Anti-efficient encoding in emergent communication
Authors: Rahma Chaabouni, Eugene Kharitonov, Emmanuel Dupoux, Marco Baroni
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We study whether the same pattern emerges when two neural networks, a speaker and a listener , are trained to play a signaling game. Surprisingly, we find that networks develop an anti-efficient encoding scheme, in which the most frequent inputs are associated to the longest messages, and messages in general are skewed towards the maximum length threshold. |
| Researcher Affiliation | Collaboration | Rahma Chaabouni1,2, Eugene Kharitonov1, Emmanuel Dupoux1,2 and Marco Baroni1,3 1Facebook AI Research 2Cognitive Machine Learning (ENS EHESS PSL Research University CNRS INRIA) 3ICREA {rchaabouni,kharitonov,dpx,mbaroni}@fb.com |
| Pseudocode | No | No structured pseudocode or algorithm blocks were found. Equation (1) is a mathematical formula, not pseudocode. |
| Open Source Code | Yes | The game is implemented using the EGG toolkit [Kharitonov et al., 2019], and the code can be found at https://github.com/facebookresearch/EGG/tree/master/egg/zoo/channel. |
| Open Datasets | No | The paper describes how 1K distinct one-hot vectors are generated from a power-law distribution for the experiment inputs, but does not provide concrete access (link, citation, or repository) to this specific training data. While natural language frequency lists are used for reference distributions, they are not the primary experimental training data. |
| Dataset Splits | No | The paper does not explicitly provide specific training/validation/test dataset splits, nor does it mention a distinct validation set being used. |
| Hardware Specification | No | No specific hardware details (e.g., GPU models, CPU types, memory) used for running the experiments are provided in the paper. |
| Software Dependencies | No | The paper mentions 'Adam optimizer' and 'LSTMs' (architectures), and the 'EGG toolkit' (a framework), but does not specify version numbers for any software dependencies like Python, PyTorch, or other libraries. |
| Experiment Setup | Yes | We train agents for 2500 episodes, each consisting of 100 mini-batches, in turn including 5120 inputs sampled from the power-law distribution with replacement. ... Speaker s hidden size H s and Listener s hidden size H l are selected from {100, 250, 500}. We train models for a variable number of episodes (2500, 5000, 10000). Learning rate for the Adam optimizer is selected from {1e-3, 5e-4, 1e-4}. Entropy regularization coefficient is selected from {0.1, 0.05, 0.01, 0.005}. |