Finetune-Informed Pretraining Boosts Downstream Performance

TL;DR

The paper addresses the suboptimality of uniform multimodal pretraining when the downstream task relies on a single modality. It introduces Finetune-Informed Pretraining (FIP), which biases learning toward the target modality via asymmetric masking, deeper target decoders, and weighted reconstruction loss within a DenoMAE framework. Applied to constellation-diagram classification in Automatic Modulation Classification (AMC), FIP yields consistent downstream gains, particularly in low-SNR regimes, without additional data or supervision. The approach is simple, architecture-agnostic, and broadly applicable to multimodal masked modeling pipelines where one modality dominates at fine-tuning.

Abstract

Multimodal pretraining is effective for building general-purpose representations, but in many practical deployments, only one modality is heavily used during downstream fine-tuning. Standard pretraining strategies treat all modalities uniformly, which can lead to under-optimized representations for the modality that actually matters. We propose Finetune-Informed Pretraining (FIP), a model-agnostic method that biases representation learning toward a designated target modality needed at fine-tuning time. FIP combines higher masking difficulty, stronger loss weighting, and increased decoder capacity for the target modality, without modifying the shared encoder or requiring additional supervision. When applied to masked modeling on constellation diagrams for wireless signals, FIP consistently improves downstream fine-tuned performance with no extra data or compute. FIP is simple to implement, architecture-compatible, and broadly applicable across multimodal masked modeling pipelines.

Figures (3)

• Figure 1: Reconstruction performance of FIP-DenoMAE.
• Figure 2: Feature representation using t-SNE.
• Figure 3: Classification accuracy at different SNRs.