Trading off Utility, Informativeness, and Complexity in Emergent Communication
Authors: Mycal Tucker, Roger Levy, Julie A. Shah, Noga Zaslavsky
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our approach in multi-agent reinforcement learning settings and in color reference games and show that: (1) VQVIB agents can continuously adapt to changing communicative needs and, in the color domain, align with human languages; (2) the emergent VQ-VIB embedding spaces are semantically meaningful and perceptually grounded; and (3) encouraging informativeness leads to faster convergence rates and improved utility, both in VQVIB and in prior neural architectures for symbolic EC, with VQ-VIB achieving higher utility for any given complexity. |
| Researcher Affiliation | Academia | Mycal Tucker M.I.T. mycal@mit.edu Roger Levy M.I.T. rplevy@mit.edu Julie Shah M.I.T. julie_a_shah@csail.mit.edu Noga Zaslavsky M.I.T. nogazs@mit.edu |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/mycal-tucker/vqvib_neurips2022 |
| Open Datasets | Yes | The target and distractor are sampled independently from the 330 color chips of the World Color Survey [WCS, 42] shown in Figure 3c, according to a humanbased prior from [13]. The WCS data are available at www1.icsi.berkeley.edu/wcs and the human-based color prior is available at https://github.com/nogazs/ib-color-naming. |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset splits by percentages or sample counts in the main text. It mentions 'validation loss' in Appendix D.1 but no details on the split itself. |
| Hardware Specification | Yes | All experiments were run on a single NVIDIA Quadro RTX 6000 GPU (24GB memory) or a single NVIDIA A100 GPU (40GB memory). |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers. |
| Experiment Setup | Yes | Full details of the architectures, hyperparameters, and environments are provided in Appendix D. |