CoRA: Boosting Time Series Foundation Models for Multivariate Forecasting through Correlation-aware Adapter

CoRA addresses a critical and well-documented weakness in current Time Series Foundation Models (TSFMs) like Google's TimesFM and Amazon's Chronos: the reliance on channel-independent (CI) modeling. While CI improves zero-shot generalization, it ignores inter-variable dependencies (e.g., how humidity affects temperature). CoRA provides a 'bolt-on' adapter to fix this. However, its defensibility is extremely low (score 2) because it is currently just a research paper implementation with zero community stars and minimal forks. Technically, it is an incremental refinement of the 'Adapter' pattern widely used in NLP. Frontier labs (OpenAI, Google, Amazon) are already iterating on 'Channel-Dependent' versions of their models; they are more likely to bake these capabilities into the next version of their base weights or release their own official fine-tuning wrappers than to adopt a third-party academic adapter. The displacement horizon is very short (6 months) given the rapid velocity of time-series foundation model research in late 2024 and early 2025.