Learning transferable representations from unlabeled time series is crucial for improving performance in data-scarce classification. Existing self-supervised methods often operate at the point level and rely on unidirectional encoding, leading to low semantic density and a mismatch between pre-training and downstream optimization. In this paper, we propose TimeMAE, a self-supervised framework that reformulates masked modeling for time series via semantic unit elevation and decoupled representation learning. Instead of modeling individual time steps, TimeMAE segments time series into non-overlapping sub-series to form semantically enriched units, enabling more informative masked reconstruction while reducing computational cost. To address the representation discrepancy introduced by masking, we design a decoupled masked autoencoder that separately encodes visible and masked regions, avoiding artificial masked tokens in the main encoder. To guide pre-training, we introduce two complementary objectives: masked codeword classification, which discretizes sub-series semantics via a learned tokenizer and masked representation regression, which aligns continuous representations through a momentum-updated target encoder. Extensive experiments on five datasets demonstrate that TimeMAE outperforms competitive baselines, particularly in label-scarce scenarios and transfer learning scenarios. Our codes are publicly available at https://github.com/Mingyue-Cheng/TimeMAE.

Time series forecasting (TSF) traditionally relies on fast-thinking paradigms that map historical observations directly to future sequences of continuous values. While effective, such approaches often frame forecasting as a pattern-matching problem and tend to overlook explicit reasoning over temporal dynamics and contextual factors, which are critical for modeling long-range dependencies and non-stationary behaviors in real-world scenarios. Recent slow-thinking large language models (LLMs), such as OpenAI o1 and DeepSeek-R1, demonstrate strong inference-time multi-step reasoning abilities. This raises a fundamental question: can slow-thinking LLMs reason over temporal dynamics to support accurate TSF, even without task-specific training? To investigate this question, we present TimeReasoner, a systematic empirical study that reformulates TSF as a conditional reasoning process performed entirely at inference time. TimeReasoner integrates hybrid instructions consisting of task directives, timestamps, sequential values, and optional contextual features, and induces multi-step temporal reasoning in pretrained slow-thinking LLMs through chain-of-thought prompting and rollout-based reasoning strategies. Extensive experiments across diverse TSF benchmarks show that slow-thinking LLMs consistently outperform prior baselines or achieve competitive training-free forecasting performance. Beyond accuracy, we analyze how different inference-time reasoning strategies influence forecasting behaviors, highlighting both the potential and limitations of slow-thinking paradigms for TSF.

Table reasoning requires models to jointly perform comprehensive semantic understanding and precise numerical operations. Although recent large language model (LLM)-based methods have achieved promising results, most of them still rely on a single-turn reasoning paradigm that processes flattened tables in a single forward pass. This paradigm suffers from inherent limitations, including context overflow on large tables, weak sensitivity to continuous numerical values, and the absence of explicit tool-use and reflection. In this paper, we propose TableMind, a tuning-based autonomous programmatic table agent that simulates the human-like cognitive schema of multi-turn interaction within a lightweight LLM. Instead of adopting a training-free workflow design, TableMind learns to internalize planning, action, and reflection through a principled two-stage training strategy. To bootstrap structured table reasoning capabilities, we construct and filter high-quality reasoning data for the supervised fine-tuning (SFT) stage. To enable precise code generation, we introduce a designed multi-perspective reward scheme and a novel optimization objective in the reinforcement learning (RL) stage. Extensive experiments on diverse benchmarks demonstrate that TableMind consistently outperforms previous baselines, validating the effectiveness of training autonomous agents to improve overall performance.

The primary form of user-internet engagement is shifting from leveraging implicit feedback signals, such as browsing and clicks, to harnessing the rich explicit feedback provided by textual interactive behaviors. This shift unlocks a rich source of user textual history, presenting a profound opportunity for a deeper form of personalization. However, prevailing approaches offer only a shallow form of personalization, as they treat user history as a flat list of texts for retrieval and fail to model the rich temporal and semantic structures reflecting dynamic nature of user interests. In this work, we propose MemWeaver, a framework that weaves the user's entire textual history into a hierarchical memory to power deeply personalized generation. The core innovation of our memory lies in its ability to capture both the temporal evolution of interests and the semantic relationships between different activities. To achieve this, MemWeaver builds two complementary memory components that both integrate temporal and semantic information, but at different levels of abstraction: behavioral memory, which captures specific user actions, and cognitive memory, which represents long-term preferences. This dual-component memory serves as a comprehensive representation of the user, allowing large language models (LLMs) to reason over both concrete behaviors and abstracted cognitive traits. This leads to content generation that is deeply aligned with their latent preferences. Experiments on the six datasets of the Language Model Personalization (LaMP) benchmark validate the efficacy of MemWeaver. Our code is available. https://github.com/fishsure/MemWeaver.

Table reasoning with large language models (LLMs) plays a critical role in building intelligent systems capable of understanding and analyzing tabular data. Despite recent progress, existing methods still face key limitations: their reasoning processes lacks depth and explicit multi-step reasoning, often relying solely on implicit language model understanding. In addition, their reasoning processes suffer from instability, primarily caused by model uncertainty. In this work, we propose STaR, a novel slow-thinking model that can achieve effective and stable table reasoning. To enable effective multi-step reasoning, we design a two-stage training framework consisting of supervised fine-tuning (SFT) warm-up followed by reinforced fine-tuning (RFT). Specifically, in the SFT stage, we construct a high-quality dataset through automatic self-verification. In the RFT stage, we introduce a difficulty-aware reinforcement learning mechanism to further enhance reasoning capabilities. Furthermore, to improve reasoning stability, we introduce trajectory-level uncertainty quantification, which fuses token-level confidence with answer-level consistency, enabling the selection of better reasoning trajectories. Extensive experiments demonstrate that STaR-8B achieves state-of-the-art performance on in-domain benchmarks and exhibits strong generalization to out-of-domain datasets, highlighting its potential for enhancing both effectiveness and stability in table reasoning.

Pre-training universal models across multiple domains to enhance downstream tasks is a prevalent learning paradigm. However, there has been minimal progress in pre-training transferable models across domains for time series representation. This dilemma is incurred by two key factors: the limited availability of training set within each domain and the substantial differences in data characteristics between domains. To address these challenges, we present a novel framework, namely CrossTimeNet, designed to perform cross-domain self-supervised pre-training to benefit target tasks. Specifically, to address the issue of data scarcity, we utilize a pre-trained language model as the backbone network to effectively capture the sequence dependencies of the input time series. Meanwhile, we adopt the recovery of corrupted region inputs as a self-supervised optimization objective, taking into account the locality of the time series. To address discrepancies in data characteristics, we introduce a novel tokenization module that converts continuous time series inputs into discrete token sequences using vector quantization techniques. This approach facilitates the learning of transferable time series models across different domains. Extensive experimental results on diverse time series tasks, including classification and forecasting, demonstrate the effectiveness of our approach. Our codes are publicly available at https://github.com/Mingyue-Cheng/CrossTimeNet.