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.

Retrieval-augmented generation (RAG) is increasingly recognized as an effective approach to mitigating the hallucination of large language models (LLMs) through the integration of external knowledge. While numerous efforts, most studies focus on a single type of external knowledge source. However, in real-world applications, most situations involve diverse knowledge from various sources, yet this area has been less explored. The main dilemma is the lack of a suitable dataset containing multiple knowledge sources and pre-exploration of the associated issues. To address these challenges, we standardize a benchmark dataset that combines structured and unstructured knowledge across diverse and complementary domains. Based on this dataset, we further develop a plug-and-play RAG framework, PruningRAG, whose main characteristic is the use of multi-granularity pruning strategies to optimize the integration of relevant information while minimizing misleading context. It consistently improves performance across various existing RAG variants, demonstrating its robustness and broad applicability. Building upon the standardized dataset and PruningRAG, we also report a series of experimental results, as well as insightful findings. Our dataset and code are publicly available. https://github.com/USTCAGI/PruningRAG, with the aim of advancing future research in the RAG community.

Modern recommender systems face a critical challenge in modeling the intricate interplay between multi-behavior interactions of users (e.g., clicks, adds-to-cart and purchases) and temporal dynamics that drive evolving preferences. While existing multi-behavior sequential recommendation methods attempt to capture these signals, they often suffer from fragmented modeling, such as decoupling behaviors and items into separate sequences, neglecting time-aware transitions, or relying on computationally intensive architectures that hinder real-world scalability. To address these limitations, we propose GEneralized Alternating Regressor (GEAR), a novel framework that unifies behaviors, items, and temporal contexts into a single autoregressive sequence through an alternating architecture. At its core, GEAR represents user interactions as triplets and processes them through a modular transformer architecture. In this architecture, each triplet is alternately modeled at lower layers to disentangle fine-grained patterns, while upper layers jointly learn cross-signal dependencies. This design mimics the interlocking mechanism of gears, enabling the seamless transitions between multi-behavior dynamics and item transitions. Additionally, we incorporate a time-bias term to quantify the decay of behavioral influence across both short- and long-term horizons. Extensive experiments on real-world datasets validate the effectiveness, generalizability, and computational efficiency of the proposed framework.

For the advancement of time series classification, we can summarize that most existing methods adopt a common learning-to-classify paradigm - a classifier model tries to learn the relation between sequence inputs and target label encoded by one-hot distribution. Although effective, this paradigm conceals two inherent limitations: (1) one-hot distribution fails to reflect the comparability and similarity between labels, and (2) it is difficult to learn transferable representation across domains. In this work, we propose InstructTime, a novel attempt to reshape time series classification as a learning-to-generate paradigm. Relying on the generative capacity of the pre-trained language model, the core idea is to formulate the classification of time series as a multimodal understanding task. Specifically, firstly, a time series discretization module is designed to convert continuous inputs into a sequence of discrete tokens to solve the inconsistency issue across modality data. Secondly, we introduce an alignment projected layer before feeding the transformed token of time series into language models. Thirdly, prior to fine-tuning the language model for the target domain, it is essential to emphasize the necessity of auto-regressive pre-training across various modality inputs. Finally, extensive experimentation are conducted on several prevalent public benchmark datasets, indicating the superior performance of the InstructTime. Our code is at https://github.com/Mingyue-Cheng/InstructTime.

Sequential recommender systems offer personalized suggestions by modeling users' interactions chronologically to capture dynamic user interest. Existing approaches typically fail to adequately describe the dynamics of the entire recommender system, including shifts in both user interest and item availability. To address this, we propose a simple yet effective framework with three key perspectives, tailored to the dynamics of recommender system by fully exploiting the time information. Firstly, we propose a dynamic candidate set construction approach to prevent the model from learning future interactions. Secondly, assuming that user behaviors remain consistent over short terms but may evolve over long terms, we employ a interval-weighted optimization target to model the correlation of users' historical interactions. Finally, we introduce a specialized time-aware attention module to enhance recommendations within specific temporal contexts. Extensive experiments demonstrate the effectiveness and generalizability of our framework. We make our codes publicly available.

Recent efforts have been dedicated to enhancing time series forecasting accuracy by introducing advanced network architectures and self-supervised pretraining strategies. Nevertheless, existing approaches still exhibit two critical drawbacks. Firstly, these methods often rely on a single dataset for training, limiting the model's generalizability due to the restricted scale of the training data. Secondly, the one-step generation schema is widely followed, which necessitates a customized forecasting head and overlooks the temporal dependencies in the output series, and also leads to increased training costs under different horizon length settings.

Deep learning-based algorithms, e.g., convolutional networks, have significantly facilitated multivariate time series classification (MTSC) task. Nevertheless, they suffer from the limitation in modeling long-range dependence due to the nature of convolution operations. Recent advancements have shown the potential of transformers to capture long-range dependence. However, it would incur severe issues, such as fixed scale representations, temporal-invariant and quadratic time complexity, with transformers directly applicable to the MTSC task because of the distinct properties of time series data. To tackle these issues, we propose FormerTime, an hierarchical representation model for improving the classification capacity for the MTSC task. In the proposed FormerTime, we employ a hierarchical network architecture to perform multi-scale feature maps. Besides, a novel transformer encoder is further designed, in which an efficient temporal reduction attention layer and a well-informed contextual positional encoding generating strategy are developed. To sum up, FormerTime exhibits three aspects of merits: (1) learning hierarchical multi-scale representations from time series data, (2) inheriting the strength of both transformers and convolutional networks, and (3) tacking the efficiency challenges incurred by the self-attention mechanism. Extensive experiments performed on 10 publicly available datasets from UEA archive verify the superiorities of the FormerTime compared to previous competitive baselines.