REATS is proposed, which leverages LLM reasoning capabilities as an intelligent ensemble router that jointly processes textual temporal pattern descriptions and numerical features to produce interpretable, sample-adaptive ensemble weights through chain-of-thought reasoning.
Abstract
Due to the diversity of real-world time series, no single forecasting model consistently dominates across all samples. Ensemble learning addresses this by combining complementary model strengths, yet existing methods rely on fixed rules or black-box models based solely on numerical inputs, failing to leverage LLM reasoning for interpretable weighting decisions. We propose REATS, which leverages LLM reasoning capabilities as an intelligent ensemble router that jointly processes textual temporal pattern descriptions and numerical features to produce interpretable, sample-adaptive ensemble weights through chain-of-thought reasoning. To enable effective LLM-based ensembling, we study its key design choices and propose: (i) a structured input pipeline that transforms raw time series into hybrid textual--numerical representations with fixed token cost, enabling rule-based chain-of-thought construction without API dependency, augmented with retrieved similar-sample priors; (ii) a diverse multi-row weight supervision scheme coupled with a token-efficient percentage-table format that reduces numerical complexity and mitigates LLM hallucinations; and (iii) a two-stage fine-tuning framework combining SFT with GRPO, where a reciprocal reward mapping transforms the continuous unbounded MSE gap into bounded signals with amplified near-oracle sensitivity, addressing the uniform sensitivity and outlier-dominated advantage compression inherent in naive reward designs for regression-based GRPO. Experiments on eight benchmarks demonstrate that REATS outperforms competitive ensemble baselines while providing natural language explanations and demonstrating strong transfer learning and out-of-domain generalization to unseen candidate models.
This work investigates LLM-based forecasting agents, meaning systems in which a language model contributes to a scored prediction about a future or currently unobserved target, and organizes architectures into three groups.
Xiaogang Xu, Jiaqi Tang, Jianmin Chen et al.· 0 citations
State-of-the-art multivariate time-series forecasters can model complex temporal and cross-variable dependencies, yet their opaque representations provide limited insight into why a particular forecast is produced. This lack of transparency restricts their use in settings where practitioners must understand and assess the factors underlying a prediction. We introduce ConceptTS, an interpretable forecasting framework that organizes its predictions around named, human-readable concepts. ConceptTS uses a large language model to propose task-relevant concepts and generate executable labeling rules, translating the language model's domain knowledge into direct supervision without costly manual concept annotation. The proposed concepts are organized into three complementary bottlenecks that describe the historical context, local forecast intervals, and the full forecast horizon. A shared decoder combines representations derived from their predicted activations to construct the forecast, making the model's decision process explicit and supporting direct concept-level interventions. Experiments on the Beijing Multi-Site Air Quality dataset show that ConceptTS achieves accuracy competitive with strong black-box baselines while producing semantically meaningful concept activations.
Time series forecasting (TSF) is critical in numerous real-world applications, yet its sequential scalar presentation limits semantic richness and the capture of complex temporal patterns. Recent advances leveraging patchwise modeling and pretrained large language models (LLMs) have achieved notable progress. However, existing methods largely focus on raw sequential patterns while overlooking intrapatch semantics, which limits the richness of time series representations, and they fail to effectively exploit complementary information from multiple views. To tackle these challenges, we propose a self-driven multiview architecture for TSF (SMArT). SMArT enriches intrapatch semantics through self-supervised multiview fusion. It jointly captures fine-grained temporal 1-D dependencies via pointwise self-attention and global temporal 2-D structures via Gramian angular fields (GAFs), all without requiring external supervision. To further bridge the gap between time series data and LLMs, SMArT pioneers a dual-prompt strategy, combining static, dataset-level guidance with dynamic, input-specific prompts derived from frequency-domain decomposition, significantly enhancing LLMs' adaptability and generalization. Extensive experiments across diverse TSF tasks validate SMArT's robustness, achieving state-of-the-art performance in long-term forecasting and excelling in few-shot and zero-shot scenarios. By integrating self-driven multiview learning with LLMs' reasoning power, SMArT establishes an effective framework for TSF. Our code and Supplementary Materials are available at https://github.com/BMRETURN/SMArT.
Wen-Bin Xing, Meng-Ran Li, Bo-Yu Zhang et al.· IEEE Transactions on Neural...· 0 citations
Most time series (TS) models are specialized for a single task, either understanding (i.e., returning text answers about a TS) or generation (i.e., returning a numeric forecast). Only recently have unified models begun to handle the two within a single architecture. Even these models, however, produce the two outputs as task-separated paths and cannot predict a series and explain why that prediction arises within a single coherent response. In this paper, we argue for a task-fused model that jointly produces 1) prediction (generation) and 2) selfexplanation (understanding), thereby integrating 1) numerical TS forecasting and 2) interpretable text reasoning within a single response. To enable the systematic study of this capability, we present both a benchmark and a recipe that jointly address the two tasks. The benchmark, ReasonTS-Bench, identifies five fundamental patterns underlying TS and enables the joint evaluation of both tasks. ReasonCast, our recipe for finetuning any LLM to perform both tasks jointly, yields a model that generates a reasoning chain and a forecast together in a single autoregressive pass. Extensive experiments show that ReasonCast outperforms both LLMs and TS models on prediction accuracy while producing verifiable, causal reasoning. Code is available at: https://github.com/seunghan96/reasoncast.
Seunghan Lee, Jun Seo, Jaehoon Lee et al.· 0 citations
A three-paradigm taxonomy (feature-based, auxiliary-based, and policy-based) based on the functional role of LLMs within the RL pipeline is proposed, which provides superior scalability and stability, though often at the expense of representational depth.
Ghusoon Hadi al-Aldaffaie, Alireza Taheri, Amirfarhad Farhadi et al.· Discover Artificial Intellig...· 0 citations
