Transfer Learning with Reconstruction Loss
Wei Cui, Wei Yu
TL;DR
This work tackles transferring learning across correlated optimization tasks that share inputs by introducing common information $\mathcal{I}(\mathbf{p})$ and a reconstruction-based loss. The method adds a reconstruction branch from a selected feature layer so that the features retain the common information, with total loss $\mathcal{L}=\mathcal{L}_S+\alpha\mathcal{L}_R$, enabling a target-task agnostic transfer via sharing the feature layer. Demonstrations on MNIST classification and wireless problems (D2D power control and MISO beamforming/localization) show superior target-task data efficiency and reduced overfitting compared with conventional transfer learning, while incurring minimal expense at deployment. The results support that aligning learned representations with the underlying common information yields transferable features across varied optimization tasks, enabling efficient adaptation to new objectives with limited data.
Abstract
In most applications of utilizing neural networks for mathematical optimization, a dedicated model is trained for each specific optimization objective. However, in many scenarios, several distinct yet correlated objectives or tasks often need to be optimized on the same set of problem inputs. Instead of independently training a different neural network for each problem separately, it would be more efficient to exploit the correlations between these objectives and to train multiple neural network models with shared model parameters and feature representations. To achieve this, this paper first establishes the concept of common information: the shared knowledge required for solving the correlated tasks, then proposes a novel approach for model training by adding into the model an additional reconstruction stage associated with a new reconstruction loss. This loss is for reconstructing the common information starting from a selected hidden layer in the model. The proposed approach encourages the learned features to be general and transferable, and therefore can be readily used for efficient transfer learning. For numerical simulations, three applications are studied: transfer learning on classifying MNIST handwritten digits, the device-to-device wireless network power allocation, and the multiple-input-single-output network downlink beamforming and localization. Simulation results suggest that the proposed approach is highly efficient in data and model complexity, is resilient to over-fitting, and has competitive performances.