Retrieval Augmented Time Series Forecasting

Time series forecasting uses historical data to predict future trends, leveraging the relationships between past observations and available features. In this paper, we propose RAFT, a retrieval-augmented time series forecasting method to provide sufficient inductive biases and complement the model's learning capacity. When forecasting the subsequent time frames, we directly retrieve historical data candidates from the training dataset with patterns most similar to the input, and utilize the future values of these candidates alongside the inputs to obtain predictions. This simple approach augments the model's capacity by externally providing information about past patterns via retrieval modules. Our empirical evaluations on ten benchmark datasets show that RAFT consistently outperforms contemporary baselines with an average win ratio of 86%.

Time series forecasting is the task of predicting future trends based on historical data. Traditionally, this involves training models to learn patterns from past observations. In our work, we explore a different approach: instead of just learning these patterns, what if we directly retrieve similar patterns that have occurred in the past for prediction?To do this, we search for patches of past data that are similar to the current input, then retrieve the future values from these patches to help make predictions. This retrieval-based method can capture patterns that lack strong temporal correlations or do not share common characteristics with other patterns, thereby reducing the learning burden and enhancing generalizability.Our approach introduces a novel way of combining the concept of retrieval with time series forecasting, which can benefit applications like weather forecasting or financial analysis. To support further research in this direction, we’ve made our code publicly available.