Non-linear Multitask Learning with Deep Gaussian Processes
This work addresses multi-task learning challenges in probabilistic modeling, offering a novel DGP-based method that is incremental but shows competitive gains.
The paper tackles multi-task learning with Deep Gaussian Processes by introducing non-linear mixtures of latent processes, separating private and shared components to capture within-task and across-task dependencies. The approach outperforms other probabilistic multi-task learning models in real-world and benchmarking settings, demonstrating improved learning performance and information transfer.
We present a multi-task learning formulation for Deep Gaussian processes (DGPs), through non-linear mixtures of latent processes. The latent space is composed of private processes that capture within-task information and shared processes that capture across-task dependencies. We propose two different methods for segmenting the latent space: through hard coding shared and task-specific processes or through soft sharing with Automatic Relevance Determination kernels. We show that our formulation is able to improve the learning performance and transfer information between the tasks, outperforming other probabilistic multi-task learning models across real-world and benchmarking settings.