arXiv subjects

Rui Li

Publications and source records attributed to Rui Li.

At least 235 records · Page 13Linked to original sources

arXiv2025

SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines

Large language models (LLMs) have demonstrated remarkable proficiency in mainstream academic disciplines such as mathematics, physics, and computer science. However, human knowledge encompasses over 200 specialized disciplines, far exceeding the scope of existing benchmarks. The capabilities of LLMs in many of these specialized fields-particularly in light industry, agriculture, and service-oriented disciplines-remain inadequately evaluated. To address this gap, we present SuperGPQA, a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines. Our benchmark employs a novel Human-LLM collaborative filtering mechanism to eliminate trivial or ambiguous questions through iterative refinement based on both LLM responses and expert feedback. Our experimental results reveal significant room for improvement in the performance of current state-of-the-art LLMs across diverse knowledge domains (e.g., the reasoning-focused model DeepSeek-R1 achieved the highest accuracy of 61.82% on SuperGPQA), highlighting the considerable gap between current model capabilities and artificial general intelligence. Additionally, we present comprehensive insights from our management of a large-scale annotation process, involving over 80 expert annotators and an interactive Human-LLM collaborative system, offering valuable methodological guidance for future research initiatives of comparable scope.

cs.CL↗

arXiv2025

Flatness Improves Backbone Generalisation in Few-shot Classification

Deployment of deep neural networks in real-world settings typically requires adaptation to new tasks with few examples. Few-shot classification (FSC) provides a solution to this problem by leveraging pre-trained backbones for fast adaptation to new classes. However, approaches for multi-domain FSC typically result in complex pipelines aimed at information fusion and task-specific adaptation without consideration of the importance of backbone training. In this work, we introduce an effective strategy for backbone training and selection in multi-domain FSC by utilizing flatness-aware training and fine-tuning. Our work is theoretically grounded and empirically performs on par or better than state-of-the-art methods despite being simpler. Further, our results indicate that backbone training is crucial for good generalisation in FSC across different adaptation methods.

cs.LG↗

arXiv2025

Towards An Efficient LLM Training Paradigm for CTR Prediction

Large Language Models (LLMs) have demonstrated tremendous potential as the next-generation ranking-based recommendation system. Many recent works have shown that LLMs can significantly outperform conventional click-through-rate (CTR) prediction approaches. Despite such promising results, the computational inefficiency inherent in the current training paradigm makes it particularly challenging to train LLMs for ranking-based recommendation tasks on large datasets. To train LLMs for CTR prediction, most existing studies adopt the prevalent ''sliding-window'' paradigm. Given a sequence of $m$ user interactions, a unique training prompt is constructed for each interaction by designating it as the prediction target along with its preceding $n$ interactions serving as context. In turn, the sliding-window paradigm results in an overall complexity of $O(mn^2)$ that scales linearly with the length of user interactions. Consequently, a direct adoption to train LLMs with such strategy can result in prohibitively high training costs as the length of interactions grows. To alleviate the computational inefficiency, we propose a novel training paradigm, namely Dynamic Target Isolation (DTI), that structurally parallelizes the training of $k$ (where $k >> 1$) target interactions. Furthermore, we identify two major bottlenecks - hidden-state leakage and positional bias overfitting - that limit DTI to only scale up to a small value of $k$ (e.g., 5) then propose a computationally light solution to effectively tackle each. Through extensive experiments on three widely adopted public CTR datasets, we empirically show that DTI reduces training time by an average of $\textbf{92%}$ (e.g., from $70.5$ hrs to $5.31$ hrs), without compromising CTR prediction performance.

cs.IR↗

arXiv2025

From Captions to Rewards (CAREVL): Leveraging Large Language Model Experts for Enhanced Reward Modeling in Large Vision-Language Models

Aligning large vision-language models (LVLMs) with human preferences is challenging due to the scarcity of fine-grained, high-quality, and multimodal preference data without human annotations. Existing methods relying on direct distillation often struggle with low-confidence data, leading to suboptimal performance. To address this, we propose CAREVL, a novel method for preference reward modeling by reliably using both high- and low-confidence data. First, a cluster of auxiliary expert models (textual reward models) innovatively leverages image captions as weak supervision signals to filter high-confidence data. The high-confidence data are then used to fine-tune the LVLM. Second, low-confidence data are used to generate diverse preference samples using the fine-tuned LVLM. These samples are then scored and selected to construct reliable chosen-rejected pairs for further training. CAREVL achieves performance improvements over traditional distillation-based methods on VL-RewardBench and MLLM-as-a-Judge benchmark, demonstrating its effectiveness. The code will be released soon.

cs.CV↗

arXiv2025

Capacitively Shunted Double-Transmon Coupler Realizing Bias-Free Idling and High-Fidelity CZ Gate

A high-fidelity CZ gate utilizing a double-transmon coupler (DTC) has recently been demonstrated as a building block for superconducting quantum processors. Like many other kinds of tunable couplers, however, the DTC requires a finite DC current for flux-biasing the coupler at the idling point to turn off the coupling, necessitating extra care for wiring and heat-load management. To address this issue, we theoretically propose and experimentally realize a novel coupling scheme by introducing a shunt capacitance between the two transmons of the DTC at zero-flux bias, which demonstrates high-fidelity CZ-gate performance comparable to the previous DTC. Through a comprehensive error budget analysis using multiple randomized benchmarking methods, we also identify that the current fidelity is limited by the decoherence through the coupler. Moreover, we experimentally demonstrate the wide operational flux range of the capacitively shunted DTC, which solves the challenging issue of remnant flux existing even with careful magnetic shielding.

quant-ph↗

arXiv2025

Spectral Radius of Graphs with Size Constraints: Resolving a Conjecture of Guiduli

We resolve a problem posed by Guiduli (1996) on the spectral radius of graphs satisfying the Hereditarily Bounded Property $P_{t,r}$, which requires that every subgraph $H$ with $|V(H)| \geq t$ satisfies $|E(H)| \leq t|V(H)| + r$. For an $n$-vertex graph $G$ satisfying $P_{t,r}$, where $t > 0$ and $r \geq -\binom{\lfloor t+1 \rfloor}{2}$, we prove that the spectral radius $ρ(G)$ is bounded above by $ρ(G) \leq c(s,t) + \sqrt{\lfloor t \rfloor n}$, where $s = \binom{\lfloor t \rfloor + 1}{2} + r$, thus affirmatively answering Guiduli's conjecture. Furthermore, we present a complete characterization of the extremal graphs that achieve this bound. These graphs are constructed as the join graph $K_{\lfloor t \rfloor} \nabla F$, where $F$ is either $K_3 \cup (n - \lfloor t \rfloor - 3)K_1$ or a forest consisting solely of star structures. The specific structure of such forests is meticulously characterized. Central to our analysis is the introduction of a novel potential function $η(F) = e(F) + (\lfloor t \rfloor - t)|V(F)|$, which quantifies the structural "positivity" of subgraphs. By combining edge-shifting operations with spectral radius maximization principles, we establish sharp bounds on $η^+(G)$, the cumulative positivity of $G$. Our results contribute to the understanding of spectral extremal problems under edge-density constraints and provide a framework for analyzing similar hereditary properties.

math.CO↗

arXiv2025

AeroReformer: Aerial Referring Transformer for UAV-based Referring Image Segmentation

As a novel and challenging task, referring segmentation combines computer vision and natural language processing to localize and segment objects based on textual descriptions. While referring image segmentation (RIS) has been extensively studied in natural images, little attention has been given to aerial imagery, particularly from unmanned aerial vehicles (UAVs). The unique challenges of UAV imagery, including complex spatial scales, occlusions, and varying object orientations, render existing RIS approaches ineffective. A key limitation has been the lack of UAV-specific datasets, as manually annotating pixel-level masks and generating textual descriptions is labour-intensive and time-consuming. To address this gap, we design an automatic labelling pipeline that leverages pre-existing UAV segmentation datasets and Multimodal Large Language Models (MLLM) for generating textual descriptions. Furthermore, we propose Aerial Referring Transformer (AeroReformer), a novel framework for UAV referring image segmentation (UAV-RIS), featuring a Vision-Language Cross-Attention Module (VLCAM) for effective cross-modal understanding and a Rotation-Aware Multi-Scale Fusion (RAMSF) decoder to enhance segmentation accuracy in aerial scenes. Extensive experiments on two newly developed datasets demonstrate the superiority of AeroReformer over existing methods, establishing a new benchmark for UAV-RIS. The datasets and code will be publicly available at: https://github.com/lironui/AeroReformer.

cs.CV↗

arXiv2025

Variable degeneracy on toroidal graphs

DP-coloring was introduced by Dvořák and Postle as a generalization of list coloring and signed coloring. A new coloring, strictly $f$-degenerate transversal, is a further generalization of DP-coloring and $L$-forested-coloring. In this paper, we present some structural results on planar and toroidal graphs with forbidden configurations, and establish some sufficient conditions for the existence of strictly $f$-degenerate transversal based on these structural results. Consequently, (i) every toroidal graph without subgraphs isomorphic to the configurations in Fig.2 is DP-$4$-colorable, and has list vertex arboricity at most $2$, (ii) every toroidal graph without $4$-cycles is DP-$4$-colorable, and has list vertex arboricity at most $2$, (iii) every planar graph without subgraphs isomorphic to the configurations in Fig.3 is DP-$4$-colorable, and has list vertex arboricity at most $2$. These results improve upon previous results on DP-$4$-coloring [Discrete Math. 341~(7) (2018) 1983--1986; Bull. Malays. Math. Sci. Soc. 43~(3) (2020) 2271--2285] and (list) vertex arboricity [Discrete Math. 333 (2014) 101--105; Int. J. Math. Stat. 16~(1) (2015) 97--105; Iranian Math. Soc. 42~(5) (2016) 1293--1303].

math.CO↗

arXiv2025

Model Adaptation: Unsupervised Domain Adaptation without Source Data

In this paper, we investigate a challenging unsupervised domain adaptation setting -- unsupervised model adaptation. We aim to explore how to rely only on unlabeled target data to improve performance of an existing source prediction model on the target domain, since labeled source data may not be available in some real-world scenarios due to data privacy issues. For this purpose, we propose a new framework, which is referred to as collaborative class conditional generative adversarial net to bypass the dependence on the source data. Specifically, the prediction model is to be improved through generated target-style data, which provides more accurate guidance for the generator. As a result, the generator and the prediction model can collaborate with each other without source data. Furthermore, due to the lack of supervision from source data, we propose a weight constraint that encourages similarity to the source model. A clustering-based regularization is also introduced to produce more discriminative features in the target domain. Compared to conventional domain adaptation methods, our model achieves superior performance on multiple adaptation tasks with only unlabeled target data, which verifies its effectiveness in this challenging setting.

cs.CV↗

arXiv2025

Be a Multitude to Itself: A Prompt Evolution Framework for Red Teaming

Large Language Models (LLMs) have gained increasing attention for their remarkable capacity, alongside concerns about safety arising from their potential to produce harmful content. Red teaming aims to find prompts that could elicit harmful responses from LLMs, and is essential to discover and mitigate safety risks before real-world deployment. However, manual red teaming is both time-consuming and expensive, rendering it unscalable. In this paper, we propose RTPE, a scalable evolution framework to evolve red teaming prompts across both breadth and depth dimensions, facilitating the automatic generation of numerous high-quality and diverse red teaming prompts. Specifically, in-breadth evolving employs a novel enhanced in-context learning method to create a multitude of quality prompts, whereas in-depth evolving applies customized transformation operations to enhance both content and form of prompts, thereby increasing diversity. Extensive experiments demonstrate that RTPE surpasses existing representative automatic red teaming methods on both attack success rate and diversity. In addition, based on 4,800 red teaming prompts created by RTPE, we further provide a systematic analysis of 8 representative LLMs across 8 sensitive topics.

cs.CL↗

arXiv2025

Self-supervised Attribute-aware Dynamic Preference Ranking Alignment

Reinforcement Learning from Human Feedback and its variants excel in aligning with human intentions to generate helpful, harmless, and honest responses. However, most of them rely on costly human-annotated pairwise comparisons for supervised alignment, which is not suitable for list-level scenarios, such as community question answering. Additionally, human preferences are influenced by multiple intrinsic factors in responses, leading to decision-making inconsistencies. Therefore, we propose \textbf{Se}lf-supervised \textbf{A}ttribute-aware \textbf{d}ynamic \textbf{p}reference \textbf{ra}nking, called \shortname. \ It quantifies preference differences between responses based on Attribute-Perceptual Distance Factors (APDF) and dynamically determines the list-wise alignment order. Furthermore, it achieves fine-grained preference difference learning and enables precise alignment with the optimal one. We specifically constructed a challenging code preference dataset named StaCoCoQA, and introduced more cost-effective and scalable preference evaluation metrics: PrefHit and PrefRecall. Extensive experimental results show that SeAdpra exhibits superior performance and generalizability on both StaCoCoQA and preference datasets from eight popular domains.

cs.CL↗

arXiv2025

Can Large Language Models Be Query Optimizer for Relational Databases?

Query optimization, which finds the optimized execution plan for a given query, is a complex planning and decision-making problem within the exponentially growing plan space in database management systems (DBMS). Traditional optimizers heavily rely on a certain cost model constructed by various heuristics and empirical tuning, probably leading to generating suboptimal plans. Recent developments of Large Language Models (LLMs) have demonstrated their potential in solving complex planning and decision-making problems, such as arithmetic and programmatic tasks. In this paper, we try to explore the potential of LLMs in handling query optimization and propose a tentative LLM-based query optimizer dubbed LLM-QO, established on PostgreSQL's execution engine. In LLM-QO, we formulate query optimization in an autoregressive fashion which directly generates the execution plan without explicit plan enumeration. To investigate the essential input of LLM-QO, we design a customized data recipe named QInstruct to collect the training data from various optimizers and serialize the database's meta data, queries and corresponding plans into a textual format. Based on QInstruct, we implement a two-stage fine-tuning pipeline, Query Instruction Tuning (QIT) and Query Direct Preference Optimization (QDPO), to empower the capability of general-purpose LLMs in handling query optimization. In our experiments, LLM-QO can generate valid and high-quality plans and consistently outperforms both traditional and learned optimizers on three query workloads. Our findings verify that LLMs can be derived as query optimizers where generalization, efficiency and adaptivity deserve further research efforts.

cs.DB↗

arXiv2025

The Nakai Conjecture for isolated hypersurface singularities of modality $\le 2$

The well-known Nakai Conjecture concerns a very natural question: For an algebra of finite type over a characteristic zero field, if the ring of its differential operators is generated by the first order derivations, is the algebra regular? And it is natural to extend the Nakai Conjecture to local domains, in this paper, we verify it for isolated hypersurface singularities of modality $\le 2$, this extends the existing works.

math.AG↗

arXiv2025

TELEClass: Taxonomy Enrichment and LLM-Enhanced Hierarchical Text Classification with Minimal Supervision

Hierarchical text classification aims to categorize each document into a set of classes in a label taxonomy, which is a fundamental web text mining task with broad applications such as web content analysis and semantic indexing. Most earlier works focus on fully or semi-supervised methods that require a large amount of human annotated data which is costly and time-consuming to acquire. To alleviate human efforts, in this paper, we work on hierarchical text classification with a minimal amount of supervision: using the sole class name of each node as the only supervision. Recently, large language models (LLM) have shown competitive performance on various tasks through zero-shot prompting, but this method performs poorly in the hierarchical setting because it is ineffective to include the large and structured label space in a prompt. On the other hand, previous weakly-supervised hierarchical text classification methods only utilize the raw taxonomy skeleton and ignore the rich information hidden in the text corpus that can serve as additional class-indicative features. To tackle the above challenges, we propose TELEClass, Taxonomy Enrichment and LLM-Enhanced weakly-supervised hierarchical text Classification, which combines the general knowledge of LLMs and task-specific features mined from an unlabeled corpus. TELEClass automatically enriches the raw taxonomy with class-indicative features for better label space understanding and utilizes novel LLM-based data annotation and generation methods specifically tailored for the hierarchical setting. Experiments show that TELEClass can significantly outperform previous baselines while achieving comparable performance to zero-shot prompting of LLMs with drastically less inference cost.

cs.CL↗

arXiv2025

A Unified Knowledge-Distillation and Semi-Supervised Learning Framework to Improve Industrial Ads Delivery Systems

Industrial ads ranking systems conventionally rely on labeled impression data, which leads to challenges such as overfitting, slower incremental gain from model scaling, and biases due to discrepancies between training and serving data. To overcome these issues, we propose a Unified framework for Knowledge-Distillation and Semi-supervised Learning (UKDSL) for ads ranking, empowering the training of models on a significantly larger and more diverse datasets, thereby reducing overfitting and mitigating training-serving data discrepancies. We provide detailed formal analysis and numerical simulations on the inherent miscalibration and prediction bias of multi-stage ranking systems, and show empirical evidence of the proposed framework's capability to mitigate those. Compared to prior work, UKDSL can enable models to learn from a much larger set of unlabeled data, hence, improving the performance while being computationally efficient. Finally, we report the successful deployment of UKDSL in an industrial setting across various ranking models, serving users at multi-billion scale, across various surfaces, geological locations, clients, and optimize for various events, which to the best of our knowledge is the first of its kind in terms of the scale and efficiency at which it operates.

cs.LG↗

arXiv2025

Automation of finding strong gravitational lenses in the Kilo Degree Survey with U-DenseLens (DenseLens + Segmentation)

In the context of upcoming large-scale surveys like Euclid, the necessity for the automation of strong lens detection is essential. While existing machine learning pipelines heavily rely on the classification probability (P), this study intends to address the importance of integrating additional metrics, such as Information Content (IC) and the number of pixels above the segmentation threshold, to alleviate the false positive rate in unbalanced data-sets. In this work, we introduce a segmentation algorithm (U-Net) as a supplementary step in the established strong gravitational lens identification pipeline (Denselens), which primarily utilizes P$_{\rm mean}$ and IC$_{\rm mean}$ parameters for the detection and ranking. The results demonstrate that the inclusion of segmentation enables significant reduction of false positives by approximately 25 per cent in the final sample extracted from DenseLens, without compromising the identification of strong lenses. The main objective of this study is to automate the strong lens detection process by integrating these three metrics. To achieve this, a decision tree-based selection process is introduced, applied to the Kilo Degree Survey (KiDS) data. This process involves rank-ordering based on classification scores, filtering based on Information Content, and segmentation score. Additionally, the study presents 14 newly discovered strong lensing candidates identified by the U-Denselens network using the KiDS DR4 data.

astro-ph.GA↗

arXiv2025

Discovery of 19 strongly-lensed quasars, dual and projected quasars in DESI-LS

We report the follow-up spectroscopic confirmation of two lensed quasars, six dual quasars, and eleven projected quasars that were previously identified as lensed-quasar candidates in \cite{He2023}. The spectroscopic data were obtained from two different sources: the P200/DBSP in California and publicly available datasets, including SDSS and DESI-EDR. The two lensed quasars (both pairs) have the following properties: $θ_E$ = 1.208'', $z_s$ = 3.105; $θ_E$ = 0.749, $z_s$ = 2.395. The six dual quasars have redshifts ranging from 0.58 to 3.28 and projected separations ranging from 15.44 to 22.54 kpc, with a mean separation of 17.95 kpc. The eleven projected quasars have projected separations ranging from 10.96 to 39.07 kpc, with a mean separation of 22.64 kpc. Additionally, there are three likely lensed quasars that cannot be definitively confirmed, contributed by two reasons. Firstly, their image separations (0.83'', 0.98'', and 0.93'') are small compared to the seeing conditions during our observations (around 1.2''). Secondly, no high SNR lensing galaxy can be detected in the Legacy Survey Imaging. Better spectroscopy and (or) imaging are needed to confirm their lensing nature.

astro-ph.GA↗

arXiv2025

Structural and Physical Properties of the Heavy Fermion Metal Ce$_2$NiAl$_6$Si$_5$

Strongly correlated electrons at the verge of quantum criticality give rise to unconventional phases of matter and behaviors, with the discovery of new quantum critical materials driving synergistic advances in both experiments and theory. In this work, we report the structural and physical properties of a new quaternary Ce-based heavy fermion compound, Ce$_2$NiAl$_6$Si$_5$, synthesized using the self-flux method. This compound forms a layered tetragonal structure (space group $P4/nmm$), with square nets of Ce atoms separated by Si-Al or Ni-Si-Ge layers. Specific heat measurements show a low temperature Sommerfeld coefficient of 1.4 J/mol-Ce K$^{2}$, with a reduced entropy indicative of significant Kondo interactions. Below 0.6 K, an upturn in resistivity and a deviation in magnetic susceptibility suggest the appearance of magnetic ordering or the development of dynamic magnetic correlations, which is further supported by a bulge in specific heat around 0.4 K. These results suggest that Ce$_2$NiAl$_6$Si$_5$ is a layered heavy fermion metal, naturally located in proximity to a spin-density-wave quantum critical point.

cond-mat.str-el↗