Two steps to risk sensitivity
Authors: Christopher Gagne, Peter Dayan
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We first adopt a conventional distributional approach to CVa R in a sequential setting and reanalyze the choices of human decision-makers in the well-known two-step task, revealing substantial risk aversion that had been lurking under stickiness and perseveration. We use simulations to examine settings in which the various forms differ in ways that have implications for human and animal planning and behavior. |
| Researcher Affiliation | Academia | Chris Gagne MPI for Biological Cybernetics Tübingen, Germany christopher.gagne@tuebingen.mpg.de Peter Dayan MPI for Biological Cybernetics University of Tübingen Tübingen, Germany dayan@tue.mpg.de |
| Pseudocode | No | The paper includes mathematical equations and descriptions of processes but no explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide a statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | We therefore used a CVa R-based form of MB and MF reasoning to fit a very substantial dataset of human behavior in this task (out of more than 2000 participants in [3], the 792 who responded on every trial). |
| Dataset Splits | No | The paper describes using a dataset of human behavior for fitting models, but it does not specify how this dataset was split into training, validation, or test sets for model development or evaluation. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU, CPU models) used for running the experiments or parameter estimation. |
| Software Dependencies | No | The paper mentions "Parameters were estimated in Python using L-BFGS-B" but does not specify any version numbers for Python or other software libraries. |
| Experiment Setup | Yes | The 7 parameters of the CVa R-model were estimated for each participant: CVa R-based risk-sensitivity α [0.1, 1], learning rate λ [0.01, 0.99], dispersion η2 [0.001, 0.09], perseveration βsticky [0, 20] and three inverse temperature parameters β2nd, βMB, βMF [0, 30]. Parameters were estimated in Python using L-BFGS-B. Preliminary recovery simulations suggested that estimating both η2 and (1 φ2) was difficult, so we determined the value of (1 φ2) that would pin the asymptotic variance to 0.1. This was done so that a never-chosen option with an estimated mean 0.5 would have a CVa R of 0 at the lower-bound value for α (i.e. 0.1), which is appropriate since the outcomes themselves were between 0 and 1. The learning rate lay between 0.01 to 0.99 and was additionally constrained such that λ < (1 φ2). |