ARTFEED — Contemporary Art Intelligence

NP-Hard Problem of Optimal Experiment Selection for Causal Bounds

other · 2026-05-11

Researchers formalized the max-potency problem for selecting cost-constrained experiments that maximally tighten partial causal identification bounds. The problem is proven NP-hard via reduction from 0-1 knapsack. Building on Duarte et al. (2023), a general procedure evaluates epistemic potency in discrete settings. Two graphical pruning criteria, including a novel path-interception rule exploiting district structure, reduce the super-exponential search space. The work appears on arXiv (2605.06993v1).

Key facts

  • Causal queries are often partially identifiable from observational data.
  • Experiments can tighten bounds but are costly.
  • Max-potency problem selects experiments to maximally tighten bounds under cost constraints.
  • Epistemic potency measures worst-case reduction in bound width.
  • Problem is NP-hard via reduction from 0-1 knapsack.
  • General procedure for evaluating epistemic potency in discrete settings from Duarte et al. (2023).
  • Two graphical pruning criteria introduced: path-interception rule and district structure.
  • Path-interception rule certifies zero potency in linear time.
  • Paper on arXiv: 2605.06993v1.

Entities

Institutions

  • arXiv

Sources