Searcharxiv⌕ Search

arXiv subjects

Fan Xia

Publications and source records attributed to Fan Xia.

At least 19 recordsLinked to original sources
arXiv2026

An Event is Worth One Token: Event Tokenization for Industrial-scale LLM Recommendation

LLM-based recommendation has scaled along model capacity and sequence length, yet each position encodes only text, semantic IDs, or a few categorical features, discarding rich user, item, context, and outcome signals available at each event. Under autoregressive modeling, this yields weak queries at each position and, since each position becomes context for the next, the degradation compounds across the sequence. We propose an event-centric paradigm that represents each interaction by its full temporal snapshot, and identify a new scaling dimension we term snapshot resolution: the amount of information encoded per event. To efficiently scale snapshot resolution, we introduce AMBER (Autoregressive Modeling via Bottlenecked Event Representation), which compresses each temporal snapshot into a compact Event Token, a new LLM input modality. The representation is learned end-to-end, while Event Tokens are pre-computed and cached for serving, decoupling snapshot resolution from real-time serving compute. On industrial-scale ranking and retrieval benchmarks, AMBER advances the compute-quality Pareto frontier relative to alternative recommendation paradigms. At sufficient capacity, a single unified tokenizer even outperforms dedicated per-entity tokenizers, demonstrating positive transfer across structurally different entity types. AMBER's Event Tokens also transfer across model architectures: when integrated into a heavily optimized non-LLM ranker as serving-time historical features, they yield statistically significant improvements. Further scaling Event Tokenizer capacity provides additional improvements.

cs.IR↗
arXiv2026

Expert-Guided g-computation with Large Language Models for Estimating Causal Effects on Timings: Applications to Hospital Quality Improvement

Hospital quality improvement (QI) programs routinely face multiple candidate interventions to optimize hospital flow, but existing methods struggle to estimate and rank the causal effects of such interventions. This work focuses on one of the most standard hospital metrics, the average length of stay (LOS), and its causal estimand, the average time saved. To characterize this causal effect, qualitative approaches rely on expert judgment to map patient trajectories, making them susceptible to cognitive biases; quantitative approaches rely on data-driven models, which fail when interventions are hypothetical with no historical data or have complex causal mechanisms that require clinical reasoning rather than data alone. We propose expert-guided g-computation, or egg-computation, which combines the complementary strengths of both approaches by connecting the Gantt charts commonly used to map patient trajectories with the causal DAG literature. We introduce a causal model over Gantt charts and establish identification using a variant of g-computation that seeks expert input only for components unidentifiable from data. To make egg-computation practical, we develop an LLM-assisted pipeline that reliably scales up expert reasoning. In simulations, egg-computation outperforms conventional causal inference methods when patients have diverse causal structures and intervention mechanisms. In a study of eleven candidate QI interventions at an urban safety-net hospital, the LLM pipeline generated graphs and time-saving estimates highly concordant with those of human experts. Beyond healthcare, egg-computation is a broadly applicable framework for estimating the average time saved for candidate interventions whose causal mechanisms can be represented using Gantt charts.

stat.ME↗
arXiv2026

ERAlign: Energy-based Representation Alignment of GNNs and LLMs on Text-attributed Graphs

Text-attributed Graphs (TAGs) incorporate textual node attributes with graph structures to describe rich relational semantics. Recent efforts to integrate Graph Neural Networks (GNNs) and Large Language Models (LLMs) have shown promise for learning on TAGs, yet achieving well-aligned representations remains challenging. Prior studies largely rely on heuristics that perform coarse-grained matching. They lack sufficient constraints and ignore distributional alignment, leading to representation drift and limited generalization. Building on Energy-based Models (EBMs), we propose an Energy-based Representation Alignment (ERAlign) framework that projects GNN-encoded graph structure and LLM-derived text embeddings in a shared latent space to achieve distribution consistency. Concretely, layer-wise alignment is quantified by a distance metric and optimized via an EBM objective. By decreasing energy values, our framework yields well-aligned representations for downstream tasks. During training, we introduce Energy Discrepancy (ED) to avoid high sampling costs associated with intractable normalization. ED also carries theoretical guarantees of higher training efficiency and reduced energy landscape distortion. Empirical evaluations on eight TAG datasets demonstrate that ERAlign obtains state-of-the-art performance across varying levels of supervision and cross-task transfer scenarios.

cs.LG↗
arXiv2026

iDiff: Interpretable Difference-aware Framework for Pairwise Image Quality Assessment

Pairwise image quality assessment (IQA) in professional photography requires a model not only to identify the preferred image between two candidates, but also to provide convincing and image-grounded reasoning. In the NTIRE 2026 RAIM challenge, this requirement is further emphasized by jointly evaluating preference prediction and rationale generation. To address this task, we propose iDiff, an Interpretable Difference-aware framework for pairwise image quality assessment. Our method adopts a dual-branch design consisting of an Answer Model and a Thinking Model. The Answer Model performs robust preference prediction by explicitly decomposing each sample into left/right global and local views, followed by content-aware specialization for person and scene images and ensemble-based aggregation across backbones. The Thinking Model focuses on rationale generation and is progressively enhanced with expert-style templates, multi-source quality features, and answer-aware supervision conditioned on the Answer Model prediction. In this way, iDiff jointly models discriminative decision making and structured explanation, improving both robustness and interpretability. Extensive experiments demonstrate the effectiveness of the proposed framework on both accuracy and reasoning-quality metrics. Our method achieved first place in the NTIRE 2026 RAIM challenge, showing the effectiveness of integrating explicit difference modeling with structured multimodal reasoning for pairwise IQA.

cs.CV↗
arXiv2026

TTE-Flash: Accelerating Reasoning-based Multimodal Representations via Think-Then-Embed Tokens

Recent research has demonstrated that Universal Multimodal Embedding (UME) benefits significantly from Chain-of-Thought (CoT) reasoning. In this paradigm, a generative model produces explicit reasoning traces for a multimodal query, with the final representation extracted from an embedding token attending to both the query and the reasoning. Despite its effectiveness, the computational overhead of generating explicit CoT traces is often prohibitive. In this work, we propose replacing explicit CoT with latent think tokens, which are interpreted as latent variables that can produce explicit CoT traces as observed variables. By optimizing think tokens using CoT generation loss and subsequent embedding tokens using contrastive loss, we produce high-performance, reasoning-aware representations at a constant inference cost. Our study investigates two key architectural designs: 1) how think and embeddings tokens should be extracted from the same LLM backbone. 2) how the tokens should be trained as two dependent tasks. We introduce TTE-Flash-2B, a reasoning-aware multimodal representation model that outperforms its explicit-CoT counterpart on the MMEB-v2 benchmark, while producing latent think tokens that are interpretable both textually and visually. Furthermore, zero-shot evaluation across 15 video datasets reveals scaling behavior as the number of think tokens increases, and motivating a pilot study of adaptive think budget allocation based on task requirements.

cs.AI↗
arXiv2026

Causal Discovery via Statistical Power (CDSP)

Causal discovery methods aim to infer causal direction from observational data. Functional causal discovery approaches use structural asymmetries to identify causal directionality but rely on strong modeling assumptions and provide limited tools for uncertainty quantification. We introduce Causal Discovery via Statistical Power (CDSP), a statistical inference framework that connects causal direction estimation with statistical power and enables uncertainty quantification. Considering the foundational setting of bivariate observational data, we show how quantities analogous to statistical power and effect size can be used in causal discovery to determine when data contain sufficient information to favor one direction over the other. We introduce the effect-size asymmetry assumption that characterizes when the probability of correctly detecting the causal direction (i.e., the power of causal discovery) exceeds that of incorrectly favoring the reverse direction. We show that the effect-size asymmetry assumption can be used for causal direction estimation with uncertainty quantification. Simulations show that CDSP direction estimation is robust to mild and moderate model misspecifications. Real data analyses on 100 cause-effect benchmark pairs further demonstrate that CDSP reduces false discovery rates by approximately 18% relative to a commonly used existing method.

stat.ME↗
arXiv2025

Virtual Width Networks

We introduce Virtual Width Networks (VWN), a framework that delivers the benefits of wider representations without incurring the quadratic cost of increasing the hidden size. VWN decouples representational width from backbone width, expanding the embedding space while keeping backbone compute nearly constant. In our large-scale experiment, an 8-times expansion accelerates optimization by over 2 times for next-token and 3 times for next-2-token prediction. The advantage amplifies over training as both the loss gap grows and the convergence-speedup ratio increases, showing that VWN is not only token-efficient but also increasingly effective with scale. Moreover, we identify an approximately log-linear scaling relation between virtual width and loss reduction, offering an initial empirical basis and motivation for exploring virtual-width scaling as a new dimension of large-model efficiency.

cs.LG↗
arXiv2025

Physics-Gated Visual Prediction of MARFE on the HL-3 Tokamak

The Multifaceted Asymmetric Radiation From the Edge (MARFE) is a critical plasma instability that often precedes density-limit disruptions in tokamaks, posing a significant risk to machine integrity and operational efficiency. We develop a physics-gated, continuous MARFE monitor for the HL-3 tokamak that outputs a per-frame intensity probability every $2$\,ms, which can potentially be used by the shape-target controller in the plasma control system. Our framework integrates two core innovations: (1) a physics-scored, weighted Expectation-Maximization (EM) pipeline that refines noisy visual labels using $(n_e, T_e, f_G, t)$ as a Bayesian prior, and (2) a continuous-time, physics-gated Neural Ordinary Differential Equation (Neural ODE) backbone whose dynamics are modulated by a sigmoid gate on $f_G$ and $T_e$. Meanwhile, the Neural ODE adopts a $40$\,ms forward forecasting horizon to accommodate the actuator-response budget. On a frozen $140$-shot held-out test set, the proposed method yields a median label-aligned lead time of $+36$\,ms, close to this design horizon. Against a Bi-LSTM baseline trained under the matched protocol, the proposed Neural ODE attains Area Under the Curve (AUC) $=0.981$ and sample-level $F_1=0.840$, compared with AUC $=0.960$ and sample-level $F_1=0.779$ for the baseline. The deployed inference service runs within a $1$-ms control-cycle budget, while new diagnostic samples are generated at the $2$-ms frame cadence.

physics.plasm-ph↗
arXiv2025

Parallel Loop Transformer for Efficient Test-Time Computation Scaling

Large Language Models (LLMs) are powerful but often too slow and costly for real-world use during inference. Looped transformers save on parameters by reusing the same weights for multiple computational steps, or "loops." However, this approach has a major flaw: the loops run one after another, causing inference latency and memory requirements to increase with each added loop. This makes them impractical for fast applications. To solve this problem, we introduce the Parallel Loop Transformer (PLT). PLT is a new architecture that delivers the performance benefits of a deep, looped model but with the low latency of a standard, non-looped model. PLT works using two key techniques. First, Cross-Loop Parallelism (CLP) breaks the sequential dependency by computing different loops for different tokens at the same time, all within a single pass. Second, to prevent memory costs from growing, we use an Efficient Representation Enhancement strategy. This method shares the memory (KV cache) from the first loop with all other loops. It then uses a Gated Sliding-Window Attention (G-SWA) to combine this shared global information with local information, maintaining high accuracy. Our experiments show that PLT achieves the high accuracy of a traditional looped model but with almost no extra latency or memory cost compared to a standard transformer.

cs.CL↗
arXiv2025

iDETEX: Empowering MLLMs for Intelligent DETailed EXplainable IQA

Image Quality Assessment (IQA) has progressed from scalar quality prediction to more interpretable, human-aligned evaluation paradigms. In this work, we address the emerging challenge of detailed and explainable IQA by proposing iDETEX-a unified multimodal large language model (MLLM) capable of simultaneously performing three key tasks: quality grounding, perception, and description. To facilitate efficient and generalizable training across these heterogeneous subtasks, we design a suite of task-specific offline augmentation modules and a data mixing strategy. These are further complemented by online enhancement strategies to fully exploit multi-sourced supervision. We validate our approach on the large-scale ViDA-UGC benchmark, where iDETEX achieves state-of-the-art performance across all subtasks. Our model ranks first in the ICCV MIPI 2025 Detailed Image Quality Assessment Challenge, demonstrating its effectiveness and robustness in delivering accurate and interpretable quality assessments.

cs.CV↗
arXiv2025

Robust and Efficient Semiparametric Inference for the Stepped Wedge Design

Stepped wedge designs (SWDs) are increasingly used to evaluate longitudinal cluster-level interventions but pose substantial challenges for valid inference. Because crossover times are randomized, intervention effects are intrinsically confounded with secular time trends, while heterogeneity across clusters, complex correlation structures, baseline covariate imbalances, and small numbers of clusters further complicate inference. We propose a unified semiparametric framework for estimating possibly time-varying intervention effects in SWDs. Under a semiparametric model on treatment contrast, we develop a nonstandard semiparametric efficiency theory that accommodates correlated observations within clusters, varying cluster-period sizes, and weakly dependent treatment assignments. The resulting estimator is consistent and asymptotically normal even under misspecified covariance structure and control cluster-period means, and is efficient when both are correctly specified. To enable inference with few clusters, we exploit the permutation structure of treatment assignment to propose a standard error estimator that reflects finite-sample variability, with a leave-one-out correction to reduce plug-in bias. The framework also allows incorporation of effect modification and adjustment for imbalanced precision variables through design-based adjustment or double adjustment that additionally incorporates an outcome-based component. Simulations and application to a public health trial demonstrate the robustness and efficiency of the proposed method relative to standard approaches.

stat.ME↗
arXiv2025

Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inference

We present Seed Diffusion Preview, a large-scale language model based on discrete-state diffusion, offering remarkably fast inference speed. Thanks to non-sequential, parallel generation, discrete diffusion models provide a notable speedup to mitigate the inherent latency of token-by-token decoding, as demonstrated recently (e.g., Mercury Coder, Gemini Diffusion). Seed Diffusion Preview achieves an inference speed of 2,146 token/s over H20 GPUs while maintaining competitive performance across a sweep of standard code evaluation benchmarks, significantly faster than contemporary Mercury and Gemini Diffusion, establishing new state of the art on the speed-quality Pareto frontier for code models.

cs.CL↗
arXiv2025

"Who experiences large model decay and why?" A Hierarchical Framework for Diagnosing Heterogeneous Performance Drift

Machine learning (ML) models frequently experience performance degradation when deployed in new contexts. Such degradation is rarely uniform: some subgroups may suffer large performance decay while others may not. Understanding where and how large differences in performance arise is critical for designing targeted corrective actions that mitigate decay for the most affected subgroups while minimizing any unintended effects. Current approaches do not provide such detailed insight, as they either (i) explain how average performance shifts arise or (ii) identify adversely affected subgroups without insight into how this occurred. To this end, we introduce a Subgroup-scanning Hierarchical Inference Framework for performance drifT (SHIFT). SHIFT first asks "Is there any subgroup with unacceptably large performance decay due to covariate/outcome shifts?" (Where?) and, if so, dives deeper to ask "Can we explain this using more detailed variable(subset)-specific shifts?" (How?). In real-world experiments, we find that SHIFT identifies interpretable subgroups affected by performance decay, and suggests targeted actions that effectively mitigate the decay.

cs.LG↗
arXiv2025

LLMs Judging LLMs: A Simplex Perspective

Given the challenge of automatically evaluating free-form outputs from large language models (LLMs), an increasingly common solution is to use LLMs themselves as the judging mechanism, without any gold-standard scores. Implicitly, this practice accounts for only sampling variability (aleatoric uncertainty) and ignores uncertainty about judge quality (epistemic uncertainty). While this is justified if judges are perfectly accurate, it is unclear when such an approach is theoretically valid and practically robust. We study these questions for the task of ranking LLM candidates from a novel geometric perspective: for $M$-level scoring systems, both LLM judges and candidates can be represented as points on an $(M-1)$-dimensional probability simplex, where geometric concepts (e.g., triangle areas) correspond to key ranking concepts. This perspective yields intuitive theoretical conditions and visual proofs for when rankings are identifiable; for instance, we provide a formal basis for the ``folk wisdom'' that LLM judges are more effective for two-level scoring ($M=2$) than multi-level scoring ($M>2$). Leveraging the simplex, we design geometric Bayesian priors that encode epistemic uncertainty about judge quality and vary the priors to conduct sensitivity analyses. Experiments on LLM benchmarks show that rankings based solely on LLM judges are robust in many but not all datasets, underscoring both their widespread success and the need for caution. Our Bayesian method achieves substantially higher coverage rates than existing procedures, highlighting the importance of modeling epistemic uncertainty.

cs.LG↗
arXiv2025

Instruction-augmented Multimodal Alignment for Image-Text and Element Matching

With the rapid advancement of text-to-image (T2I) generation models, assessing the semantic alignment between generated images and text descriptions has become a significant research challenge. Current methods, including those based on Visual Question Answering (VQA), still struggle with fine-grained assessments and precise quantification of image-text alignment. This paper presents an improved evaluation method named Instruction-augmented Multimodal Alignment for Image-Text and Element Matching (iMatch), which evaluates image-text semantic alignment by fine-tuning multimodal large language models. We introduce four innovative augmentation strategies: First, the QAlign strategy creates a precise probabilistic mapping to convert discrete scores from multimodal large language models into continuous matching scores. Second, a validation set augmentation strategy uses pseudo-labels from model predictions to expand training data, boosting the model's generalization performance. Third, an element augmentation strategy integrates element category labels to refine the model's understanding of image-text matching. Fourth, an image augmentation strategy employs techniques like random lighting to increase the model's robustness. Additionally, we propose prompt type augmentation and score perturbation strategies to further enhance the accuracy of element assessments. Our experimental results show that the iMatch method significantly surpasses existing methods, confirming its effectiveness and practical value. Furthermore, our iMatch won first place in the CVPR NTIRE 2025 Text to Image Generation Model Quality Assessment - Track 1 Image-Text Alignment.

cs.CV↗
arXiv2025

Factors affecting power in stepped wedge trials when the treatment effect varies with time

Stepped wedge cluster randomized trials (SW-CRTs) have historically been analyzed using immediate treatment (IT) models, which assume the effect of the treatment is immediate after treatment initiation and subsequently remains constant over time. However, recent research has shown that this assumption can lead to severely misleading results if treatment effects vary with exposure time, i.e. time since the intervention started. Models that account for time-varying treatment effects, such as the exposure time indicator (ETI) model, allow researchers to target estimands such as the time-averaged treatment effect (TATE) over an interval of exposure time, or the point treatment effect (PTE) representing a treatment contrast at one time point. However, this increased flexibility results in reduced power. In this paper, we use public power calculation software and simulation to characterize factors affecting SW-CRT power. Key elements include choice of estimand, study design considerations, and analysis model selection. or common SW-CRT designs, the sample size (clusters per sequence or individuals per cluster-period) must be increased substantially, commonly by a factor of 1.5 to 3, but often by much more, to maintain 90\% power when switching from an IT model to an ETI model (targeting the TATE over the study). However, the inflation factor is lower for TATE estimands over shorter periods that exclude longer exposure times. In general, SW-CRT designs (including the "staircase" variant) have much greater power for estimating "short-term effects" relative to "long-term effects". For an ETI model targeting a TATE estimand, substantial power can be gained by adding time points to the start of the study or increasing baseline sample size, but surprisingly little power is gained from adding time points to the end of the study. More restrictive choices for modeling the exposure... [truncated]

stat.ME↗
arXiv2025

Flexible radiofrequency carbon nanotube transistors operating at frequencies above 100 GHz

The development of the sixth generation of wireless communications technology (6G) requires terminals that can operate at frequencies above 100 GHz. For human-centric applications, these terminals should also be flexible and have low power. However, current flexible radiofrequency transistors typically have lower maximum frequencies, in part due to the poor thermal conductivity of flexible substrates. Here, we report radiofrequency transistors that are based on aligned carbon nanotube arrays on flexible substrates and have current gain cutoff frequencies ($f_{\text{T}}$) and power gain cutoff frequencies ($f_{\text{max}}$) above 100 GHz. This is achieved by using electro-thermal co-design to improve the heat dissipation and radiofrequency performance of the devices. The transistors exhibit an on-state current of 0.947 mA $\mu$m$^{-1}$, a transconductance of 0.728 mS $\mu$m$^{-1}$, a peak extrinsic $f_{\text{T}}$ of 152 GHz, a peak extrinsic $f_{\text{max}}$ of 102 GHz, and a power consumption under 200 mW mm$^{-1}$. We also show that the devices can be used to create flexible radiofrequency amplifiers with an output power of 64 mW mm$^{-1}$ and a 11 dB power gain in the K-band.

physics.app-ph↗
arXiv2025

Survey and Improvement Strategies for Gene Prioritization with Large Language Models

Rare diseases are challenging to diagnose due to limited patient data and genetic diversity. Despite advances in variant prioritization, many cases remain undiagnosed. While large language models (LLMs) have performed well in medical exams, their effectiveness in diagnosing rare genetic diseases has not been assessed. To identify causal genes, we benchmarked various LLMs for gene prioritization. Using multi-agent and Human Phenotype Ontology (HPO) classification, we categorized patients based on phenotypes and solvability levels. As gene set size increased, LLM performance deteriorated, so we used a divide-and-conquer strategy to break the task into smaller subsets. At baseline, GPT-4 outperformed other LLMs, achieving near 30% accuracy in ranking causal genes correctly. The multi-agent and HPO approaches helped distinguish confidently solved cases from challenging ones, highlighting the importance of known gene-phenotype associations and phenotype specificity. We found that cases with specific phenotypes or clear associations were more accurately solved. However, we observed biases toward well-studied genes and input order sensitivity, which hindered gene prioritization. Our divide-and-conquer strategy improved accuracy by overcoming these biases. By utilizing HPO classification, novel multi-agent techniques, and our LLM strategy, we improved causal gene identification accuracy compared to our baseline evaluation. This approach streamlines rare disease diagnosis, facilitates reanalysis of unsolved cases, and accelerates gene discovery, supporting the development of targeted diagnostics and therapies.

q-bio.GN↗