arXiv subjects

Zixuan Yang

Publications and source records attributed to Zixuan Yang.

At least 19 recordsLinked to original sources

arXiv2026

Not Every Token Is Worth Distilling: Selective Supervision for Direct-OPD

Direct On-Policy Distillation (Direct-OPD) transfers reinforcement-learning-induced policy improvements from a small model to a larger student by using the token-level log-ratio between post-RL and pre-RL checkpoints as dense supervision on the student's own rollouts. This transfer rewards the policy shift at every state, yet the log-ratio measures only relative change: it can stay fixed even as the probability mass that both checkpoints assign to the student's candidate tokens vanishes. Through an exact construction, we show that the Direct-OPD reward and its update can remain unchanged while the Jensen-Shannon divergence (JSD) and both KL directions between the checkpoints vanish with this mass, and we note that a small JSD bounds how much the teacher's behavior changed. Motivated by this analysis, we propose Selective Supervision for Direct-OPD (S$^2$D-OPD), which ranks student-sampled states by their teacher-reference JSD and masks Direct-OPD supervision at low-divergence states, retaining only the top 10% of states per response. Across two teacher pairs and four student models ranging from 1.7B to 8B parameters, S$^2$D-OPD improves held-out accuracy over dense Direct-OPD on AIME and HMMT benchmarks in seven of eight settings and matches it in the eighth, without extra forward passes. Our code is available at https://anonymous.4open.science/r/S2D-OPD-8868.

cs.LG↗

arXiv2026

Data storytelling meets interpretable machine learning: Decoding AI decisions for non-experts without revealing sensitive data and model details

AI-driven automated decision-making requires both predictive performance and interpretability. Recent advances in interpretable machine learning (IML) provide tools for explaining model predictions, but the technical complexity of these explanations may hinder accessibility to non-experts. To address this challenge, this study integrates data storytelling with IML to enhance the explainability of AI-generated decisions for a broader audience. Following the design science research (DSR) paradigm, this study proposes a formal definition of data storytelling in IML, introduces the DIST Pyramid to align data storytelling with IML, and presents the I-P-O Model to describe their interactions. It further develops an architecture to explain AI decisions through distinct "What-if" and "Why-not" event-generation processes. The architecture also employs data desensitization to protect sensitive input data. To validate the approach, a case study is conducted with the Boston Housing dataset, using SHapley Additive exPlanations (SHAP) values and large language models (LLMs) to generate data stories with And-But-Therefore (ABT) structures. An empirical evaluation shows that 76.4% and 74.3% of respondents rated the "What-if" and "Why-not" data stories as more comprehensible, with significantly higher accessibility scores than traditional SHAP visualizations. The paper concludes with the presentation of a narrative interpretation framework that integrates IML and data storytelling, thereby expanding the research scope as well as the practical applicability of AI decision-making.

cs.AI↗

arXiv2026

MERIT: Matching Expertise via Rubric-Informed Training for Reviewer Assignment

Matching submissions with suitable reviewers at scale is a growing challenge for major venues, yet existing approaches either rely on coarse proxy signals that conflate general relatedness with true suitability, or require expensive human annotations that are difficult to scale for training. We propose MERIT, a two-stage framework that bridges this gap by converting criterion-level expertise matching into scalable suitability supervision. In the first stage, we train a reviewer assessor via reinforcement learning to identify the expertise dimensions a paper requires, match them against the reviewer's prior work, and produce a suitability decision, with rewards provided by an LLM judge guided by paper-specific expertise rubrics. In the second stage, we distill the assessor's predictions into an embedding-based retriever for efficient large-scale assignment. Experiments show that our 4B reviewer assessor outperforms larger general-purpose LLMs on suitability classification, and the resulting retriever achieves state-of-the-art performance across LR-Bench and the CMU Gold dataset. Our code is available at https://github.com/Luli3220/MERIT.

cs.CL↗

arXiv2026

Tournament-GRPO: Group-Wise Tournament Rewards for Reinforcement Learning in Open-Ended Long-Form Generation

Reinforcement learning in open-ended long-form generation is challenging because reliable reference answers and automatic metrics are often unavailable. Existing rubric-based methods typically rely on pointwise LLM-as-a-judge scoring, but absolute scores are difficult to calibrate across complex responses, may provide weak discrimination among same-query rollouts, and can become saturated during optimization. We propose Tournament-GRPO, a group-wise reward framework that converts rubric-guided LLM judgments into relative rewards through repeated multi-round tournaments among same-query rollouts. Tournament-GRPO compares candidates within groups, accumulates tournament outcomes, and normalizes them into group-wise rewards for GRPO training. Experiments on Deep Research Bench show that Tournament-GRPO consistently outperforms existing reward-design baselines, achieving a 4.52-point overall-score improvement over the strongest baseline. Further analyses show that tournament rewards provide a favorable effectiveness--efficiency trade-off and that tournament design affects training dynamics. These results suggest that rubric-guided tournament comparison provides an effective reward signal for reinforcement learning in open-ended long-form generation.

cs.CL↗

arXiv2026

Action-Conditioned World Model for Goal Plane Probe Guidance in Robotic Ultrasound

We present an action-conditioned world model framework for goal plane probe guidance in robotic ultrasound, with a focus on neck ultrasound scanning. Autonomous ultrasound tasks often require large numbers of probe-motion trajectories for training, but collecting high-quality demonstrations is labor-intensive and explicit simulators are difficult to build because ultrasound appearance depends on contact, tissue deformation, and view-dependent acoustic artifacts. We address this problem with a two-stage model-based learning pipeline. First, a latent conditional diffusion world model predicts future ultrasound observations from recent context frames, probe motions and temporal offset. Second, a goal-conditioned temporal transformer predicts ordered probe motions and is fine-tuned using rewards from the frozen world model. Experiments on the self-collected dataset show that the world model preserves action-dependent anatomical structure on target-directed scans. In real-world closed loop experiments, the framework achieves success rates of 70.0\% for carotid guidance and 65.0\% for thyroid guidance. These results demonstrate the potential of learned ultrasound dynamics for training goal-directed robotic probe navigation.

cs.RO↗

arXiv2026

On the maximal anti-Ramsey problem of Burr, Erdős, Graham, and Sós for $P_4$

Given a graph $L$, the maximal anti-Ramsey function $\chiS(n,e,L)$ denotes the minimum integer $\chiS$ for which there exists an $n$-vertex graph $G$ with at least $e$ edges admitting an edge-coloring with $\chiS$ colors in which each copy of $L$ in $G$ is rainbow. In 1989, Burr, Erdős, Graham, and Sós posed the following problem: Is it true that for all $ε>0$, there exists $c(ε)>0$ such that for all sufficiently large $n$, $ \chiS\left(n,\binom{n}{2}-\lfloor n^{2-ε}\rfloor,P_4\right)>c(ε)n^2. $ Very recently, Li, Ning, and Xie gave a negative answer to the problem for all $0< ε< 1/2$. In this note, we establish that a quadratic lower bound holds in the complementary regime $ ε\geq 1/2$. More specifically, we prove that for all $ε\ge 1/2$ and sufficiently large $n$, there is an absolute constant $c>0$ such that $ \chiS\left(n,\binom{n}{2}-\lfloor n^{2-ε}\rfloor,P_4\right)>c n^2. $

math.CO↗

arXiv2026

On the generalized Turán number of the complete bipartite graph $K_{3,b+1}$

For graphs $F$ and $H$, let $\mathrm{ex}(n,H,F)$ denote the maximum number of copies of $H$ in an $n$-vertex $F$-free graph. Very recently, Janzer, Longbrake, and Yepremyan proved that for $3<a\leq b$ and sufficiently large $t$, \begin{equation*} \mathrm{ex}(n,K_{a,b},K_{3,t})=Θ_{a,b,t}(n^3). \end{equation*} Later, Hou, Hu, and Wang made this threshold explicit by showing that the conclusion holds for all $t\geq 2\max\{3,\lceil b/2\rceil\}+1$. In particular, for every even $b\geq 6$, this matches the necessary threshold $t=b+1$. In this paper, we resolve the remaining case where $b$ is odd. More precisely, we prove that for all fixed integers $b\geq 5$ and $3<a\leq b$, \begin{equation*} \mathrm{ex}(n,K_{a,b},K_{3,b+1})=Θ_{a,b}(n^3). \end{equation*} Our construction uses a finite-field point set in $\mathrm{PG}(5,q)$ together with an orthogonal polarity. The key new ingredient is the polynomial splitting lemma due to Andrade, Bary-Soroker, and Rudnick, which produces many planes whose intersections with the point set and their polar planes both have size $b$. This gives a $K_{3,b+1}$-free incidence graph while preserving $Ω_{a,b}(n^3)$ copies of $K_{a,b}$.

math.CO↗

arXiv2026

Simply Stabilizing the Loop via Fully Looped Transformer

Scaling model performance typically requires increasing model size. Looped Transformer offers a compelling alternative by iteratively reusing the same Transformer blocks, trading additional computation for improved performance without increasing parameter count or context length. Because the number of loop iterations can be adjusted at inference, it also provides a natural mechanism for balancing performance and test-time compute. However, Looped Transformer still suffers from training instability when the number of loop iterations increases. Our analysis reveals that this instability stems from two sources: gradient oscillation and residual explosion. To address these two problems, we propose the Fully Looped Transformer, which introduces two parameter-free modifications: (1) Fully Looped Architecture, which distributes inter-loop signals across all layers to mitigate residual explosion; (2) Attention Injection, which reuses the existing attention block to suppress gradient oscillation. These modifications stabilize training dynamics, enabling the Fully Looped Transformer to be trained stably up to 12 loop iterations, whereas other baseline looped models collapse in this regime. In milder settings where Looped Transformer does not collapse, Fully Looped Transformer still improves average downstream-task performance by up to 13.2\%. Overall, our experiments demonstrate that Fully Looped Transformer improves training stability, enhances downstream performance, and provides preliminary adaptability under different test-time compute budgets by varying loop iterations at inference.

cs.LG↗

arXiv2026

Most probably trangle-free graphs

The celebrated Mantel's theorem states that any triangle-free graph on $n$ vertices contains at most $\left\lfloor n^2/4\right\rfloor$ edges. It is natural to ask how many triangles must exist in a graph with more than $\left\lfloor n^2/4\right\rfloor$ edges--a problem known as the Erdős-Rademacher problem. In this paper, we propose a probabilistic variant of this classic problem. Specifically, given an $n$-vertex graph $G$ with $\left\lfloor n^2/4\right\rfloor+i$ ($i>0$) edges, we choose the edges of $G$ independently with probability $p$, and the resulting new graph is triangle-free with a certain probability. Our goal is to maximize this probability by choosing $G$ appropriately. For the case where $G$ has $ \left\lfloor n^2/4\right\rfloor +1$ edges, we determine the exact maximum probability.

math.CO↗

arXiv2026

JADE: Bridging the Strategic-Operational Gap in Dynamic Agentic RAG

The evolution of Retrieval-Augmented Generation (RAG) has shifted from static retrieval pipelines to dynamic, agentic workflows where a central planner orchestrates multi-turn reasoning. However, existing paradigms face a critical dichotomy: they either optimize modules jointly within rigid, fixed-graph architectures, or empower dynamic planning while treating executors as frozen, black-box tools. We identify that this \textit{decoupled optimization} creates a ``strategic-operational mismatch,'' where sophisticated planning strategies fail to materialize due to unadapted local executors, often leading to negative performance gains despite increased system complexity. In this paper, we propose \textbf{JADE} (\textbf{J}oint \textbf{A}gentic \textbf{D}ynamic \textbf{E}xecution), a unified framework for the joint optimization of planning and execution within dynamic, multi-turn workflows. By modeling the system as a cooperative multi-agent team unified under a single shared backbone, JADE enables end-to-end learning driven by outcome-based rewards. This approach facilitates \textit{co-adaptation}: the planner learns to operate within the capability boundaries of the executors, while the executors evolve to align with high-level strategic intent. Empirical results demonstrate that JADE transforms disjoint modules into a synergistic system, yielding remarkable performance improvements via joint optimization and enabling a flexible balance between efficiency and effectiveness through dynamic workflow orchestration.

cs.AI↗

arXiv2026

RATE: Reviewer Profiling and Annotation-free Training for Expertise Ranking in Peer Review Systems

Reviewer assignment is increasingly critical yet challenging in the LLM era, where rapid topic shifts render many pre-2023 benchmarks outdated and where proxy signals poorly reflect true reviewer familiarity. We address this evaluation bottleneck by introducing LR-bench, a high-fidelity, up-to-date benchmark curated from 2024-2025 AI/NLP manuscripts with five-level self-assessed familiarity ratings collected via a large-scale email survey, yielding 1055 expert-annotated paper-reviewer-score annotations. We further propose RATE, a reviewer-centric ranking framework that distills each reviewer's recent publications into compact keyword-based profiles and fine-tunes an embedding model with weak preference supervision constructed from heuristic retrieval signals, enabling matching each manuscript against a reviewer profile directly. Across LR-bench and the CMU gold-standard dataset, our approach consistently achieves state-of-the-art performance, outperforming strong embedding baselines by a clear margin. We release LR-bench at https://huggingface.co/datasets/Gnociew/LR-bench, and a GitHub repository at https://github.com/Gnociew/RATE-Reviewer-Assign.

cs.CL↗

arXiv2026

The Turán number of Berge matchings

Given a graph $F$, an $r$-uniform hypergraph $\mathcal{H}$ is a {\em Berge-$F$} if there is a bijection $ϕ:E(F)\to E(\mathcal{H})$ such that $e\subseteq ϕ(e)$ for each $e\in E(F)$. Given a family $\mathcal{F}$ of $r$-uniform hypergraphs, an $r$-uniform hypergraph is $\mathcal{F}$-free if it does not contain any member of $\mathcal{F}$ as a subhypergraph. The Turán number of $\mathcal{F}$ is the maximum number of hyperedges in an $\mathcal{F}$-free $r$-graph on $n$ vertices. Let $M_{s+1}$ denote a matching of size $s+1$, i.e., the graph consisting of $s+1$ independent edges. Khormali and Palmer [\textit{European J. Combin.} 102 (2022) 103506] completely determined the Turán number of Berge matchings for sufficiently large $n$. Subsequently, Kang, Ni, and Shan [\textit{Discrete Math.} 345 (2022) 112901] determined the exact value of the Turán number of Berge-$M_{s+1}$ for all $n$ when $r \le s-1$ or $r \ge 2s+2$. In this paper, we settle the final open case $s \le r \le 2s+1$, thereby completing the determination of the Turán number of Berge matchings.

math.CO↗

arXiv2026

Beyond Monolithic Architectures: A Multi-Agent Search and Knowledge Optimization Framework for Agentic Search

Agentic search has emerged as a promising paradigm for complex information seeking by enabling Large Language Models (LLMs) to interleave reasoning with tool use. However, prevailing systems rely on monolithic agents that suffer from structural bottlenecks, including unconstrained reasoning outputs that inflate trajectories, sparse outcome-level rewards that complicate credit assignment, and stochastic search noise that destabilizes learning. To address these challenges, we propose \textbf{M-ASK} (Multi-Agent Search and Knowledge), a framework that explicitly decouples agentic search into two complementary roles: Search Behavior Agents, which plan and execute search actions, and Knowledge Management Agents, which aggregate, filter, and maintain a compact internal context. This decomposition allows each agent to focus on a well-defined subtask and reduces interference between search and context construction. Furthermore, to enable stable coordination, M-ASK employs turn-level rewards to provide granular supervision for both search decisions and knowledge updates. Experiments on multi-hop QA benchmarks demonstrate that M-ASK outperforms strong baselines, achieving not only superior answer accuracy but also significantly more stable training dynamics.\footnote{The source code for M-ASK is available at https://github.com/chenyiqun/M-ASK.}

cs.AI↗

arXiv2025

Rainbow matchings in edge-colored graphs

Let $G$ be an edge-colored graph. We use $e(G)$ and $c(G)$ to denote the number of edges and colors in $G$, respectively. A subgraph $H$ is called rainbow if $c(H)=e(H)$. Li et al. (European J. Combin., 36 (2014), 453-459) proved that every edge-colored graph on $n$ vertices with $e(G)+c(G) \geq n(n+1)/2$ contains rainbow triangles. Later, Xu et al. (European J. Combin., 54 (2016), 193-200) generalized the previous results concerning rainbow triangles to rainbow cliques $Kr$, where $r\geq 4$. In this paper, we consider the existence of rainbow matchings of size $k$ in general edge-colored graphs $G$ under the condition of $e(G)+c(G)$, and the condition in our result is tight.

math.CO↗

arXiv2025

Scaling Laws for Online Advertisement Retrieval

The scaling law is a notable property of neural network models and has significantly propelled the development of large language models. Scaling laws hold great promise in guiding model design and resource allocation. Recent research increasingly shows that scaling laws are not limited to NLP tasks or Transformer architectures; they also apply to domains such as recommendation. However, there is still a lack of literature on scaling law research in online advertisement retrieval systems. This may be because 1) identifying the scaling law for resource cost and online revenue is often expensive in both time and training resources for industrial applications, and 2) varying settings for different systems prevent the scaling law from being applied across various scenarios. To address these issues, we propose a lightweight paradigm to identify online scaling laws of retrieval models, incorporating a novel offline metric and an offline simulation algorithm. We prove that under mild assumptions, the correlation between the novel metric and online revenue asymptotically approaches 1 and empirically validates its effectiveness. The simulation algorithm can estimate the machine cost offline. Based on the lightweight paradigm, we can identify online scaling laws for retrieval models almost exclusively through offline experiments, and quickly estimate machine costs and revenues for given model configurations. We further validate the existence of scaling laws across mainstream model architectures (e.g., Transformer, MLP, and DSSM) in our real-world advertising system. With the identified scaling laws, we demonstrate practical applications for ROI-constrained model designing and multi-scenario resource allocation in the online advertising system. To the best of our knowledge, this is the first work to study identification and application of online scaling laws for online advertisement retrieval.

cs.IR↗

arXiv2025

On saturation problems for matchings with regularity constraints

A graph $G$ is $F$-saturated if $G$ is $F$-free but for any edge $e$ in the complement of $G$ the graph $G + e$ contains $F$. Gerbner et al. (Discrete Math., 345 (2022), 112921) initiated the study of $rsat(n,F)$, the minimum number of edges in a regular $n$-vertex $F$-saturated graph, and they posed the problem of for which graphs $rsat(n, F )$ exists. Regarding this problem, we obtain the precise value of $rsat(n,(m+1)K_2)$ for all possible cases, where $(m+1)K_2$ denotes a matching of size $m+1$. As a natural counterpart, we also determine the maximum number of edges in a regular $n$-vertex $(m+1)K_2$-free graph for all $m\ge 1$ and $n\ge 2m+2$.

math.CO↗

arXiv2025

Learning Cascade Ranking as One Network

Cascade Ranking is a prevalent architecture in large-scale top-k selection systems like recommendation and advertising platforms. Traditional training methods focus on single-stage optimization, neglecting interactions between stages. Recent advances have introduced interaction-aware training paradigms, but still struggle to 1) align training objectives with the goal of the entire cascade ranking (i.e., end-to-end recall of ground-truth items) and 2) learn effective collaboration patterns for different stages. To address these challenges, we propose LCRON, which introduces a novel surrogate loss function derived from the lower bound probability that ground truth items are selected by cascade ranking, ensuring alignment with the overall objective of the system. According to the properties of the derived bound, we further design an auxiliary loss for each stage to drive the reduction of this bound, leading to a more robust and effective top-k selection. LCRON enables end-to-end training of the entire cascade ranking system as a unified network. Experimental results demonstrate that LCRON achieves significant improvement over existing methods on public benchmarks and industrial applications, addressing key limitations in cascade ranking training and significantly enhancing system performance.

cs.IR↗

arXiv2025

Exploring Human-Like Thinking in Search Simulations with Large Language Models

Simulating user search behavior is a critical task in information retrieval, which can be employed for user behavior modeling, data augmentation, and system evaluation. Recent advancements in large language models (LLMs) have opened up new possibilities for generating human-like actions including querying, browsing, and clicking. In this work, we explore the integration of human-like thinking into search simulations by leveraging LLMs to simulate users' hidden cognitive processes. Specifically, given a search task and context, we prompt LLMs to first think like a human before executing the corresponding action. As existing search datasets do not include users' thought processes, we conducted a user study to collect a new dataset enriched with users' explicit thinking. We investigate the impact of incorporating such human-like thinking on simulation performance and apply supervised fine-tuning (SFT) to teach LLMs to emulate both human thinking and actions. Our experiments span two dimensions in leveraging LLMs for user simulation: (1) with or without explicit thinking, and (2) with or without fine-tuning on the thinking-augmented dataset. The results demonstrate the feasibility and potential of incorporating human-like thinking in user simulations, though performance improvements on some metrics remain modest. We believe this exploration provides new avenues and inspirations for advancing user behavior modeling in search simulations.

cs.IR↗