +using a dataset consisting of images and textual descriptions of mammals, where
 +the agents are tasked with identifying the correct object. Our experiments indicate
 +that a robust and efficient communication protocol emerges, where gradual information
 +exchange informs better predictions and higher communication bandwidth
 +improves generalization. https://​github.com/​nyu-dl/​MultimodalGame
 +