Time Series Forecasting via Reasoning: A Slow-Thinking Approach with Reinforcement Fine-Tuned LLMs
The discussion covers 'Time Series Forecasting via Reasoning,' a model that achieves accuracy through a dual approach. It first uses Supervised Fine-Tuning (SFT) to teach the LLM structured reasoning, then employs Reinforcement Learning (RL) with specific rewards. This method generates explainable forecasts that surpass traditional deep learning models on complex datasets.