ION-C: Integration of Overlapping Networks via Constraints
Praveen Nair, Payal Bhandari, Mohammadsajad Abavisani, Sergey Plis,, David Danks
TL;DR
This paper introduces ION-C, a more efficient ASP-based algorithm for integrating overlapping causal networks across multiple datasets, validated on synthetic and real-world social survey data.
Contribution
The paper formulates the overlapping network integration problem as an ASP problem and improves computational efficiency over previous methods.
Findings
Overlap significantly affects runtime and solution set size.
ION-C successfully applied to synthetic graphs with varying parameters.
Validated on European Social Survey data with consistent results.
Abstract
In many causal learning problems, variables of interest are often not all measured over the same observations, but are instead distributed across multiple datasets with overlapping variables. Tillman et al. (2008) presented the first algorithm for enumerating the minimal equivalence class of ground-truth DAGs consistent with all input graphs by exploiting local independence relations, called ION. In this paper, this problem is formulated as a more computationally efficient answer set programming (ASP) problem, which we call ION-C, and solved with the ASP system clingo. The ION-C algorithm was run on random synthetic graphs with varying sizes, densities, and degrees of overlap between subgraphs, with overlap having the largest impact on runtime, number of solution graphs, and agreement within the output set. To validate ION-C on real-world data, we ran the algorithm on overlapping graphs…
Peer Reviews
Decision·Submitted to ICLR 2025
The paper uses good English, and shows experimental results suggesting the described goal is achieved of addressing the problem as an ASP problem. At a high level the proposed approach seems plausible.
#### General In general, the paper is probably only readable to specialists. The paper doesn't provide introductory definitions which would allow a non-expert reader to understand concepts and notations or would allow a more expert reader to disambiguate between definitions of which multiple different ones have been considered in the literature. See "details below" for a few examples. The paper doesn't demonstrate clearly what is the added value of representing and solving the problem as a
- The problem itself is well-motivated and Sec 1 introduction to Sec 3 problem setting and method are easy to follow. - This work revisits an old algorithm and reformulate in a simple clingo problem specification. (I appreciate a simple solution over a unnecessarily complex solution.) - The combination of ASP and clingo scales better than the original ION algorithm.
- No new notable theoretical contribution and the main contribution seems rewriting the conditions/constraints in ASP/clingo.
The tackled problem is of practical importance as exploiting data from different sources, even when defined on different variables, is fundamental for real-world applications of causal discovery. Due to the high-computational complexity, the existing ION algorithm could not scale to medium-sized graphs — as the authors say ION was only tested on 6 nodes DAGs in the original evaluation. Therefore, the presented solution might have important applications for larger graphs.
The main point of the proposed ION-C algorithm is to fasten the ION algorithm. However, the computational complexity of solving the ASP program is not reported in the paper. If the authors could provide such complexity and compare it directly to the complexity of the original ION algorithm, it would help to understand whether ION-C has theoretical guarantees of being faster or if it's only an empirical result. Finally, for the experimental side, it would help to have a clear visualization of ION
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Optical Network Technologies · Semiconductor Lasers and Optical Devices · Photonic and Optical Devices
MethodsSparse Evolutionary Training