arXiv subjects

Aixin Sun

Publications and source records attributed to Aixin Sun.

At least 19 recordsLinked to original sources

arXiv2026

Defining Page-Centric Visual Document Retrieval and Evaluating Existing Benchmarks

In modern information retrieval (IR), documents are increasingly encountered as structured sequences of visually rich, paginated units, rather than as flat text. Example documents encountered in our daily life include Word files, PDFs, and slide decks. Free text-centric retrieval systems either ignore this structure entirely or destroy it through arbitrary passage chunking. In this paper, we formulate page-centric Visual Document Retrieval (VDR) with pages as the fundamental indexing units and document context available for assessing page relevance. The formulation distinguishes two retrieval granularities, page and document, and three retrieval paradigms: independent-item, evidence-set, and iterative retrieval. We then assess 42 benchmarks using eight criteria covering corpus characteristics, query construction, relevance annotation, and ranking support. Our analysis shows two main findings on current VDR research. First, evaluation remains concentrated on page-level independent-item retrieval, with fewer resources supporting document-level, evidence-set, and iterative settings. Second, queries originating from real user information needs independently of target corpus content and graded relevance judgments are uncommon, while the coverage and verification of relevance judgments vary substantially. Both findings reveal gaps between the capabilities applications require and those existing evaluation protocols can establish. We provide benchmark selection guidance and identify priorities for future benchmark design, emphasizing alignment between task requirements, relevance semantics, and evaluation protocols.

cs.IR↗

arXiv2026

ProVe: Proposer-Verifier Decoding

Speculative decoding accelerates autoregressive (AR) language models by letting a small model draft tokens serially and a large model verify them in parallel. We reverse these roles. Starting from a standard AR checkpoint, we fine-tune the large model into a proposer that predicts a block of future tokens in a single forward pass, then use one additional forward pass to verify the proposed block. This gives ProVe, Proposer-Verifier decoding. The verifier can be a smaller AR model or the proposer itself; the latter gives ProVe-lossless, which reproduces the proposer's greedy AR decoding. Block proposals already provide substantial room for acceleration: with a block size of 16, the proposed block has an average AR-matching prefix of 6.6 tokens on GSM8K. Across six benchmarks with a 32B proposer, ProVe-lossless reaches 2.5x average AR throughput without training a separate draft head. With a small 3B verifier, ProVe reaches 2.7x average throughput at 2.6 accuracy points below AR on the six-task average, providing a controllable tradeoff between speed and accuracy.

cs.CL↗

arXiv2026

Alignment by Stereotyping: How LLMs Sacrifice Individual Distinctiveness for Cultural Adaptation

Large language models are increasingly deployed for personalized interaction, and demographic conditioning via user profiles is a widely adopted strategy for cultural adaptation. We ask whether this approach genuinely serves individual users or achieves accuracy by erasing individual distinctiveness. Studying seven models including frontier GPT-5.1 on the World Values Survey, we find that demographic profiles improve value alignment accuracy for most models, but at a systematic cost to individuality. That is, models pull responses toward demographic group centroids rather than preserving individual differences, a behavioral pattern we term alignment by stereotyping. Permutation tests (10,000 permutations, six demographic attributes, seven models) certify that top-performing models compress individuals far above the human baseline; within-family scaling amplifies this tradeoff while degrading intrinsic cultural understanding. Using a synthetic dialogue dataset validated on real human-chatbot conversations from PRISM (Kirk et al., 2024), we further show that distributing demographic signals across conversational turns partially suppresses prototype retrieval compared to compact demographic labels, a finding validated on real conversations via PRISM but requiring replication at larger scale.

cs.CL↗

arXiv2026

EMemBench: Interactive Benchmarking of Episodic Memory for VLM Agents

We introduce EMemBench, a programmatic benchmark generator for evaluating long-term episodic memory of agents through interactive games. Rather than using a fixed set of questions, EMemBench generates questions from environment-grounded trajectories, covering both text-only and visual game environments. Each template computes verifiable ground truth from underlying game signals, with controlled answerability and balanced coverage over memory skills: single/multi-hop recall, induction, temporal, spatial, logical, and adversarial. We evaluate memory agents with strong LMs/VLMs as backbones, using in-context prompting as baselines. Across 15 text games and multiple visual seeds, results are far from saturated: induction and spatial reasoning are persistent bottlenecks, especially in visual settings. Persistent memory yields clear gains for open backbones on text games, but improvements are less consistent for VLM agents, suggesting that visually grounded episodic memory remains an open challenge. A human study further contextualizes the difficulty and interpretability of EMemBench.

cs.CL↗

arXiv2026

Skill-as-Pseudocode: Refactoring Skill Libraries to Pseudocode for LLM Agents

Markdown skill libraries for LLM agents ship as free-form prose, forcing the agent to re-derive both the input schema and the concrete invocation syntax on every retrieval. This produces a "confused $\to$ re-retrieve $\to$ still confused" loop: the agent issues a partially-correct action, receives uninformative feedback, and re-retrieves the same prose. We propose Skill-as-Pseudocode (SaP), an automatic conversion of markdown skill libraries into typed pseudocode with deterministic quality control. From each cluster of similar procedural passages, SaP extracts a typed contract and filters it through a four-check deterministic verifier (coverage, binding, replacement, risk). Promoted contracts are inlined into a rewritten skill skeleton alongside restored action templates, giving the agent two complementary signals: a typed signature for what a skill does and a concrete template for how to invoke it. On the ALFWorld unseen split (134 games, gpt-4o-mini, three seeds), SaP wins 82/402 paired games versus 47/402 for the Graph-of-Skills (GoS) baseline (pooled McNemar $p = 8.2 \times 10^{-5}$), at $-22.8 \pm 6.4$% input tokens and $-14.5 \pm 4.1$% LLM calls per game. A bundle-component ablation attributes the gain to the pairing of typed contracts with concrete action templates: the contract alone falls below the prose baseline.

cs.PL↗

arXiv2026

A Position Paper on Recommender Systems in the Era of Autonomous Agents

For decades, recommender systems have been optimized to serve human users. However, the rapid deployment of autonomous agents introduces a paradigm shift: recommendation consumers are predicted to be increasingly a mixture of humans and authorized agents acting on their behalf. This position paper reviews insights from prior human-centric RecSys research and outlines the transition to this hybrid environment. In our discussion, we treat agents as independent, user-aligned assistants that act as end-users of recommender systems, rather than platform-built components. We characterize the resulting tripartite interactions among humans, agents, and platforms, highlighting the dynamics that can arise across these relationships. This shift presents new opportunities for RecSys research, while introducing unique challenges for evaluation, alignment, and system design.

cs.IR↗

arXiv2026

Revisiting Parameter-Based Knowledge Editing in Large Language Models: Theoretical Limits and Empirical Evidence

Parameter-based knowledge editing updates the internal knowledge of large language models (LLMs) via localized weight modifications and has attracted significant attention. However, most existing methods overlook fundamental theoretical limitations and are rarely evaluated under realistic, practice-oriented settings. In this paper, we first present a theoretical analysis based on the dimensional Collapse Hypothesis, explaining how localized parameter edits can propagate along fragile directions in the representation space, inducing global interference and ultimately causing reasoning collapse. Building on this insight, we conduct a comprehensive empirical evaluation by systematically varying knowledge complexity, number of edits, evaluation dimensions, and baseline methods. Our results show that parameter-based editing methods consistently damage core LLM capabilities. In contrast, a simple retrieval-based baseline achieves consistently stronger performance than all parameter-editing methods across all evaluated conditions. These findings highlight that preserving the fundamental capabilities of LLMs after knowledge editing should be a central concern for future research.

cs.CL↗

arXiv2026

RecNextEval: A Reference Implementation for Temporal Next-Batch Recommendation Evaluation

A good number of toolkits have been developed in Recommender Systems (RecSys) research to promote fair evaluation and reproducibility. However, recent critical examinations of RecSys evaluation protocols have raised concerns regarding the validity of existing evaluation pipelines. In this demonstration, we present RecNextEval, a reference implementation of an evaluation framework specifically designed for next-batch recommendation. RecNextEval utilizes a time-window data split to ensure models are evaluated along a global timeline, effectively minimizing data leakage. Our implementation highlights the inherent complexities of RecSys evaluation and encourages a shift toward model development that more accurately simulates production environments. The RecNextEval library and its accompanying GUI interface are open-source and publicly accessible.

cs.IR↗

arXiv2026

If an LLM Were a Character, Would It Know Its Own Story? Evaluating Lifelong Learning in LLMs

Large language models (LLMs) can carry out human-like dialogue, but unlike humans, they are stateless due to the superposition property. However, during multi-turn, multi-agent interactions, LLMs begin to exhibit consistent, character-like behaviors, hinting at a form of emergent lifelong learning. Despite this, existing benchmarks often fail to capture these dynamics, primarily focusing on static, open-ended evaluations. To address this gap, we introduce LIFESTATE-BENCH, a benchmark designed to assess lifelong learning in LLMs. It features two episodic datasets: Hamlet and a synthetic script collection, rich in narrative structure and character interactions. Our fact checking evaluation probes models' self-awareness, episodic memory retrieval, and relationship tracking, across both parametric and non-parametric approaches. Experiments on models like Llama3.1-8B, GPT-4-turbo, and DeepSeek R1, we demonstrate that nonparametric methods significantly outperform parametric ones in managing stateful learning. However, all models exhibit challenges with catastrophic forgetting as interactions extend, highlighting the need for further advancements in lifelong learning.

cs.CL↗

arXiv2026

MARS: Enabling Autoregressive Models Multi-Token Generation

Autoregressive (AR) language models generate text one token at a time, even when consecutive tokens are highly predictable given earlier context. We introduce MARS (Mask AutoRegreSsion), a lightweight fine-tuning method that teaches an instruction-tuned AR model to predict multiple tokens per forward pass. MARS adds no architectural modifications, no extra parameters, and produces a single model that can still be called exactly like the original AR model with no performance degradation. Unlike speculative decoding, which maintains a separate draft model alongside the target, or multi-head approaches such as Medusa, which attach additional prediction heads, MARS requires only continued training on existing instruction data. When generating one token per forward pass, MARS matches or exceeds the AR baseline on six standard benchmarks. When allowed to accept multiple tokens per step, it maintains baseline-level accuracy while achieving 1.5-1.7x throughput. We further develop a block-level KV caching strategy for batch inference, achieving up to 1.71x wall-clock speedup over AR with KV cache on Qwen2.5-7B. Finally, MARS supports real-time speed adjustment via confidence thresholding: under high request load, the serving system can increase throughput on the fly without swapping models or restarting, providing a practical latency-quality knob for deployment.

cs.CL↗

arXiv2026

Hesitation and Tolerance in Recommender Systems

Users' interactions with recommender systems often involve more than simple acceptance or rejection. We highlight two overlooked states: hesitation, when people deliberate without certainty, and tolerance, when this hesitation escalates into unwanted engagement before ending in disinterest. Across two large-scale surveys (N=6,644 and N=3,864), hesitation was nearly universal, and tolerance emerged as a recurring source of wasted time, frustration, and diminished trust. Analyses of e-commerce and short-video platforms confirm that tolerance behaviors, such as clicking without purchase or shallow viewing, correlate with decreased activity. Finally, an online field study at scale shows that even lightweight strategies treating tolerance as distinct from interest can improve retention while reducing wasted effort. By surfacing hesitation and tolerance as consequential states, this work reframes how recommender systems should interpret feedback, moving beyond clicks and dwell time toward designs that respect user value, reduce hidden costs, and sustain engagement.

cs.IR↗

arXiv2026

Can LLM Safety Be Ensured by Constraining Parameter Regions?

Large language models (LLMs) are often assumed to contain ``safety regions'' -- parameter subsets whose modification directly influences safety behaviors. We conduct a systematic evaluation of four safety region identification methods spanning different parameter granularities, from individual weights to entire Transformer layers, across four families of backbone LLMs with varying sizes. Using ten safety identification datasets, we find that the identified safety regions exhibit only low to moderate overlap, as measured by IoU. The overlap drops significantly when the safety regions are further refined using utility datasets (\ie non-harmful queries). These results suggest that current techniques fail to reliably identify a stable, dataset-agnostic safety region.

cs.LG↗

arXiv2026

OneTrans: Unified Feature Interaction and Sequence Modeling with One Transformer in Industrial Recommender

In recommendation systems, scaling up feature-interaction modules (e.g., Wukong, RankMixer) or user-behavior sequence modules (e.g., LONGER) has achieved notable success. However, these efforts typically proceed on separate tracks, which not only hinders bidirectional information exchange but also prevents unified optimization and scaling. In this paper, we propose OneTrans, a unified Transformer backbone that simultaneously performs user-behavior sequence modeling and feature interaction. OneTrans employs a unified tokenizer to convert both sequential and non-sequential attributes into a single token sequence. The stacked OneTrans blocks share parameters across similar sequential tokens while assigning token-specific parameters to non-sequential tokens. Through causal attention and cross-request KV caching, OneTrans enables precomputation and caching of intermediate representations, significantly reducing computational costs during both training and inference. Experimental results on industrial-scale datasets demonstrate that OneTrans scales efficiently with increasing parameters, consistently outperforms strong baselines, and yields a 5.68% lift in per-user GMV in online A/B tests.

cs.IR↗

arXiv2026

FLM-Audio: Natural Monologues Improves Native Full-Duplex Chatbots via Dual Training

Full-duplex dialog models aim to listen and speak simultaneously, delivering rapid responses to dynamic user input. Among different solutions to full-duplexity, a native solution merges multiple channels in each time step, achieving the lowest latency. However, prevailing designs break down the textual monologue sentences for word-level alignment with audio streams, which degrades language modeling abilities. To help address this issue, we introduce "contiguous monologues", which are composed by continuous sentences and "waiting" intervals, mimicking human-like cognitive behavior in dialogs. We find a proper training paradigm to be critical for semantically aligning contiguous monologues with audio. To this end, we develop a "dual" training paradigm that alternates the position of the monologues, either leading or trailing the audio, across different training stages. A combination of our contiguous monologue and dual training strategy is applied in developing FLM-Audio, our 7B spoken dialog chatbot with native full-duplexity. As confirmed by experimental results, FLM-Audio achieves superior response qualities and chatting experiences while requiring significantly less training data.

cs.SD↗

arXiv2026

EgoMem: Lifelong Memory Agent for Full-duplex Omnimodal Models

We introduce EgoMem, the first lifelong memory agent tailored for full-duplex models that process real-time omnimodal streams. EgoMem enables real-time models to recognize multiple users directly from raw audiovisual streams, to provide personalized response, and to maintain long-term knowledge of users' facts, preferences, and social relationships extracted from audiovisual history. EgoMem operates with three asynchronous processes: (i) a retrieval process that dynamically identifies user via face and voice, and gathers relevant context from a long-term memory; (ii) an omnimodal dialog process that generates personalized audio responses based on the retrieved context; and (iii) a memory management process that automatically detects dialog boundaries from omnimodal streams, and extracts necessary information to update the long-term memory. Unlike existing memory agents for LLMs, EgoMem relies entirely on raw audiovisual streams, making it especially suitable for lifelong, real-time, and embodied scenarios. Experimental results demonstrate that EgoMem's retrieval and memory management modules achieve over 95% accuracy on the test set. When integrated with a fine-tuned RoboEgo omnimodal chatbot, the system achieves fact-consistency scores above 87% in real-time personalized dialogs, establishing a strong baseline for future research.

cs.AI↗

arXiv2026

Demystifying the Slash Pattern in Attention: The Role of RoPE

Large Language Models (LLMs) often exhibit slash attention patterns, where attention scores concentrate along the $Δ$-th sub-diagonal for some offset $Δ$. These patterns play a key role in passing information across tokens. But why do they emerge? In this paper, we demystify the emergence of these Slash-Dominant Heads (SDHs) from both empirical and theoretical perspectives. First, by analyzing open-source LLMs, we find that SDHs are intrinsic to models and generalize to out-of-distribution prompts. To explain the intrinsic emergence, we analyze the queries, keys, and Rotary Position Embedding (RoPE), which jointly determine attention scores. Our empirical analysis reveals two characteristic conditions of SDHs: (1) Queries and keys are almost rank-one, and (2) RoPE is dominated by medium- and high-frequency components. Under these conditions, queries and keys are nearly identical across tokens, and interactions between medium- and high-frequency components of RoPE give rise to SDHs. Beyond empirical evidence, we theoretically show that these conditions are sufficient to ensure the emergence of SDHs by formalizing them as our modeling assumptions. Particularly, we analyze the training dynamics of a shallow Transformer equipped with RoPE under these conditions, and prove that models trained via gradient descent exhibit SDHs. The SDHs generalize to out-of-distribution prompts.

cs.LG↗

arXiv2025

Error Analyses of Auto-Regressive Video Diffusion Models: A Unified Framework

Auto-Regressive Video Diffusion Models (AR-VDMs) have shown strong capabilities in generating long, photorealistic videos, but suffer from two key limitations: (i) history forgetting, where the model loses track of previously generated content, and (ii) temporal degradation, where frame quality deteriorates over time. Yet a rigorous theoretical analysis of these phenomena is lacking, and existing empirical understanding remains insufficiently grounded. In this paper, we introduce Meta-ARVDM, a unified analytical framework that studies both errors through the shared autoregressive structure of AR-VDMs. We show that history forgetting is characterized by the conditional mutual information between the generated output and preceding frames, conditioned on inputs, and prove that incorporating more past frames monotonically alleviates history forgetting, thereby theoretically justifying a common belief in existing works. Moreover, our theory reveals that standard metrics fail to capture this effect, motivating a new evaluation protocol based on a ``needle-in-a-haystack'' task in closed-ended environments (DMLab and Minecraft). We further show that temporal degradation can be quantified by the cumulative sum of per-step errors, enabling prediction of degradation for different schedulers without video rollout. Finally, our evaluation uncovers a strong empirical correlation between history forgetting and temporal degradation, a connection not previously reported.

cs.CV↗

arXiv2025

On the Role of Discreteness in Diffusion LLMs

Diffusion models offer appealing properties for language generation, such as parallel decoding and iterative refinement, but the discrete and highly structured nature of text challenges the direct application of diffusion principles. In this paper, we revisit diffusion language modeling from the view of diffusion process and language modeling, and outline five properties that separate diffusion mechanics from language-specific requirements. We first categorize existing approaches into continuous diffusion in embedding space and discrete diffusion over tokens. We then show that each satisfies only part of the five essential properties and therefore reflects a structural trade-off. Through analyses of recent large diffusion language models, we identify two central issues: (i) uniform corruption does not respect how information is distributed across positions, and (ii) token-wise marginal training cannot capture multi-token dependencies during parallel decoding. These observations motivate diffusion processes that align more closely with the structure of text, and encourage future work toward more coherent diffusion language models.

cs.CL↗