Fast and Flexible Inference of Joint Distributions from their Marginals
Authors: Charlie Frogner, Tomaso Poggio
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We investigate empirically both the computational cost of our algorithm and the accuracy of the new models on real datasets, showing favorable performance in both cases and illustrating the impact of increased flexibility in modeling enabled by this work. |
| Researcher Affiliation | Academia | 1CSAIL, Massachusetts Institute of Technology, Cambridge, Massachusetts 2Center for Brains, Minds, and Machines, Massachusetts Institute of Technology, Cambridge, Massachusetts. |
| Pseudocode | Yes | Algorithm 1 Dykstra s method for MAP estimation; Algorithm 2 Dykstra s method for MAP estimation, dual parameterization; Algorithm 3 Newton-Raphson method for the projection PCu onto a marginal constraint |
| Open Source Code | No | The paper provides a link for a baseline implementation ('The TROT implementation is from the original author, available at https://github.com/Boris Muzellec/TROT.') but does not explicitly state or provide a link for the authors' own method's source code. |
| Open Datasets | Yes | 1. CDC cause of death by state, 1999-2016 (CDC, 2018); 2. Indian educational attainment by district, 2001 (India, 2018); 3. Florida voter registration by county, 2012 (Imai & Khanna, 2016); 4. U.S. total personal income by state, 2016 (IPUMS, 2018); 5. U.S. health insurance coverage by state, 2016 (IPUMS, 2018) |
| Dataset Splits | No | The paper describes using five real datasets and treating the collected tables as 'ground truth' for evaluation, but it does not specify any train/validation/test splits or mention cross-validation. The evaluation is implied to be on the full datasets. |
| Hardware Specification | Yes | All methods were implemented in Python using numpy, and were run on a Mac Book Pro with a dual-core 2.9GHz processor. |
| Software Dependencies | No | The paper mentions 'implemented in Python using numpy' but does not specify version numbers for Python or numpy. It also mentions 'The TROT implementation is from the original author', but this refers to a baseline, not their specific software dependencies with versioning. |
| Experiment Setup | Yes | After each outer iteration we check the Frobenius norm of the deviation of the dual variable Θ from its previous value, halting when this deviation is less than 1e-4. The inner Newton optimization is run for 20 steps within each outer iteration. ... We set the parameter λ from Muzellec et al. (2017) to 1. For normal models, we fix σ = 0.1. |