Depth Uncertainty Networks for Active Learning
This addresses active learning challenges for practitioners by improving model adaptation as dataset size changes, though it appears incremental as a variant of existing BNN methods.
The paper tackled the problem of model bias and overfitting in active learning by introducing Depth Uncertainty Networks (DUNs), a Bayesian neural network variant that infers network depth to adapt complexity, resulting in outperforming other BNN variants on several tasks with notably less overfitting.
In active learning, the size and complexity of the training dataset changes over time. Simple models that are well specified by the amount of data available at the start of active learning might suffer from bias as more points are actively sampled. Flexible models that might be well suited to the full dataset can suffer from overfitting towards the start of active learning. We tackle this problem using Depth Uncertainty Networks (DUNs), a BNN variant in which the depth of the network, and thus its complexity, is inferred. We find that DUNs outperform other BNN variants on several active learning tasks. Importantly, we show that on the tasks in which DUNs perform best they present notably less overfitting than baselines.