Online Causal Structure Learning in the Presence of Latent Variables

In this thesis, we propose to use Causal Bayesian Networks (CBNs), which play a central role in dealing with uncertainty in Artificial Intelligence (AI). Causal models can be created based on information, data, or both. Regardless of the source of information used to create the model, there may be inaccuracies, or the application area may vary. Therefore, the model needs constant improvement during use. Most of existing structure learning algorithms are batch. However, industrial companies store vast amounts of data every day in real-world scenarios. Existing batch methods cannot process the significant quantity of continuously incoming data in a reasonable amount of time and memory. Therefore, batch methods may become computationally expensive and infeasible for large dataset. It is inappropriate to handle such changes with existing batch-learning approaches, and instead, a structure should be learned in an online manner. In this way, we present three online causal structure learning algorithms to fill this gap. These algorithms can track changes in a causal structure and process data in a dynamic real-time manner. Standard causal structure learning algorithms as- sume that causal structure does not change during the data collection process, but in real-world scenarios, it does often change. The online causal structure learning algorithms we present here can revise correlation values without reprocessing the entire dataset and use an existing model to avoid re-learning the causal links in the prior model, which still fit data. The algorithms update the correlations of causes and effects with the estimation of the weight of each causal interaction. Proposed algorithms are tested on synthetic and real-world datasets. Firstly, we performed the desired algorithms and Fast Causal Inference (FCI) algorithm on synthetic datasets generated from structures that change over time. We com- pared these algorithms in the respect of both the learning performance and learn- ing speed. And then, we illustrated the benefits of this approach by applying to real-world data which is a seasonally adjusted commodity price index dataset (monthly) for the U.S. from 1967 to 2018. The online causal structure learning algorithms outperformed standard FCI by a large margin in learning the changed causal structure correctly and efficiently when latent variables were present.