Inverse Decision Modeling: Learning Interpretable Representations of Behavior

Authors: Daniel Jarrett, Alihan Hüyük, Mihaela Van Der Schaar

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5. Illustrative Use Case So far, we have argued for a systematic, unifying perspective on inverse decision modeling ( IDM ) for behavior representation learning, and presented inverse bounded rational control ( IBRC ) as a concrete example of the formalism. Three aspects of this approach deserve empirical illustration: Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability Interpretability: IBRC gives a transparent parameterization of behavior that can be successfully learned from data. Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity Expressivity: IBRC more finely differentiates between imperfect behaviors, while standard reward learning cannot. Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability Applicability: IDM can be used in real-world settings, as an investigative device for understanding human decisions.
Researcher Affiliation Academia 1 Department of Applied Mathematics and Theoretical Physics, Uni- versity of Cambridge, UK; 2Department of Electrical Engineering, University of California, Los Angeles, USA.
Pseudocode No The paper provides mathematical formulations and theorems (e.g., Theorem 4 and 5), but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository.
Open Datasets Yes For our real-world setting, we consider 6-monthly clinical data for 1,737 patients in the Alzheimer s Disease Neuroimaging Initiative [163] study (ADNI).
Dataset Splits No The paper mentions using 1,000 generated trajectories as a basis for learning and describes the configuration for ADNI, but it does not specify explicit training, validation, or test splits by percentage or sample count, nor does it mention cross-validation.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions 'inference is performed via MCMC' and 'Bayesian IRL', but it does not specify any software names with version numbers for libraries, frameworks, or specific tools used.
Experiment Setup Yes In DIAG, our normative specification (for υ) is that diagnostic tests cost 1, correct diagnoses award 10, incorrect 36, and γ =0.95. Accuracies are 70% in both directions (!env), and patients arrive in equal proportions ( env). ... In ADNI, the configuration is similar except each MRI costs 1, while 2.5 is awarded once beliefs reach >90% certainty in any direction; also, σ is centered at the IOHMM learned from the data. For simplicity, for , % we use uniform priors in both settings.