Invariant representation learning aims to extract features from data that remain consistent across different environments or conditions, enhancing model generalization and robustness. This approach is crucial for tasks where the underlying causal mechanisms are stable but the observed data distribution may vary due to confounding factors or domain shifts.