Searcharxiv⌕ Search

arXiv subjects

Zhichao Xu

Publications and source records attributed to Zhichao Xu.

At least 19 recordsLinked to original sources

Multi-Granularity Context-Enhanced RAG over Multimodal Knowledge Graphs

Knowledge graph (KG)-based retrieval-augmented generation (RAG) leverages structured knowledge to provide large language models (LLMs) with high-quality external information. Building on these works, recent studies have explored multimodal knowledge graphs (MMKGs) as knowledge bases for GraphRAG. This enables GraphRAG to integrate knowledge across multiple modalities, thereby further enhancing its performance. However, existing MMKG-based RAG methods generally follow a common pipeline in which different modalities are largely processed independently before being fused. As a result, textual context is only used to a limited extent during visual information extraction and subsequent multimodal knowledge fusion. This brings a semantic gap between images and text which limits the multimodal GraphRAG performance. To address this issue, we propose a novel framework for constructing a Context-Enhanced MMKG (CEMMKG) to better support multimodal GraphRAG. The proposed CEMMKG enriches each image with complementary textual context at both local and global scopes. Local context goes beyond the surrounding text by incorporating sentences that are semantically related to the image, while global context provides a summary of the entire document. We further introduce a multi-granularity design for the local context, allowing it to capture semantically relevant information at different levels of detail. Extensive experiments on multiple datasets validate that CEMMKG is effective in leveraging contextual information to improve MMKG-based RAG performance. Moreover, its effectiveness across different MMKG-based RAG methods demonstrates its broad applicability.

cs.AI↗

ViRA: Toward Ultra-High-Resolution Visual Agents via Cross-Input Relative Policy Optimization

Visual agents answer questions about ultra-high-resolution photos by zooming into the relevant regions. However, existing agents often answer from common sense instead of reading the evidence. Reinforcement learning (RL) on the final answer cannot correct this, because a guess earns the same reward as reading. We propose Cross-Input Relative Policy Optimization (CrossRPO), which trains agents on contrast sets. A contrast set asks one question about several versions of an image that differ only in a small region with the evidence and have different answers, so only reading the evidence solves the whole set. CrossRPO rewards solving the whole set, and we prove that it does not reinforce a set answered from common sense. We build contrast sets for training (ViRA-Train) and evaluation (ViRA-Bench) by editing one small object in images of 21 to 120 megapixels, and we photograph further sets with real changes (ViRA-Bench-Real) to exclude editing traces. We train two Visual Reasoning Agents, ViRA-9B and ViRA-27B, with CrossRPO on only about 3,500 items of ViRA-Train each. CrossRPO achieves 63.3% set accuracy on ViRA-Bench at the maximum resolution, against 59.3% for GRPO. ViRA-9B is the most accurate open model of its size on all benchmarks, and ViRA-27B outperforms Claude Haiku 5.5 and GPT-6-Luna, both at their highest thinking level, on most benchmarks.

cs.CV↗

Tevatron Meets Megatron: Expert-Parallel LLM Reranker Training on an Academic Budget

Training large dense and mixture-of-experts (MoE) rerankers is constrained by GPU memory and distributed-execution efficiency, which has kept academic reranker research largely on dense models. Tevatron trains rerankers through the Hugging Face Trainer with DeepSpeed and PyTorch FSDP1, a stack that stalls on three fronts: DeepSpeed ZeRO-2 has a known gradient-gathering failure on this path, FSDP1 full sharding is memory-safe but throughput-bound, and neither offers the expert parallelism (EP) that MoE training needs. We present Tevatron 3.0, which integrates Megatron-Core as a reranker training backend while preserving Tevatron's data processing, evaluation pipeline, and Hugging Face checkpoint format. For full-parameter dense 8B training, Megatron matches FSDP1 on BEIR-15 NDCG@10 with approximately 4% lower epoch time and twice the peak GPU memory under matched data-parallel and recomputation settings, and approximately 22% lower epoch time in its recommended configuration; it supports both LoRA and full-parameter fine-tuning. Expert parallelism makes it possible to train a 30B-parameter MoE reranker (Qwen3-30B-A3B) on two nodes. A controlled study of architecture, LoRA, and distillation on BEIR-15, with three first-stage retrievers and Hugging Face and vLLM serving, shows that the MoE reranker matches dense-8B quality at less than half the activated parameters and at higher inference throughput. We open-source the framework and trained checkpoints.

cs.IR↗

Tevatron-Elastic: A Unified Abstraction for Training Elastic Retrievers and Rerankers

A single model scale challenges the flexibility of a production retrieval system: some settings need it faster, others need a smaller index, and the right trade-off changes with the workload. In the context of information retrieval (IR), a transformer-based model can be made smaller in three ways --- using fewer layers, passing fewer tokens through the upper layers, or producing a shorter embedding --- and each way saves a different resource. These options have been studied in isolation, each as its own method with its own code and training setup, which makes them hard to combine or adapt to a new model. In this work, we present Tevatron-Elastic to bring all three under one unified abstraction: a single object names any size the model can run at, and a short schedule lists the sizes to train. Training then produces one checkpoint that serves all of those sizes, and at deployment the user has the liberty to pick any of them. The same abstraction covers both retrievers and rerankers and both encoder and decoder models, as it works through interfaces that Hugging Face transformers already expose; adding a new backbone can be a configuration change rather than a model rewrite. Prior methods --- Matryoshka embeddings, early exit, 2D Matryoshka (Starbucks), and layerwise token compression (LTC) --- become special cases of our unified abstraction. With Tevatron-Elastic, we also propose a new elastic algorithm Matryoshka~LTC (MLTC) to jointly train several token-compression ratios in one retriever checkpoint. To validate our framework, we train 20 checkpoints across three backbones and two tasks: the quality curves are smooth, full-point quality remains close to that of models trained for a single size, and a controlled study confirms the wall-clock speedups. We release the framework and all checkpoints to support elastic retrieval research.

cs.CL↗

Defragmenting Language Models: An Interpretability-based Approach for Vocabulary Expansion

All languages are equal; when it comes to tokenization, some are more equal than others. Tokens are the hidden currency that dictates the cost and latency of contemporary LLM access. However, many languages written in non-Latin scripts observe a poor exchange rate: LLMs take several multiples of tokens to encode the same information in many languages as they do for English. We observe that this issue, known as token overfragmentation, persists in modern open-weight LLMs. The standard remedy is vocabulary expansion that adds target language items missing from the model's vocabulary. We focus on two core decisions in the vocabulary expansion process: What items should we add? and How should we initialize their corresponding input and output embeddings? We comprehensively study and advance the new research direction of interpretability-based vocabulary expansion along these two core decisions, making four key contributions: First, we question the conventional use of frequency-based methods to choose candidate vocabulary items to add (a decision long treated as settled), and show that interpretability-based methods offer a superior performance-token efficiency tradeoff. Second, we strengthen the case for interpretability-based embedding initialization by showing large gains (~20 pts) over baseline initialization methods for several languages written in non-Latin scripts. Third, we identify the phenomenon of "subword detokenization," where models progressively merge fragmented subword tokens into larger subwords across layers. Fourth, we propose FragMend, a method that leverages this phenomenon to push the efficiency ceiling of interpretability-based expansion, and validate its effectiveness.

cs.CL↗

CLEAR: Context Augmentation from Contrastive Learning of Experience via Agentic Reflection

Large language model agents rely on effective model context to obtain task-relevant information for decision-making. Many existing context engineering approaches primarily rely on the context generated from the past experience and retrieval mechanisms that reuse these context. However, retrieved context from past tasks must be adapted by the execution agent to fit new situations, placing additional reasoning burden on the underlying LLM. To address this limitation, we propose a generative context augmentation framework using Contrastive Learning of Experience via Agentic Reflection (CLEAR). CLEAR first employs a reflection agent to perform contrastive analysis over past execution trajectories and summarize useful context for each observed task. These summaries are then used as supervised fine-tuning data to train a context augmentation model (CAM). Then we further optimize CAM using reinforcement learning, where the reward signal is obtained by running the task execution agent. By learning to generate task-specific knowledge rather than retrieve knowledge from the past, CAM produces context that is better tailored to the current task. We conduct comprehensive evaluations on the AppWorld and WebShop benchmarks. Experimental results show that CLEAR consistently outperforms strong baselines. It improves task completion rate from 72.62% to 81.15% on AppWorld test set and averaged reward from 0.68 to 0.74 on a subset of WebShop, compared with baseline agent. Our code is publicly available at https://github.com/awslabs/CLEAR.

cs.AI↗

DualOPSD: Closing the Feedback Loop in On-Policy Self-Distillation

On-policy self-distillation (OPSD) combines student generated trajectories with dense supervision from a privileged teacher of the same model. Its pointwise clipping protects the student from locally incompatible teacher directions, but the fixed teacher receives no feedback about the supervision the student rejects, which limits effective learning from the teacher. We introduce DualOPSD, which alternates updates in a coupled teacher-student system. The student retains OPSD's clipped forward KL objective. The privileged teacher adapts to the updated student after each student step. A residual target based on the frozen initial model preserves the effect of privileged context during teacher adaptation. Across Qwen3-8B/4B/1.7B and three math benchmarks, DualOPSD improves avg@12 accuracy over OPSD by 9.83 percentage points. Further diagnostic analyses show lower bidirectional KL, less persistent mismatch, and retained sensitivity to privileged reference. Together, these results demonstrate that DualOPSD can be a more effective form of privileged self-distillation. Code is available at: https://github.com/on1262/DualOPSD.

cs.LG↗

Explore with Memory, Learn without It: In-Context Exploration and Self-Distillation for LLM Reasoning

Exploration bounds learning efficiency in reinforcement learning with verifiable rewards (RLVR). Current methods explore mostly through independent sampling, without fully using the information gathered along the way, such as known wrong answers and environment feedback. We propose LEAF, which injects a dynamic, problem-specific memory into the context before every rollout. The memory contains the most recent response, its reward, and other environment feedback. It is updated after every verifier call. We decouple policy updates from rollouts and update the policy through self-distillation. On four benchmarks and two open models, LEAF accelerates exploration by about 4x compared with a random sampling baseline. On unseen test sets, it improves accuracy over the strongest baseline by 1.59 and 1.83 percentage points on the two models, while using less compute than any RL baseline. These results highlight the potential of LEAF in improving both exploration and learning efficiency. Code is available at: https://github.com/on1262/LEAF

cs.LG↗

APEX: Speculate smarter, not deeper

Speculative decoding reduces large language model inference latency by drafting multiple tokens before target-model verification, but its effectiveness depends on both the proposal mechanism and draft depth. Fixed configurations cannot respond to changes in predictability, repetition, and acceptance during generation, so deeper drafting can increase wasted computation without proportional speedup. We introduce APEX, a learned controller that balances decoding speed and draft-token waste through request-level expert selection and block-level depth adaptation. APEX-Router selects among EAGLE-3, n-gram, and draft-model speculation for each request, while APEX-Depth adjusts draft length at each verification block using causal decoding signals and recent verifier feedback. APEX models accepted draft length as censored survival feedback, learning position-wise rejection hazards, block execution costs, and an action utility that balances throughput, accepted progress, and wasted tokens. This allows the controller to adapt speculation while retaining the target model's verification procedure. We integrate APEX into vLLM and evaluate it with Qwen3-8B across six workloads, achieving up to 5.24X speedup over autoregressive decoding. Across the aggregate evaluation, APEX-S achieves 4.27X speedup, while APEX-B achieves 3.27X speedup with a 41.0% relative reduction in wasted-token percentage compared with fixed n-gram speculation at k=16, providing distinct operating points for balancing acceleration and draft-token utilization.

cs.CL↗

Rethinking Self-Distillation for Multi-Teacher Capability Merging

Combining capabilities of multiple expert models trained starting from the same base checkpoint has become increasingly common in frontier language-model post-training. Recent trends suggest that multi-teacher on-policy distillation (MOPD) outperforms conventional off-policy methods. However, despite the higher inference and environment interaction costs incurred by MOPD, we find that much of its reported accuracy gain is due to certain training design choices and hyperparameter optimization, rather than the algorithm itself. We conduct a controlled self-distillation study across two multi-teacher settings, four models, and eleven benchmarks, comparing off-policy methods, namely supervised fine-tuning (SFT) and soft-label distillation, to hybrid teacher-prefix distillation and MOPD. We found that all four methods achieve \textit{nearly identical} accuracy. However, MOPD uses $14.8$--$23.1\times$ SFT's training GPU-hours. We also revisit four recently published comparisons between on-policy and off-policy distillation and find that the reported on-policy gains shrink substantially when SFT baselines are trained on rejection-sampled teacher trajectories and use independently tuned hyperparameters. As a training-free alternative, we also find that simple weight-merging methods can recover expert capabilities with minutes of CPU merging time, although their accuracy degrades as model size decreases and task interference increases. Overall, our results question recent gains reported due to MOPD and suggest careful tuning of more efficient off-policy baselines as a viable alternative.

cs.LG↗

An Empirical Study of Automating Agent Evaluation

Agent evaluation requires assessing complex multi-step behaviors involving tool use and intermediate reasoning, making it costly and expertise-intensive. A natural question arises: can frontier coding assistants reliably automate this evaluation process? Our study shows that simply prompting coding assistants is insufficient for this task. Without domain-specific evaluation knowledge, frontier coding assistants achieve only a 30% execution success rate and produce over-engineered evaluations averaging 12+ metrics per agent, indicating that strong coding ability does not automatically translate to reliable agent evaluation. We introduce EvalAgent, an AI assistant that automates the end-to-end agent evaluation pipeline. EvalAgent encodes evaluation domain expertise as evaluation skills (procedural instructions, reusable code and templates, and dynamically retrieved API documentation) that compose into a trace-based pipeline producing complete evaluation artifacts including metrics, executable code, and reports. To systematically assess generated evaluations, we introduce a meta-evaluation framework alongside AgentEvalBench, a benchmark comprising 20 agents, each paired with evaluation requirements and test scenarios. We further propose the Eval@1 metric to measure whether generated evaluation code both executes and yields meaningful results on the first run. Our experiments show that EvalAgent produces focused evaluations, improving Eval@1 from 17.5% to 65%, and achieving 79.5% human expert preference over baseline approaches. Further ablation studies show that evaluation skills are critical for handling complex evaluation: removing them causes Eval@1 to drop significantly from 65% to 30%.

cs.CL↗

VerTox: Verifiable Reward-Guided Corpus Poisoning Against Neural Ranking Models

Neural ranking models have become core components of modern information retrieval systems and important building blocks of AI systems such as retrieval-augmented generation (RAG) pipelines. However, their robustness remains insufficiently understood in the presence of large language models (LLMs), which can generate fluent and deceptive content at scale. This work investigates the vulnerability of neural ranking models to corpus poisoning attacks, in which an adversary injects a small number of maliciously crafted documents into the corpus to distort ranking behavior. We propose VerTox, the first framework to formulate corpus poisoning as a verifiable reward-guided reinforcement learning (RLVR) problem. By explicitly coupling ranking distortion with factual corruption through specialized reward shaping, we fine-tune compact LLMs into adversarial generators. Experiments demonstrate that our method achieves near-perfect attack success rates, producing adversarial documents that frequently rank higher than target documents across major neural ranking architectures, as well as a proprietary commercial embedding model. The generated adversarial documents are fluent and exhibit low perplexity, making them difficult to detect. Furthermore, by explicitly encouraging factual corruption, our adversarial documents significantly degrade the performance of a downstream RAG application.

cs.CL↗

LACONIC: Dense-Level Effectiveness for Scalable Sparse Retrieval via a Two-Phase Training Curriculum

While dense retrieval models have been the standard for state-of-the-art information retrieval, their deployment is often constrained by high memory requirements and reliance on GPU accelerators for vector similarity search at scale. Learned sparse retrieval offers a compelling alternative by enabling efficient search via inverted indices, yet it has historically received less attention than dense approaches. In this paper, we introduce LACONIC, a family of learned sparse retrievers based on the Llama3 architecture (1B, 3B, and 8B). We propose a streamlined two-phase training curriculum consisting of (1) weakly supervised pre-finetuning to adapt causal LLMs for bidirectional contextualization and (2) high-signal finetuning using curated hard negatives. Our results demonstrate that LACONIC effectively bridges the performance gap with dense models: the 8B variant achieves a state-of-the-art 60.2 nDCG@10 on the MTEB Retrieval benchmark, ranking 15th on the leaderboard as of February 5th, 2026, while utilizing 74% less index memory than an equivalent dense model. By delivering high retrieval effectiveness on commodity CPU hardware with a fraction of the compute budget required by competing models, LACONIC provides a scalable and efficient solution for real-world search applications. We fully open source our code implementation and trained checkpoints to facilitate reproducibility.

cs.IR↗

PERSONAJUDGE: Simulating Individual Human Preference Judgments with Evaluator-Specific Demonstration Data

Large language models increasingly serve as judges in AI evaluation, but current approaches rely on consensus preferences that ignore individual evaluator variation. We propose a novel simulation approach that combines categorical judgments with evaluator-specific auxiliary data--retrospective reasoning traces and interface telemetry--to enable LLM-based simulation of individual evaluators via in-context learning. We conduct a systematic empirical study of this approach using multi-facet data from 32 trained annotators across 4,200 preference judgments in a 4 x 4 x 4 factorial design. Our key findings: (1) The simulation approach achieves up to 9.9 percentage point improvements over the Base Judge; (2) Reasoning traces provide the largest gains with higher collection efforts, while interface telemetry often hurts rather than helps performance despite being cheaper to collect. (3) Simulation difficulty is systematic, predicted by an evaluator's neutral usage (most clearly on Helpfulness) and divergence from consensus; the neutral-usage tendency--rather than simulatability itself--is the cross-task-stable property (r = 0.728). These results establish both the potential and limits of evaluator-specific auxiliary data for personalized evaluation, offering methodological insights for scaling individual aware AI assessment.

cs.HC↗

Beyond Supervised Clarification: Input Rewriting with LLMs for Dialogue Discourse Parsing

Rewriting inputs to improve frozen downstream models has become a common strategy in modern NLP pipelines. Prior work on incremental dialogue discourse parsing (DDP) shows that supervised clarification models can rewrite fragmentary or underspecified utterances, such as resolving ellipsis or references, to improve parsing accuracy. In this work, we revisit this idea under realistic deployment conditions, where no clarification supervision is available and the clarifier must rely on zero-shot prompting or feedback from a frozen parser. Across three Segmented Discourse Representation Theory (SDRT) datasets and multiple parsers, we find that last-utterance clarification is far less reliable than suggested by supervised settings. Parser-agnostic rewriting often introduces more regressions than repairs, as edits that enable fixes also disrupt discourse cues relied upon by the parser. A best-of-8 rewriting analysis further reveals a practical ceiling: a large fraction of errors are not repairable through input rewriting alone. A parser-aware clarifier trained with GRPO reduces regressions by up to 37% by learning conservative abstention, yet still fails to produce selectivity-aware clarifications that consistently improve parsing. Together, these findings recast clarification as a selective intervention problem. We identify rewritability prediction, deciding whether an utterance is repairable before intervention, as the key missing capability for input-side optimization of frozen discourse parsers, and a critical direction for improving agentic pipelines more broadly.

cs.CL↗

Rethinking On-policy Optimization for Query Augmentation

Recent advances in large language models (LLMs) have led to a surge of interest in query augmentation for information retrieval (IR). Two main approaches have emerged. The first prompts LLMs to generate answers or pseudo-documents that serve as new queries, relying purely on the model's parametric knowledge or contextual information. The second applies reinforcement learning (RL) to fine-tune LLMs for query rewriting, directly optimizing retrieval metrics. While having respective advantages and limitations, the two approaches have not been compared under consistent experimental conditions. In this work, we present the first systematic comparison of prompting-based and RL-based query augmentation across diverse benchmarks, including evidence-seeking, ad hoc, and tool retrieval. Our key finding is that under a compute-aware comparison setting, simple, training-free query augmentation often performs on par with, or even surpasses, more expensive RL-based counterparts, especially when using powerful LLMs. Motivated by this discovery, we introduce a novel hybrid method, On-policy Pseudo-document Query Expansion (OPQE), in which the LLM policy learns to generate a pseudo-document that maximizes retrieval performance, rather than rewriting the query, thus merging the flexibility and generative structure of prompting with the targeted optimization of RL. We show OPQE outperforms both standalone prompting and RL-based rewriting, demonstrating that a synergistic approach yields the best results. We open source our implementation to facilitate reproducibility.

cs.CL↗

Improving Long-Context Retrieval with Multi-Prefix Embedding

Long-context retrieval exposes a tension: single-vector embeddings lose fine-grained detail, while token-level multi-vector methods incur prohibitive storage. We propose Multi-Prefix Embedding (MPE), which partitions a document into chunks separated by EOS tokens, encodes the full sequence in a single causal forward pass, and extracts one embedding at each prefix boundary. MPE retains cross-chunk context, enables chunk-level MaxSim matching, and trains with only document-level relevance labels. Experiments on MLDR-en, BrowseComp-Plus, and LongEmbed show that MPE is competitive with or outperforms single-vector, independent-chunk, and multi-vector baselines, while providing a natural source attribution mechanism for locating evidence chunks.

cs.IR↗

A Sober Look at Agentic Misalignment in Automated Workflows

We study a class of emergent misalignment in multi-agent systems (MAS), with a focus on automated workflows, which we refer to agentic misalignment. Although these systems can solve complex tasks, they often fail because agents act according to implicit proxy utilities that do not align with the intended human goals. We formally define these behaviors and analyze them within a Bayesian framework, showing that generic utilities naturally lead to posterior collapse of agents in automated workflows. To address this issue, we propose Agentic Evidence Attribution (AEA), a novel alignment paradigm that improves agent posteriors using context-specific evidence. AEA reasons over agent actions and provides structured evidence to correct misaligned behavior during collaboration. To better understand the role of evidence, we study two instantiations of AEA: self-reflection (internal evidence from the model) and weak-to-strong generalization (external evidence on the agentic trajectory). We show that a small evidence model effectively aligns the MAS by providing orthogonal failure attribution. Our results clarify the sources of agentic misalignment in automated workflows and show that evidence-based alignment can effectively improve agent collaboration and leads to reliable multi-agent systems built on automated workflows.

cs.AI↗