Mingqi Wu

Ming Zhang, Jiabao Zhuang, Wenqing Jing et al.

Deep Research Agents increasingly automate survey generation, yet whether they match human experts at retrieving essential papers and organizing them into expert-like taxonomies remains unclear. Existing benchmarks emphasize writing quality or citation correctness, while standard clustering metrics ignore hierarchical structure. We introduce TaxoBench, a benchmark of 72 highly-cited LLM surveys with expert-authored taxonomy trees and 3,815 papers mapped to paper categories. TaxoBench evaluates (1) retrieval via Recall/Precision/F1, and (2) organization at a leaf level (paper-to-category assignment) and a hierarchy level via novel metrics, namely Unordered Semantic Tree Edit Distance US-TED/US-NTED and Semantic Path Similarity Sem-Path. Two modes are supported: Deep Research (topic-only, end-to-end) and Bottom-Up (expert paper set provided, organization-only). To distinguish disagreement with a single expert reference from genuine model failure, we explicitly partition findings into capability-based (reference-free) and alignment-based (reference-dependent). Evaluating 7 Deep Research Agents and 12 frontier LLMs reveals a dual bottleneck: capability-side, the best agent retrieves only 20.92% of expert-cited papers, and 1,000 model taxonomies show 75.9% sibling overlap, 51.2% MECE violations, and 83.4% structural imbalance, all detectable without any reference; alignment-side, all 12 LLMs converge to Sem-Path 28--29%, well below 47--58% achieved by three independent human-annotator groups on the same paper sets. Our benchmark is publicly available at https://github.com/KongLongGeFDU/TaxoBench

Yuntao Du, Jianxun Lian, Jing Yao et al.

Collaborative Filtering (CF) is a widely used and effective technique for recommender systems. In recent decades, there have been significant advancements in latent embedding-based CF methods for improved accuracy, such as matrix factorization, neural collaborative filtering, and LightGCN. However, the explainability of these models has not been fully explored. Adding explainability to recommendation models can not only increase trust in the decisionmaking process, but also have multiple benefits such as providing persuasive explanations for item recommendations, creating explicit profiles for users and items, and assisting item producers in design improvements. In this paper, we propose a neat and effective Explainable Collaborative Filtering (ECF) model that leverages interpretable cluster learning to achieve the two most demanding objectives: (1) Precise - the model should not compromise accuracy in the pursuit of explainability; and (2) Self-explainable - the model's explanations should truly reflect its decision-making process, not generated from post-hoc methods. The core of ECF is mining taste clusters from user-item interactions and item profiles.We map each user and item to a sparse set of taste clusters, and taste clusters are distinguished by a few representative tags. The user-item preference, users/items' cluster affiliations, and the generation of taste clusters are jointly optimized in an end-to-end manner. Additionally, we introduce a forest mechanism to ensure the model's accuracy, explainability, and diversity. To comprehensively evaluate the explainability quality of taste clusters, we design several quantitative metrics, including in-cluster item coverage, tag utilization, silhouette, and informativeness. Our model's effectiveness is demonstrated through extensive experiments on three real-world datasets.

Hao Wang, Jianxun Lian, Mingqi Wu et al.

Sequential user modeling, a critical task in personalized recommender systems, focuses on predicting the next item a user would prefer, requiring a deep understanding of user behavior sequences. Despite the remarkable success of Transformer-based models across various domains, their full potential in comprehending user behavior remains untapped. In this paper, we re-examine Transformer-like architectures aiming to advance state-of-the-art performance. We start by revisiting the core building blocks of Transformer-based methods, analyzing the effectiveness of the item-to-item mechanism within the context of sequential user modeling. After conducting a thorough experimental analysis, we identify three essential criteria for devising efficient sequential user models, which we hope will serve as practical guidelines to inspire and shape future designs. Following this, we introduce ConvFormer, a simple but powerful modification to the Transformer architecture that meets these criteria, yielding state-of-the-art results. Additionally, we present an acceleration technique to minimize the complexity associated with processing extremely long sequences. Experiments on four public datasets showcase ConvFormer's superiority and confirm the validity of our proposed criteria.

Yuming Yang, Mingyoung Lai, Wanxu Zhao et al.

Long chain-of-thought (CoT) trajectories provide rich supervision signals for distilling reasoning from teacher to student LLMs. However, both prior work and our experiments show that trajectories from stronger teachers do not necessarily yield better students, highlighting the importance of data-student suitability in distillation. Existing methods assess suitability primarily through student likelihood, favoring trajectories that closely align with the model's current behavior but overlooking more informative ones. Addressing this, we propose Rank-Surprisal Ratio (RSR), a simple metric that captures both alignment and informativeness to assess the suitability of a reasoning trajectory. RSR is motivated by the observation that effective trajectories typically combine low absolute probability with relatively high-ranked tokens under the student model, balancing learning signal strength and behavioral alignment. Concretely, RSR is defined as the ratio of a trajectory's average token-wise rank to its average negative log-likelihood, and is straightforward to compute and interpret. Across five student models and reasoning trajectories from 11 diverse teachers, RSR strongly correlates with post-training performance (average Spearman 0.86), outperforming existing metrics. We further demonstrate its practical utility in both trajectory selection and teacher selection.

Yuming Yang, Yang Nan, Junjie Ye et al.

Data diversity is crucial for the instruction tuning of large language models. Existing studies have explored various diversity-aware data selection methods to construct high-quality datasets and enhance model performance. However, the fundamental problem of precisely defining and measuring data diversity remains underexplored, limiting clear guidance for data engineering. To address this, we systematically analyze 11 existing diversity measurement methods by evaluating their correlation with model performance through extensive fine-tuning experiments. Our results indicate that a reliable diversity measure should properly account for both inter-sample differences and the information density in the sample space. Building on this, we propose NovelSum, a new diversity metric based on sample-level "novelty." Experiments on both simulated and real-world data show that NovelSum accurately captures diversity variations and achieves a 0.97 correlation with instruction-tuned model performance, highlighting its value in guiding data engineering practices. With NovelSum as an optimization objective, we further develop a greedy, diversity-oriented data selection strategy that outperforms existing approaches, validating both the effectiveness and practical significance of our metric. The code is available at https://github.com/UmeanNever/NovelSum.

Wenlong Shi, Jianxun Lian, Mingqi Wu et al.

Large language models (LLMs) increasingly serve as interactive social agents, yet their ability to maintain coherent and authentic persona-level role-playing remains limited, particularly in realistic social scenarios. Existing research predominantly focuses on character-level settings and relies on static evaluation formats, failing to capture the complexity of everyday social interactions. In this work, we present PersonaArena, a dynamic simulation framework for evaluating and improving persona-level role-playing in LLMs. PersonaArena leverages a large, filtered corpus of user-generated social content to construct a nuanced persona bank, and elicits multi-turn, context-rich interactions within simulated social environments. Our framework features a multi-agent debating judge for holistic and unbiased assessment. Through extensive experiments, we demonstrate that PersonaArena enables rigorous evaluation and enhancement of LLMs' role-playing capabilities, advancing the development of more authentic and socially adept AI agents.

Fengxiang Li, Han Zhang, Haoyang Huang et al.

We present KAT-Coder-V2, an agentic coding model developed by the KwaiKAT team at Kuaishou. KAT-Coder-V2 adopts a "Specialize-then-Unify" paradigm that decomposes agentic coding into five expert domains - SWE, WebCoding, Terminal, WebSearch, and General - each undergoing independent supervised fine-tuning and reinforcement learning, before being consolidated into a single model via on-policy distillation. We develop KwaiEnv, a modular infrastructure sustaining tens of thousands of concurrent sandbox instances, and scale RL training along task complexity, intent alignment, and scaffold generalization. We further propose MCLA for stabilizing MoE RL training and Tree Training for eliminating redundant computation over tree-structured trajectories with up to 6.2x speedup. KAT-Coder-V2 achieves 79.6% on SWE-bench Verified (vs. Claude Opus 4.6 at 80.8%), 88.7 on PinchBench (surpassing GLM-5 and MiniMax M2.7), ranks first across all three frontend aesthetics scenarios, and maintains strong generalist scores on Terminal-Bench Hard (46.8) and tau^2-Bench (93.9). Our model is publicly available at https://streamlake.com/product/kat-coder.

Hao Liao, Wensheng Lu, Jianxun Lian et al.

Large Language Models (LLMs) have shown promise for generative recommender systems due to their transformative capabilities in user interaction. However, ensuring they do not recommend out-of-domain (OOD) items remains a challenge. We study two distinct methods to address this issue: RecLM-ret, a retrieval-based method, and RecLM-cgen, a constrained generation method. Both methods integrate seamlessly with existing LLMs to ensure in-domain recommendations. Comprehensive experiments on three recommendation datasets demonstrate that RecLM-cgen consistently outperforms RecLM-ret and existing LLM-based recommender models in accuracy while eliminating OOD recommendations, making it the preferred method for adoption. Additionally, RecLM-cgen maintains strong generalist capabilities and is a lightweight plug-and-play module for easy integration into LLMs, offering valuable practical benefits for the community. Source code is available at https://github.com/microsoft/RecAI

Mingqi Wu, Qiang Sun

Interpolators are unstable. For example, the mininum $\ell_2$ norm least square interpolator exhibits unbounded test errors when dealing with noisy data. In this paper, we study how ensemble stabilizes and thus improves the generalization performance, measured by the out-of-sample prediction risk, of an individual interpolator. We focus on bagged linear interpolators, as bagging is a popular randomization-based ensemble method that can be implemented in parallel. We introduce the multiplier-bootstrap-based bagged least square estimator, which can then be formulated as an average of the sketched least square estimators. The proposed multiplier bootstrap encompasses the classical bootstrap with replacement as a special case, along with a more intriguing variant which we call the Bernoulli bootstrap. Focusing on the proportional regime where the sample size scales proportionally with the feature dimensionality, we investigate the out-of-sample prediction risks of the sketched and bagged least square estimators in both underparametrized and overparameterized regimes. Our results reveal the statistical roles of sketching and bagging. In particular, sketching modifies the aspect ratio and shifts the interpolation threshold of the minimum $\ell_2$ norm estimator. However, the risk of the sketched estimator continues to be unbounded around the interpolation threshold due to excessive variance. In stark contrast, bagging effectively mitigates this variance, leading to a bounded limiting out-of-sample prediction risk. To further understand this stability improvement property, we establish that bagging acts as a form of implicit regularization, substantiated by the equivalence of the bagged estimator with its explicitly regularized counterpart. We also discuss several extensions.

Zhihao Zhang, Qiaole Dong, Qi Zhang et al.

Post-training algorithms such as Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) are widely used to adapt multimodal large language models to downstream tasks. While effective at task adaptation, their impact on prior knowledge remains unclear. In this paper, we introduce jigsaw puzzles as a novel task absent from existing pretraining corpora and systematically study the behavior of SFT and RFT on open-source multimodal model, Qwen2.5-VL series. Our experiments reveal a sharp trade-off: SFT enables rapid task acquisition but leads to catastrophic forgetting, whereas RFT learns more slowly but maintains prior knowledge. We study this phenomenon through learning dynamics by examining both the magnitude and direction of how training data influence prior knowledge. Our analysis shows that RFT mainly reinforces correct samples naturally aligned with the base model's probability landscape, leading to weaker interference with prior knowledge. Moreover, training on RFT-simulated rollouts, which exert a small magnitude of influence and are well aligned in direction to prior knowledge, allows SFT to preserve prior knowledge better while rapidly learning new tasks. These findings suggest that distribution of training data, rather than algorithmic differences, plays a central role in forgetting, and highlight RFT's potential for stable continual learning in multimodal large language models.

Mingqi Wu, Zhihao Zhang, Qiaole Dong et al.

Reasoning in large language models has long been a central research focus, and recent studies employing reinforcement learning (RL) have introduced diverse methods that yield substantial performance gains with minimal or even no external supervision. Surprisingly, some studies even suggest that random or incorrect reward signals can enhance performance. However, these breakthroughs are predominantly observed for the mathematically strong Qwen2.5 series on benchmarks such as MATH-500, AMC, and AIME, and seldom transfer to models like Llama, which warrants a more in-depth investigation. In this work, our empirical analysis reveals that pre-training on massive web-scale corpora leaves Qwen2.5 susceptible to data contamination in widely used benchmarks. Consequently, conclusions derived from contaminated benchmarks on Qwen2.5 series may be unreliable. To obtain trustworthy evaluation results, we introduce a generator that creates fully clean arithmetic problems of arbitrary length and difficulty, dubbed RandomCalculation. Using this leakage-free dataset, we show that only accurate reward signals yield steady improvements that surpass the base model's performance boundary in mathematical reasoning, whereas random or incorrect rewards do not. Moreover, we conduct more fine-grained analyses to elucidate the factors underlying the different performance observed on the MATH-500 and RandomCalculation benchmarks. Consequently, we recommend that future studies evaluate models on uncontaminated benchmarks and, when feasible, test various model series to ensure trustworthy conclusions about RL and related methods.

Mingqi Wu, Qiang Sun, Yi Yang

High-dimensional data often contain low-dimensional signals obscured by structured background noise, which limits the effectiveness of standard PCA. Motivated by contrastive learning, we address the problem of recovering shared signal subspaces from positive pairs, paired observations sharing the same signal but differing in background. Our baseline, PCA+, uses alignment-only contrastive learning and succeeds when background variation is mild, but fails under strong noise or high-dimensional regimes. To address this, we introduce PCA++, a hard uniformity-constrained contrastive PCA that enforces identity covariance on projected features. PCA++ has a closed-form solution via a generalized eigenproblem, remains stable in high dimensions, and provably regularizes against background interference. We provide exact high-dimensional asymptotics in both fixed-aspect-ratio and growing-spike regimes, showing uniformity's role in robust signal recovery. Empirically, PCA++ outperforms standard PCA and alignment-only PCA+ on simulations, corrupted-MNIST, and single-cell transcriptomics, reliably recovering condition-invariant structure. More broadly, we clarify uniformity's role in contrastive learning, showing that explicit feature dispersion defends against structured noise and enhances robustness.

Mingqi Wu, Archer Y. Yang, Qiang Sun

Iterative self-training (self-distillation) repeatedly refits a model on pseudo-labels generated by its own predictions. We study this procedure in overparameterized linear regression: an initial estimator is trained on noisy labels, and each subsequent iterate is trained on fresh covariates with noiseless pseudo-labels from the previous model. In the high-dimensional regime, we derive deterministic-equivalent recursions for the prediction risk and effective noise across iterations, and prove that the empirical quantities concentrate sharply around these limits. The recursion separates two competing forces: a systematic component that grows with iteration due to progressive signal forgetting, and a stochastic component that decays due to denoising via repeated data-dependent projections. Their interaction yields a $U$-shaped test-risk curve and an optimal early-stopping time. In spiked covariance models, iteration further acts as an iteration-dependent spectral filter that preserves strong eigendirections while suppressing weaker ones, inducing an implicit form of soft feature selection distinct from ridge regression. Finally, we propose an iterated generalized cross-validation criterion and prove its uniform consistency for estimating the risk along the self-training trajectory, enabling fully data-driven selection of the stopping time and regularization. Experiments on synthetic covariances validate the theory and illustrate the predicted denoising-forgetting trade-off.

Ming Zhang, Yujiong Shen, Jingyi Deng et al.

Existing evaluation of Large Language Models (LLMs) on static benchmarks is vulnerable to data contamination and leaderboard overfitting, critical issues that obscure true model capabilities. To address this, we introduce LLMEval-3, a framework for dynamic evaluation of LLMs. LLMEval-3 is built on a proprietary bank of 220k graduate-level questions, from which it dynamically samples unseen test sets for each evaluation run. Its automated pipeline ensures integrity via contamination-resistant data curation, a novel anti-cheating architecture, and a calibrated LLM-as-a-judge process achieving 90% agreement with human experts, complemented by a relative ranking system for fair comparison. An 20-month longitudinal study of nearly 50 leading models reveals a performance ceiling on knowledge memorization and exposes data contamination vulnerabilities undetectable by static benchmarks. The framework demonstrates exceptional robustness in ranking stability and consistency, providing strong empirical validation for the dynamic evaluation paradigm. LLMEval-3 offers a robust and credible methodology for assessing the true capabilities of LLMs beyond leaderboard scores, promoting the development of more trustworthy evaluation standards.

Konstantin Donhauser, Mingqi Wu, Fanny Yang

Kernel ridge regression is well-known to achieve minimax optimal rates in low-dimensional settings. However, its behavior in high dimensions is much less understood. Recent work establishes consistency for kernel regression under certain assumptions on the ground truth function and the distribution of the input data. In this paper, we show that the rotational invariance property of commonly studied kernels (such as RBF, inner product kernels and fully-connected NTK of any depth) induces a bias towards low-degree polynomials in high dimensions. Our result implies a lower bound on the generalization error for a wide range of distributions and various choices of the scaling for kernels with different eigenvalue decays. This lower bound suggests that general consistency results for kernel ridge regression in high dimensions require a more refined analysis that depends on the structure of the kernel beyond its eigenvalue decay.