Dynamic Inverse Reinforcement Learning for Characterizing Animal Behavior

Authors: Zoe Ashwood, Aditi Jha, Jonathan W. Pillow

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate DIRL in simulated experiments and then apply it to a dataset of mice exploring a labyrinth.
Researcher Affiliation Academia 1Princeton Neuroscience Institute, Princeton University 2Dept. of Computer Science, Princeton University 3Dept. of Electrical and Computer Engineering, Princeton University {zashwood, aditijha, pillow}@princeton.edu
Pseudocode Yes Algorithm 1: DIRL Inference Procedure
Open Source Code Yes We provide the code as a supplemental material.
Open Datasets Yes The real data studied in this work is publicly available at https://github.com/markusmeister/Rosenberg-2021-Repository.
Dataset Splits Yes We then fit the goal maps and time-varying weights to 160 of these trajectories, and held out the remaining 20% of trajectories as a validation set.
Hardware Specification No Our method is computationally efficient and infers the time-varying reward function from 4000 decisions in the 127-node labyrinth environment in 20 minutes on a laptop. This is not specific enough (e.g., no CPU/GPU model or memory).
Software Dependencies No The paper does not explicitly list specific software dependencies with version numbers in the main text.
Experiment Setup Yes We, first, initialize the parameters randomly, such that the elements of the goal maps are chosen from U(0, 1) and the time-varying weights are Gaussian distributed.