Predicate Exchange: Inference with Declarative Knowledge

Authors: Zenna Tavares, Javier Burroni, Edgar Minasyan, Armando Solar-Lezama, Rajesh Ranganath

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We implement predicate exchange as a language agnostic tool which performs a nonstandard execution of a probabilistic program. We demonstrate the approach on sequence models of health and inverse rendering.
Researcher Affiliation Academia 1MIT, USA 2College of Information and Computer Science, University of Massachusetts, Amherst, USA. 3Princeton University, USA 4NYU, USA. Correspondence to: Zenna Tavares <zenna@mit.edu>.
Pseudocode Yes Algorithm 1 Soft Execution: softexecute(π, α, D) Algorithm 2 Predicate Exchange: predexchange
Open Source Code Yes For a concrete implementation, we build predicate exchange into the OMEGA probabilistic programming language1 (Tavares et al., 2019), 1OMEGA is available at http://github.com/zenna/Omega.jl
Open Datasets Yes We compare the independent RNN model to the one with declarative knowledge on a second patient from Physionet (Moody et al., 2001).
Dataset Splits No The paper mentions learning from a 'partial trajectory' or using 'first ten data points' for the glucose model, but does not provide specific train/validation/test split percentages or counts, nor does it specify a separate validation set.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments, such as GPU or CPU models, or memory specifications.
Software Dependencies No The paper mentions software components like 'No U-Turn Sampler', 'Hamiltonion Monte Carlo', 'reverse-mode automatic differentiation', and 'OMEGA probabilistic programming language', but does not specify their version numbers.
Experiment Setup Yes In practice, we simulate four chains with α logarithmically spaced between log10(α1) = 5 and log10(αM) = 5, and swap states that are adjacent in temperature (α1 with α2, α2 with α3, etc) every 10 iterations. (...) For finite dimensional continuous models we use the No U-Turn Sampler (Hoffman & Gelman, 2014), a variant of Hamiltonion Monte Carlo (HMC).