arXiv subjects

Wu Li

Publications and source records attributed to Wu Li.

At least 19 recordsLinked to original sources

arXiv2026

Mitigating Context-Memory Conflicts in LLMs through Dynamic Cognitive Reconciliation Decoding

Large language models accumulate extensive parametric knowledge through pre-training. However, knowledge conflicts occur when outdated or incorrect parametric knowledge conflicts with external knowledge in the context. Existing methods address knowledge conflicts through contrastive decoding, but in conflict-free scenarios, static approaches disrupt output distribution. Other dynamic decoding methods attempt to measure the degree of conflict but still struggle with complex real-world situations. In this paper, we propose a two-stage decoding method called Dynamic Cognitive Reconciliation Decoding (DCRD), to predict and mitigate context-memory conflicts. DCRD first analyzes the attention map to assess context fidelity and predict potential conflicts. Based on this prediction, the input is directed to one of two decoding paths: (1) greedy decoding, or (2) context fidelity-based dynamic decoding. This design enables DCRD to handle conflicts efficiently while maintaining high accuracy and decoding efficiency in conflict-free cases. Additionally, to simulate scenarios with frequent knowledge updates, we constructed ConflictKG, a knowledge conflict QA benchmark. Experiments on four LLMs across six QA datasets show that DCRD outperforms all baselines, achieving state-of-the-art performance.

cs.CL↗

arXiv2026

Team-Based Self-Play With Dual Adaptive Weighting for Fine-Tuning LLMs

While recent self-training approaches have reduced reliance on human-labeled data for aligning LLMs, they still face critical limitations: (i) sensitivity to synthetic data quality, leading to instability and bias amplification in iterative training; (ii) ineffective optimization due to a diminishing gap between positive and negative responses over successive training iterations. In this paper, we propose Team-based self-Play with dual Adaptive Weighting (TPAW), a novel self-play algorithm designed to improve alignment in a fully self-supervised setting. TPAW adopts a team-based framework in which the current policy model both collaborates with and competes against historical checkpoints, promoting more stable and efficient optimization. To further enhance learning, we design two adaptive weighting mechanisms: (i) a response reweighting scheme that adjusts the importance of target responses, and (ii) a player weighting strategy that dynamically modulates each team member's contribution during training. Initialized from a SFT model, TPAW iteratively refines alignment without requiring additional human supervision. Experimental results demonstrate that TPAW consistently outperforms existing baselines across various base models and LLM benchmarks. Our code is publicly available at https://github.com/lab-klc/TPAW.

cs.CL↗

arXiv2026

Beyond the Individual: Virtualizing Multi-Disciplinary Reasoning for Clinical Intake via Collaborative Agents

The initial outpatient consultation is critical for clinical decision-making, yet it is often conducted by a single physician under time pressure, making it prone to cognitive biases and incomplete evidence capture. Although the Multi-Disciplinary Team (MDT) reduces these risks, they are costly and difficult to scale to real-time intake. We propose Aegle, a synchronous virtual MDT framework that brings MDT-level reasoning to outpatient consultations via a graph-based multi-agent architecture. Aegle formalizes the consultation state using a structured SOAP representation, separating evidence collection from diagnostic reasoning to improve traceability and bias control. An orchestrator dynamically activates specialist agents, which perform decoupled parallel reasoning and are subsequently integrated by an aggregator into a coherent clinical note. Experiments on ClinicalBench and a real-world RAPID-IPN dataset across 24 departments and 53 metrics show that Aegle consistently outperforms state-of-the-art proprietary and open-source models in documentation quality and consultation capability, while also improving final diagnosis accuracy. Our code is available at https://github.com/HovChen/Aegle.

cs.MA↗

arXiv2026

Directed Ordinal Diffusion Regularization for Progression-Aware Diabetic Retinopathy Grading

Diabetic Retinopathy (DR) progresses as a continuous and irreversible deterioration of the retina, following a well-defined clinical trajectory from mild to severe stages. However, most existing ordinal regression approaches model DR severity as a set of static, symmetric ranks, capturing relative order while ignoring the inherent unidirectional nature of disease progression. As a result, the learned feature representations may violate biological plausibility, allowing implausible proximity between non-consecutive stages or even reverse transitions. To bridge this gap, we propose Directed Ordinal Diffusion Regularization (D-ODR), which explicitly models the feature space as a directed flow by constructing a progression-constrained directed graph that strictly enforces forward disease evolution. By performing multi-scale diffusion on this directed structure, D-ODR imposes penalties on score inversions along valid progression paths, thereby effectively preventing the model from learning biologically inconsistent reverse transitions. This mechanism aligns the feature representation with the natural trajectory of DR worsening. Extensive experiments demonstrate that D-ODR yields superior grading performance compared to state-of-the-art ordinal regression and DR-specific grading methods, offering a more clinically reliable assessment of disease severity. Our code is available on https://github.com/HovChen/D-ODR.

cs.CV↗

arXiv2026

Echoes as Anchors: Probabilistic Costs and Attention Refocusing in LLM Reasoning

Test-time compute allocation in large reasoning models (LRMs) is widely used and has applications in mathematical problem solving, code synthesis, and planning. Recent work has addressed this problem by scaling self-consistency and parallel thinking, adding generic ``thinking tokens'' and prompting models to re-read the question before answering. Unfortunately, these approaches either inject task-agnostic tokens or mandate heuristics that do not explain -- and often ignore -- the \emph{spontaneous} repetition that many LRMs exhibit at the head of their internal chains. In contrast, we analyze and harness the model's tendency to restate the question, which we term the \emph{Echo of Prompt (EOP)}, as a front-loaded, compute-shaping mechanism. We formalize its probabilistic cost by casting echo removal as rejection-based conditioning and defining the \emph{Echo Likelihood Gap} $\Delta\mathcal{L}$ as a computable proxy. This provides the missing theoretical link that links early repetition to likelihood gains and downstream accuracy. However, it does not by itself specify how to exploit EOP. Consequently, we develop \emph{Echo-Distilled SFT (ED-SFT)} to instill an ``echo-then-reason'' pattern through supervised finetuning, and \emph{Echoic Prompting (EP)} to re-ground the model mid-trace without training. While promising, quantifying benefits beyond verbosity is non-trivial. Therefore, we conduct length and suffix-controlled likelihood analyses together with layer-wise attention studies, showing that EOP increases answer to answer-prefix attention in middle layers, consistent with an \emph{attention refocusing} mechanism. We evaluate on GSM8K, MathQA, Hendrycks-MATH, AIME24, and MATH-500 under identical decoding settings and budgets, and find consistent gains over baselines. Code is available at https://github.com/hhh2210/echoes-as-anchors.

cs.CL↗

arXiv2023

Multi-View Variational Autoencoder for Missing Value Imputation in Untargeted Metabolomics

Background: Missing data is a common challenge in mass spectrometry-based metabolomics, which can lead to biased and incomplete analyses. The integration of whole-genome sequencing (WGS) data with metabolomics data has emerged as a promising approach to enhance the accuracy of data imputation in metabolomics studies. Method: In this study, we propose a novel method that leverages the information from WGS data and reference metabolites to impute unknown metabolites. Our approach utilizes a multi-view variational autoencoder to jointly model the burden score, polygenetic risk score (PGS), and linkage disequilibrium (LD) pruned single nucleotide polymorphisms (SNPs) for feature extraction and missing metabolomics data imputation. By learning the latent representations of both omics data, our method can effectively impute missing metabolomics values based on genomic information. Results: We evaluate the performance of our method on empirical metabolomics datasets with missing values and demonstrate its superiority compared to conventional imputation techniques. Using 35 template metabolites derived burden scores, PGS and LD-pruned SNPs, the proposed methods achieved R^2-scores > 0.01 for 71.55% of metabolites. Conclusion: The integration of WGS data in metabolomics imputation not only improves data completeness but also enhances downstream analyses, paving the way for more comprehensive and accurate investigations of metabolic pathways and disease associations. Our findings offer valuable insights into the potential benefits of utilizing WGS data for metabolomics data imputation and underscore the importance of leveraging multi-modal data integration in precision medicine research.

q-bio.GN↗

arXiv2023

Electron-induced non-monotonic pressure dependence of the lattice thermal conductivity of {\theta}-TaN

Recent theoretical and experimental research suggests that $\theta$-TaN is a semimetal with high thermal conductivity ($\kappa$), primarily due to the contribution of phonons ($\kappa_\texttt{ph}$). By using first-principles calculations, we show a non-monotonic pressure dependence of the $\kappa$ of $\theta$-TaN. $\kappa_\texttt{ph}$ first increases until it reaches a maximum at around 60~GPa, and then decreases. This anomalous behaviour is a consequence of the competing pressure responses of phonon-phonon and phonon-electron interactions, in contrast to the known materials BAs and BP, where the non-monotonic pressure dependence is caused by the interplay between different phonon-phonon scattering channels. Although TaN has phonon dispersion features similar to BAs at ambient pressure, its response to pressure is different and an overall stiffening of the phonon branches takes place. Consequently, the relevant phonon-phonon scattering weakens as pressure increases. However, the increased electronic density of states near the Fermi level, and specifically the emergence of additional pockets of the Fermi surface at the high-symmetry L point in the Brillouin zone, leads to a substantial increase in phonon-electron scattering at high pressures, driving a decrease in $\kappa_{\mathrm{ph}}$. At intermediate pressures ($\sim$~20$-$70~GPa), the $\kappa$ of TaN surpasses that of BAs. Our work provides deeper insight into phonon transport in semimetals and metals where phonon-electron scattering is relevant.

cond-mat.mtrl-sci↗

arXiv2023

A high-performance GPU implementation of the electron-phonon Wannier interpolation and the related transport properties

The electron-phonon Wannier interpolation (EPWI) method is an efficient way to compute the properties of electron-phonon interactions (EPIs) accurately. This study presents a GPU-accelerated implementation of the EPWI method for computing transport properties, followed by a performance analysis. The implementation is tested on common systems such as aluminum and silicon. The results show complete consistency with those obtained through CPU computations. The proposed algorithm has the capability of computing the conductivity of aluminum in 20 minutes on a single NVIDIA Tesla V100 GPU, adopting a $200^3$ electron and phonon sampling grid. This speed is 173 times higher than the CPU-based algorithm, running on two nodes of the Intel Xeon Platinum 8260 CPU. Such impressive acceleration is achieved by carefully designing the algorithm to exploit the GPU's specific features. Furthermore, this methodology establishes a generic foundation for EPWI algorithms, which can be applied to other EPI-related properties.

cond-mat.mtrl-sci↗

arXiv2022

Single-dislocation phonons: atomic-scale measurement and their thermal properties

Nanoscale defects such as dislocations, have a significant impact on the phonon thermal transport properties in non-metallic materials. To unravel these effects, understanding of defect phonon modes is essential. Herein, at the atomic scale, the localized phonons of individual dislocation at a Si/Ge interface are measured via monochromated electron energy loss spectroscopy in a scanning transmission electron microscope. These modes are then correlated with the local microstructure, further revealing the dislocation effects on the local thermal transport properties. The dislocation causes phonon redshift in several milli-electron-volts within about two to four nanometers of the core, where both of the strain field and Ge-segregation play roles. With the presence of dislocation, the local interfacial thermal conductance can be either enhanced or reduced, depending on the complex interaction and competition between lattice-disorder (dislocation) and element-disorder (heterointerface mixing and Ge-segregation) at the interface. These findings provide valuable insights to improve the thermal properties of thermoelectric generators and thermal management systems through proper defect engineering.

cond-mat.mtrl-sci↗

arXiv2021

Raman Linewidth Contributions from Four-Phonon and Electron-Phonon Interactions in Graphene

The Raman peak position and linewidth provide insight into phonon anharmonicity and electron-phonon interactions (EPI) in materials. For monolayer graphene, prior first-principles calculations have yielded decreasing linewidth with increasing temperature, which is opposite to measurement results. Here, we explicitly consider four-phonon anharmonicity, phonon renormalization, and electron-phonon coupling, and find all to be important to successfully explain both the $G$ peak frequency shift and linewidths in our suspended graphene sample at a wide temperature range. Four-phonon scattering contributes a prominent linewidth that increases with temperature, while temperature dependence from EPI is found to be reversed above a doping threshold ($\hbar\omega_G/2$, with $\omega_G$ being the frequency of the $G$ phonon).

cond-mat.mtrl-sci↗

arXiv2021

FourPhonon: An extension module to ShengBTE for computing four-phonon scattering rates and thermal conductivity

FourPhonon is a computational package that can calculate four-phonon scattering rates in crystals. It is built within ShengBTE framework, which is a well-recognized lattice thermal conductivity solver based on Boltzmann transport equation. An adaptive energy broadening scheme is implemented for the calculation of four-phonon scattering rates. In analogy with $thirdorder.py$ in ShengBTE, we also provide a separate python script, $Fourthorder.py$, to calculate fourth-order interatomic force-constants. The extension module preserves all the nice features of the well-recognized lattice thermal conductivity solver ShengBTE, including good parallelism and straightforward workflow. In this paper, we discuss the general theory, program design, and example calculations on Si, BAs and $\mathrm{LiCoO_2}$.

cond-mat.mtrl-sci↗

arXiv2019

Combined treatment of phonon scattering by electrons and point defects explains the thermal conductivity reduction in highly-doped Si

The mechanisms causing the reduction in lattice thermal conductivity in highly P- and B-doped Si are looked into in detail. Scattering rates of phonons by point defects, as well as by electrons, are calculated from first principles. Lattice thermal conductivities are calculated considering these scattering mechanisms both individually and together. It is found that at low carrier concentrations and temperatures phonon scattering by electrons is dominant and can reproduce the experimental thermal conductivity reduction. However, at higher doping concentrations the scattering rates of phonons by point defects dominate the ones by electrons except for the lowest phonon frequencies. Consequently, phonon scattering by point defects contributes substantially to the thermal conductivity reduction in Si at defect concentrations above $10^{19}$ cm$^{-3}$ even at room temperature. Only when, phonon scattering by both point defects and electrons are taken into account, excellent agreement is obtained with the experimental values at all temperatures.

cond-mat.mtrl-sci↗

arXiv2019

Optimizing phonon scattering by tuning surface-interdiffusion-driven intermixing to break the random-alloy limit of thermal conductivity

We investigate the evolution of the cross-plane thermal conductivity $κ$ of superlattices (SLs) as interfaces change from perfectly abrupt to totally intermixed, by using non-equilibrium molecular dynamics simulations in combination with the spectral heat current calculations. We highlight the role of surface-interdiffusion-driven intermixing by calculating the $κ$ of SLs with changing interface roughness, whose tuning allows for the $κ$ values much lower than the "alloy limit" and the abrupt interface limit in same cases. The interplay between alloy and interface scattering in different frequency ranges provides a physical basis to predict a minimum of thermal conductivity. More specifically, we also explore how the interface roughness affects the thermal conductivities for SLs materials with a broad span of atomic mass and bond strength. In particular, we find that (i) only when the "spacer" thickness of SLs increases up to a critical value the $κ$ of rough SLs can break the corresponding "alloy limit". (ii) Whether the $κ$ changes monotonically as interface roughness strongly depends on the period length and intrinsic behavior of phonon transport for SLs materials. Especially, for the SL with large period length, there exists an optimal interface roughness which can minimize the thermal conductivity. (iii) Surface-interdiffusion-driven intermixing is more effective in achieving the low $κ$ below the alloy limit for SL materials with large mass mismatch than with small one. (iv) It's possible for SLs materials with large lattice mismatch (i.e., bond strength) to design an ideally abrupt interface structure with $κ$ much below the "alloy limit". These results have a clear implications for optimization of thermal transport for heat management and for the development of thermoelectric materials.

cond-mat.mtrl-sci↗

arXiv2019

Transport phenomena in a free-standing two-dimensional sodium sheet

The advances in the growth techniques provide numerous scope to explore the possibilities of new 2D materials for potential applications. With the aid of first-principle calculations we show that 2D Na can be a new addition to the family of thermodynamically stable 2D materials for device applications. Not surprisingly, due to half-occupied $3s$ orbital 2D Na possesses the features of the 2D electron gas (2DEG). The transport properties are examined based on the accurate solution of Boltzmann transport equation. With practically tunable carrier density in 2D materials, the intrinsic electrical resistivity of electron doped 2D Na is $\sim$ 1.4 times larger than that of graphene and falls below the latter 450 K onwards. The Bloch-Grüneisen temperature is almost constant at 50 K, independent of the type or density of the charge carriers. The electronic thermal conductivity of pure 2D Na is $\sim$ 1.24 times larger than that of its bulk counterpart at 300 K. The Wiedemann-Franz law stands tall in 2D Na with calculated Lorenz number 2.41 $\times 10^{-8} V^2/deg^2$ at room temperature. The transport mechanism presented here is expected to occur in all Na like systems with a clean Fermi surface.

cond-mat.mtrl-sci↗

arXiv2019

Significant reduction of lattice thermal conductivity in suspended graphene by charge doping

Declining the lattice thermal conductivity in graphene is essential for its thermoelectric applications. In high electron density systems, scatterings of phonons by electrons are no less than the phonon scatterings by other phonons. With the aid of first-principle calculations we examine the lattice thermal conductivity in graphene by inducing electron-phonon scattering externally. With experimentally tunable charge carrier density we find $\sim 50\%$ reduction of the lattice thermal conductivity at 200 K. The present study opens up new avenues for potential thermoelectric applications of graphene.

cond-mat.mtrl-sci↗

arXiv2019

Effect of local chemistry and structure on thermal transport in doped GaAs

Using a first-principles approach, we analyze the impact of \textit{DX} centers formed by S, Se, and Te dopant atoms on the thermal conductivity of GaAs. Our results are in good agreement with experiments and unveil the physics behind the drastically different effect of each kind of defect. We establish a causal chain linking the electronic structure of the dopants to the thermal conductivity of the bulk solid, a macroscopic transport coefficient. Specifically, the presence of lone pairs leads to the formation of structurally asymmetric \textit{DX} centers that cause resonant scattering of incident phonons. The effect of such resonances is magnified when they affect the part of the spectrum most relevant for thermal transport. We show that these resonances are associated with localized vibrational modes in the perturbed phonon spectrum. Finally, we illustrate the connection between flat adjacent minima in the energy landscape and resonant phonon scattering through detailed analyses of the energy landscape of the defective structures.

cond-mat.mtrl-sci↗

arXiv2016

Intrinsic ultralow lattice thermal conductivity of the unfilled skutterudite FeSb$_3$

It has been generally accepted that unfilled skutterudites process high lattice thermal conductivity ($κ_{l}$) that can be efficiently reduced upon filling. Here by using first principles Boltzmann-Peierls transport calculations, we find pure skutterudite of FeSb$_3$ with no filler in fact has an intrinsic ultralow $κ_{l}$ smaller than that of CoSb$_3$ by one order of magnitude. The value is even smaller than those of most of the fully filled skutterudites. This finding means that with FeSb$_3$ as a reference, filling does not necessarily lower $κ_{l}$. The ultralow $κ_{l}$ of FeSb$_3$ is a consequence of much softened optical phonon branches associated with the weakly bonded Sb$_4$ rings. They overlap more with heat-carrying acoustic phonons and significantly increase the phase space for three-phonon anharmonic scattering processes. This provides an alternative non-filling related mechanism for lowering the $κ_{l}$ of skutterudites.

cond-mat.mtrl-sci↗

arXiv2016

Physically founded phonon dispersions of few-layer materials, and the case of borophene

An increasing number of theoretical calculations on few-layer materials have been reporting a non-zero sound velocity for all three acoustic phonon modes. In contrast with these reports, here we show that the lowest phonon dispersion branch of atomistically described few-layer materials should be quadratic, and this can have dramatic consequencies on calculated properties, such as the thermal conductivity. By reformulating the interatomic force constants (IFC) in terms of internal coordinates, we find that a delicate balance between the IFCs is responsible for this quadraticity. This balance is hard to obtain in ab-initio calculations even if all the symmetries are numerically enforced a posteriori, but it arises naturally in our approach. We demonstrate the phenomenon in the case of borophene, where a very subtle correction to the ab-initio IFCs yields the physically correct quadratic dispersion, while leaving the rest of the spectrum virtually unmodified. Such quadraticity nevertheless has a major effect on the computed lattice thermal conductivity, which in the case of borophene changes by more than a factor 2, and reverses its anisotropy, when the subtle IFC correction is put in place.

cond-mat.mtrl-sci↗