pnx2003 / emergent_language

Emergent_Language

Cognitive and Reasoning (2023 Fall) course project by Core123.

Group Members

Weinan Qian, Feiyang Xie, Ningxin Pan, Haochen Zhao

Model pipeline

train0(Env info --agent0 encode--> message0 --agent1 decoder--> state rep --table lookup)train1(--agent1 encoder--> message1 --agent0 decoder--> agent0 action --> Env reward)
(split into train0 and train1 because lookup table truncates gradient)

Current setting:

1. state: multidiscrete([3,3,3]), message is out of env setting
2. single round communication
3. message length 1, vocab size > #state, expecting to learn (state,char) pairs