TS-Memory: Plug-and-Play Memory for Time Series Foundation Models

TL;DR

TS-Memory introduces a lightweight, plug-and-play memory adapter for Time Series Foundation Models, enhancing zero-shot forecasting under distribution shifts without increasing inference latency.

Contribution

It proposes a novel parametric memory distillation method, TS-Memory, that improves TSFM adaptation efficiency and effectiveness through confidence-aware retrieval synthesis and distillation.

Findings

Consistent improvements in point forecasting accuracy.

Enhanced probabilistic forecasting performance.

Efficient retrieval-free deployment with constant-time overhead.

Abstract

Time Series Foundation Models (TSFMs) achieve strong zero-shot forecasting through large-scale pre-training, but adapting them to downstream domains under distribution shift remains challenging. Existing solutions face a trade-off: Parametric Adaptation can cause catastrophic forgetting and requires costly multi-domain maintenance, while Non-Parametric Retrieval improves forecasts but incurs high inference latency due to datastore search. We propose Parametric Memory Distillation and implement it as TS-Memory, a lightweight memory adapter that augments frozen TSFMs. TS-Memory is trained in two stages. First, we construct an offline, leakage-safe kNN teacher that synthesizes confidence-aware quantile targets from retrieved futures. Second, we distill this retrieval-induced distributional correction into a lightweight memory adapter via confidence-gated supervision. During inference, TS-Memory fuses memory and backbone predictions with constant-time overhead, enabling retrieval-free deployment. Experiments across diverse TSFMs and benchmarks demonstrate consistent improvements in both point and probabilistic forecasting over representative adaptation methods, with efficiency comparable to the frozen backbone.