Hans Jarett Ong

Causal discovery from observational data is a fundamental challenge. Greedy search algorithms like Regression with Subsequent Independence Test (RESIT), commonly used for learning Additive Noise Models (ANMs), are susceptible to making irreversible errors, especially in high-variance contexts. Such settings can be caused by unmeasured confounders or by high statistical noise from finite samples. To address this, we introduce a novel generalization of RESIT that replaces its local, greedy search with a more robust beam search, framing the task as a path search on a state-space graph. Through extensive simulation experiments, we demonstrate that structural accuracy, measured by Structural Hamming Distance (SHD) and Structural Intervention Distance (SID), consistently improves as the beam width (w) increases. Crucially, we also show that this performance gain comes at a manageable, approximately linear increase in computational cost relative to w. Furthermore, our analysis across different sample sizes shows these gains are most statistically significant in intermediate regimes (n=250,500). This suggests that at these sample sizes, the statistical noise is high enough to mislead the greedy search into a suboptimal ordering, an error our wider beam search corrects, while performance converges at large sample sizes (n=1000). Our framework provides a practical, tunable algorithm that bridges the gap between fast but brittle local search methods and computationally infeasible global searches, thereby enhancing the reliability of causal discovery in complex, high-variance settings where such local errors are common.

Unsupervised representation learning seeks to recover latent generative factors, yet standard methods relying on statistical independence often fail to capture causal dependencies. A central challenge is identifiability: as established in disentangled representation learning and nonlinear ICA literature, disentangling causal variables from observational data is impossible without supervision, auxiliary signals, or strong inductive biases. In this work, we propose the Latent Additive Noise Model Causal Autoencoder (LANCA) to operationalize the Additive Noise Model (ANM) as a strong inductive bias for unsupervised discovery. Theoretically, we prove that while the ANM constraint does not guarantee unique identifiability in the general mixing case, it resolves component-wise indeterminacy by restricting the admissible transformations from arbitrary diffeomorphisms to the affine class. Methodologically, arguing that the stochastic encoding inherent to VAEs obscures the structural residuals required for latent causal discovery, LANCA employs a deterministic Wasserstein Auto-Encoder (WAE) coupled with a differentiable ANM Layer. This architecture transforms residual independence from a passive assumption into an explicit optimization objective. Empirically, LANCA outperforms state-of-the-art baselines on synthetic physics benchmarks (Pendulum, Flow), and on photorealistic environments (CANDLE), where it demonstrates superior robustness to spurious correlations arising from complex background scenes.

Uncovering the underlying causal mechanisms of complex real-world systems remains a significant challenge, as these systems often entail high data collection costs and involve unknown interventions. We introduce MetaCaDI, the first framework to cast the joint discovery of a causal graph and unknown interventions as a meta-learning problem. MetaCaDI is a Bayesian framework that learns a shared causal graph structure across multiple experiments and is optimized to rapidly adapt to new, few-shot intervention target prediction tasks. A key innovation is our model’s analytical adaptation, which uses a closed-form solution to bypass expensive and potentially unstable gradient-based bilevel optimization. Extensive experiments on synthetic and complex gene expression data demonstrate that MetaCaDI significantly outperforms state-of-the-art methods. It excels at both causal graph recovery and identifying intervention targets from as few as 10 data instances, proving its robustness in data-scarce scenarios.