Problem
Research questions and friction points this paper is trying to address.
Time Series Forecasting
Decoder-Only
Latent Steering
Retrieval-Based
Inference-Time Improvement
SteerCast: Retrieval-Based Latent Steering for Decoder-Only Time Series Forecasting
🤖 AI Summary
This study addresses the limited inference accuracy of decoder-only time series forecasting models by proposing a retrieval-based latent guidance method. During inference, the approach dynamically corrects autoregressive generation trajectories using latent-space guidance vectors. Its core mechanism integrates nearest-neighbor retrieval, latent vector aggregation, and hidden-state injection to achieve plug-and-play optimization without updating model parameters. Extensive evaluations across multiple benchmarks demonstrate that the proposed method significantly improves forecasting accuracy with no additional training overhead. By enabling effective inference-time adaptation, this work establishes an efficient new paradigm for enhancing the predictive capabilities of pretrained time series models.
📝 Abstract
Time series forecasting aims to predict future values from historical observations and auxiliary features. We propose \textbf{SteerCast}, a retrieval-based latent steering method that improves decoder-only forecaster at inference time, without updating its parameters. SteerCast constructs a database from the training set by storing a representation of each history window together with a \emph{steering vector} computed in the forecaster's latent space, defined as the difference between representations induced by the ground-truth continuation and by the model's own prediction. At test time, SteerCast retrieves nearest neighbors for a query history, aggregates their steering vectors, and injects the resulting signal into the forecaster's hidden states at every step of autoregressive generation, guiding predictions toward trajectories consistent with similar training cases. Experiments across diverse multivariate benchmarks and multiple horizons show that SteerCast consistently improves forecasting accuracy over the fine-tuned backbone and retrieval-based baselines, while requiring no additional training beyond the original fine-tuning and using only the training set as a retrieval corpus.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Retrieval-Based Latent Steering
Decoder-Only Time Series Forecasting
Steering Vector
Inference-Time Enhancement
Autoregressive Generation
🔎 Similar Papers
No similar papers found.
