While Large Language Models (LLMs) excel at many reasoning tasks, their native inability to produce calibrated, multi-class probability distributions limits their use in high-stakes Web applications like content moderation and fraud detection. Existing methods to elicit probabilities from LLMs either sacrifice their crucial Chain-of-Thought (CoT) reasoning capabilities or suffer from poor calibration. To address this, we introduce a new paradigm, Verbalized Probability Distribution, and a novel training framework, RLVP (Reinforcement Learning with Verbalized Probabilities). RLVP fine-tunes an LLM to generate both an interpretable CoT and a complete, verbalized probability distribution. We overcome the ''insufficient reward granularity'' problem in standard Reinforcement Learning (RL) for classification by using soft probabilities from expert tabular models as a dense reward curriculum. Through large-scale joint training on 169 tabular tasks, we demonstrate that a single RLVP-trained model can surpass a strong, task-specific XGBoost baseline on up to 55% of tasks. More importantly, the trained model achieves state-of-the-art few-shot performance on unseen, heterogeneous Web benchmarks that mix structured data with free text, achieving performance comparable to or superior than expert models trained on the same limited data. This showcases a strong capability for generalization and knowledge transfer to complex Web data. Our work presents a viable path toward building general-purpose, probabilistically-sound, and interpretable foundation models for the Web.

Accurate personalized headline generation hinges on precisely capturing user interests from historical behaviors. However, existing methods neglect personalized-irrelevant click noise in entire historical clickstreams, which may lead to hallucinated headlines that deviate from genuine user preferences. In this paper, we reveal the detrimental impact of click noise on personalized generation quality through rigorous analysis in both user and news dimensions. Based on these insights, we propose a novel Personalized Headline Generation framework via Denoising Fake Interests from Implicit Feedback (PHG-DIF). PHG-DIF first employs dual-stage filtering to effectively remove clickstream noise, identified by short dwell times and abnormal click bursts, and then leverages multi-level temporal fusion to dynamically model users' evolving and multi-faceted interests for precise profiling. Moreover, we release DT-PENS, a new benchmark dataset comprising the click behavior of 1,000 carefully curated users and nearly 10,000 annotated personalized headlines with historical dwell time annotations. Extensive experiments demonstrate that PHG-DIF substantially mitigates the adverse effects of click noise and significantly improves headline quality, achieving state-of-the-art (SOTA) results on DT-PENS. Our framework implementation and dataset are available at https://github.com/liukejin-up/PHG-DIF.

Large language models (LLMs) have increasingly been explored as powerful text embedders.Existing LLM-based text embedding approaches often leverage the embedding of the final token, typically a reserved special token such as [EOS].However, these tokens have not been intentionally trained to capture the semantics of the whole context, limiting their capacity as text embeddings, especially for retrieval and re-ranking tasks.We propose to add a new training stage before contrastive learning to enrich the semantics of the final token embedding.This stage employs bidirectional generative reconstruction tasks, namely EBQ2D (Embedding-Based Query-to-Document) and EBD2Q (Embedding-Based Document-to-Query), which interleave to anchor the [EOS] embedding and reconstruct either side of Query-Document pairs.Experimental results demonstrate that our additional training stage significantly improves LLM performance on the Massive Text Embedding Benchmark (MTEB), achieving new state-ofthe-art results across different LLM base models and scales. 1

In recent years, support vector regression (SVR) has attracted much attention for nonlinear system identification. It can solve nonlinear problems in the form of linear expressions within the linearly transformed space. Commonly, the convenient kernel trick is applied, which leads to implicit nonlinear mapping by replacing the inner product with a positive definite kernel function. However, only a limited number of kernel functions have been found to work well for the real applications. Moreover, it has been pointed that the implicit nonlinear kernel mapping is not always good, since it may faces the potential over-fitting for some complex and noised learning task. In this paper, explicit nonlinear mapping is learnt by means of the quasi-ARX modeling, and the associated inner product kernel, which is named quasi-linear kernel, is formulated with nonlinearity tunable between the linear and nonlinear kernel functions. Numerical and real systems are simulated to show effectiveness of the quasi-linear kernel, and the proposed identification method is also applied to microarray missing value imputation problem.

Quasi-ARX neurofuzzy (Q-ARX-NF) models have shown great approximation ability and usefulness in nonlinear system identification and control. However, the incorporated neurofuzzy networks suffer from the curse-of-dimensionality problem, which may result in high computational complexity and over-fitting. In this paper, support vector regressor (SVR) based identification approach is used to reduce computational complexity with the help of transforming the original problem into Lagrange space, which is only sensitive to the number of data samples. Furthermore, to improve the generalization capability, a parsimonious model structure is obtained by eliminating insignificant input variables for the incorporated neurofuzzy network, which is implemented by genetic algorithm (GA) based input selection method with a novel fitness evaluation function. Two numerical simulations are tested to show the effectiveness of the proposed method.

This paper presents a novel approach for designing adaptive controller of nonlinear dynamical systems based on an improved quasi-ARX neural network prediction model. The improved quasi-ARX neural network prediction model has two parts: the linear part is used for stability and the nonlinear part is used to satisfy accuracy requirement. Then, we can obtain a linear controller and a nonlinear controller based on the characteristic of the improved quasi-ARX neural network prediction model. A fuzzy switching algorithm is designed between the two controllers. Theory analysis and simulations are given to show the effectiveness of the proposed method both on stability and accuracy.