AmadeusGPT: a natural language interface for interactive animal behavioral analysis

Authors: Shaokai Ye, Jessy Lauer, Mu Zhou, Alexander Mathis, Mackenzie Mathis

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We used the MABe 2022 behavior challenge tasks to benchmark Amadeus GPT and show excellent performance.
Researcher Affiliation Academia Shaokai Ye EPFL Geneva, CH Jessy Lauer EPFL Geneva, CH Mu Zhou EPFL Geneva, CH Alexander Mathis EPFL Geneva, CH Mackenzie W. Mathis EPFL Geneva, CH mackenzie.mathis@epfl.ch
Pseudocode No The paper provides examples of Python code generated by Amadeus GPT, but it does not include pseudocode blocks or algorithms labeled as such.
Open Source Code Yes Code and demos can be found at: https://github.com/Adaptive Motor Control Lab/Amadeus GPT.
Open Datasets Yes We used the MABe 2022 behavior challenge tasks to benchmark Amadeus GPT... The third is video data of three mice interacting from the MABe 2022 Challenge [4]... The first is an open-access Elevated Plus Maze (EPM) dataset from Sturman et al. [13].
Dataset Splits No The paper mentions using the 'evaluation split' for the MABe 2022 Benchmark, implying predefined splits from the benchmark itself, but does not provide specific percentages or counts for training, validation, or test splits used in their own experiments.
Hardware Specification No The paper does not specify the hardware (e.g., GPU, CPU models) used for running the experiments. It mentions using 'GPT3.5 or 4' and 'Super Animals models [17]' but not the underlying hardware for execution.
Software Dependencies No The paper mentions using 'GPT3.5' and 'GPT4' (which are specific models, not general software dependencies with versions), and also refers to 'Super Animal models [17]' and 'SAM [11]'. It states Python code is generated and executed, but does not specify version numbers for Python or any other general software libraries or frameworks. The prompt requires specific version numbers for key software components.
Experiment Setup Yes Our approach is purely rule-based, thus no machine learning is needed and only three parameters are needed to be given or tuned: a smoothing window size for merging neighboring bouts, a minimal window size for dropping short events, and the pixel per centimeter.