Bayesian Symbolic Regression
A method that incorporates symbolic regression into a Bayesian inference framework by defining a prior over the space of expressions and performing posterior inference in combination with a likelihood, sampling closed-form mathematical expressions from the posterior distribution and thereby naturally quantifying model uncertainty.
Timeline (last 90 days)
A recent paper proposed reformulating causal discovery as a Bayesian symbolic regression problem, quantifying causal strength by sampling symbolic expressions from the posterior distribution.
The method defines the frequency with which each candidate variable appears in the sampled expressions as its posterior inclusion probability, which serves as the causal score.
Bayesian symbolic regression defines a prior distribution over expressions and performs posterior inference combined with the likelihood, characterizing which expressions are plausible with a probability distribution and thereby quantifying model uncertainty.
The method was evaluated on coupled nonlinear bivariate dynamical data, the CauseMe benchmark suite, and a four-variable biogeoscience structural causal model.
Experimental results show that the method is competitive with purpose-designed causal discovery algorithms and often outperforms them.
The method can return closed-form mechanisms, thereby achieving explicit causality and interpretability.
The method overcomes the limitations of traditional methods in bivariate and strongly autocorrelated settings, providing explicit causal directions rather than equivalence classes.
The method improves robustness through ensemble posterior sampling, avoiding the bias introduced by selecting a single model.
The main limitation of the method is that computational cost may be high, since a large number of symbolic expressions must be sampled.
All Facts (9)
The main limitation of the method is that computational cost may be high, since a large number of symbolic expressions must be sampled.
50%UnverifiedThe method defines the frequency with which each candidate variable appears in the sampled expressions as its posterior inclusion probability, which serves as the causal score.
50%UnverifiedBayesian symbolic regression defines a prior distribution over expressions and performs posterior inference combined with the likelihood, characterizing which expressions are plausible with a probability distribution and thereby quantifying model uncertainty.
50%UnverifiedA recent paper proposed reformulating causal discovery as a Bayesian symbolic regression problem, quantifying causal strength by sampling symbolic expressions from the posterior distribution.
50%UnverifiedExperimental results show that the method is competitive with purpose-designed causal discovery algorithms and often outperforms them.
50%UnverifiedThe method can return closed-form mechanisms, thereby achieving explicit causality and interpretability.
50%UnverifiedThe method overcomes the limitations of traditional methods in bivariate and strongly autocorrelated settings, providing explicit causal directions rather than equivalence classes.
50%UnverifiedThe method improves robustness through ensemble posterior sampling, avoiding the bias introduced by selecting a single model.
50%UnverifiedThe method was evaluated on coupled nonlinear bivariate dynamical data, the CauseMe benchmark suite, and a four-variable biogeoscience structural causal model.
50%