arXiv subjects

Wanli Yang

Publications and source records attributed to Wanli Yang.

At least 19 recordsLinked to original sources

arXiv2026

Beyond Final Scores: A Systematic Evaluation of Agents for Long-Horizon AI Research and Development

Autonomous agents are increasingly capable of improving models, systems, and other technical artifacts through long-horizon experimentation. To understand the current state of this capability, however, evaluation must go beyond final scores, which neither reveal where progress is gained or lost nor indicate whether accumulated experience improves later decisions. We therefore present a systematic evaluation of seven frontier models on 36 long-horizon tasks based on a new framework that uses rule-based metrics to characterize within-run behavior through Solution Framing, Execution, and Feedback Control and controlled comparisons to assess experience reuse within and across tasks. The results show that current agents operate more like engineering optimizers than fully autonomous researchers: they can formulate and implement practical solutions, but their performance varies substantially across runs, their strongest solutions mainly adapt or combine established techniques, and genuine methodological novelty remains rare. Detailed analysis reveals that observed performance is shaped by multiple factors, including distinct process bottlenecks behind similar final outcomes, experience reuse that can help or mislead subsequent decisions, and harness designs that affect performance stability. These findings suggest concrete directions for improving model training, inference-time strategies, experience management, and harness design.

cs.AI↗

arXiv2026

Spectroscopic evidence for a molecular orbital Kondo insulator

A Kondo insulator (KI) is a prototypical example of a highly entangled phase of matter, where many-body interactions between local moments and delocalized electrons engender the non-magnetic insulating ground state. Conventionally, the local moments arise from atomic multiplet states with a narrow bandwidth, limiting Kondo coherence to low temperatures. Here, we realize a new paradigm for constructing the KI state with hybridized molecular orbitals in FeSb2. Resonant inelastic X-ray scattering (RIXS) at the Fe L-edge reveals distinct signatures of band-like continuum states and localized states. Comparisons with first-principles calculations establish a mixed-configuration ground state with hybridized Fe d-Sb p molecular orbitals as basis states. By systematically investigating the RIXS momentum, temperature, and doping dependences, we find propagating collective modes commensurate with many-body charge and spin excitations. Our results pave the way for understanding the emerging class of unconventional d electron insulators and engineering high temperature Kondo many-body states.

cond-mat.str-el↗

arXiv2026

Beyond Reasoning: Reinforcement Learning Unlocks Parametric Knowledge in LLMs

Reinforcement learning (RL) has achieved remarkable success in LLM reasoning, but whether it can also improve direct recall of parametric knowledge remains an open question. We study this question in a controlled zero-shot, one-hop, closed-book QA setting with no chain-of-thought, training only on binary correctness rewards and applying fact-level train-test deduplication to ensure gains reflect improved recall rather than reasoning or memorization. Across three model families and multiple factual QA benchmarks, RL yields ~27% average relative gains, surpassing both training- and inference-time baselines alike. Mechanistically, RL primarily redistributes probability mass over existing knowledge rather than acquiring new facts, moving correct answers from the low-probability tail into reliable greedy generations. Our data-attribution study reveals that the hardest examples are the most informative: those whose answers never appear in 128 pre-RL samples (only ~18% of training data) drive ~83% of the gain, since rare correct rollouts still emerge during training and get reinforced. Together, these findings broaden the role of RL beyond reasoning, repositioning it as a tool for unlocking rather than acquiring latent parametric knowledge.

cs.CL↗

arXiv2026

Interlayer Five-Spin Polaron in Superconducting Bilayer Nickelates

The discovery of high-$T_c$ superconductivity in Ruddlesden-Popper nickelates has sparked substantial effort towards understanding unconventional electronic states beyond a traditional cuprate-like $d^9$ configurational ground state. An understanding of the interplay between magnetic ground states and multi-orbital physics is key for establishing a microscopic mechanism for superconductivity. In the bilayer nickelates, spin density wave (SDW) order is a prominent feature in the non-superconducting regime, yet its relation to superconducting pairing remains an open question. Here, we use resonant x-ray scattering to examine the existence of SDW order in superconducting bilayer nickelate thin films La$_2$PrNi$_2$O$_7$ (LPNO). Comparing superconducting and oxygen-deficient LPNO thin films, we find that superconductivity occurs in SDW-free, oxygen-stoichiometric regions, whereas oxygen-deficiency promotes SDW order, indicating phase segregation of SDW and superconductivity. Furthermore, Ni-$L_3$ and O-$K$ edge spectroscopy reveals distinct electronic structures - particularly along the $c$-axis - between the two regions. Our results identify oxygen stoichiometry as a key parameter controlling interlayer coupling and thus the electronic structure of bilayer nickelates. In concert with theory, we propose that a ligand hole primarily resides at the inter-bilayer apical oxygen, forming a robust interlayer five-spin polaron state, which serves as the ground state for superconducting bilayer nickelates.

cond-mat.str-el↗

arXiv2026

FAIR-ESI: Feature Adaptive Importance Refinement for Electrophysiological Source Imaging

An essential technique for diagnosing brain disorders is electrophysiological source imaging (ESI). While model-based optimization and deep learning methods have achieved promising results in this field, the accurate selection and refinement of features remains a central challenge for precise ESI. This paper proposes FAIR-ESI, a novel framework that adaptively refines feature importance across different views, including FFT-based spectral feature refinement, weighted temporal feature refinement, and self-attention-based patch-wise feature refinement. Extensive experiments on two simulation datasets with diverse configurations and two real-world clinical datasets validate our framework's efficacy, highlighting its potential to advance brain disorder diagnosis and offer new insights into brain function.

cs.CV↗

arXiv2026

BaseCal: Unsupervised Confidence Calibration via Base Model Signals

Reliable confidence is essential for trusting the outputs of LLMs, yet widely deployed post-trained LLMs (PoLLMs) typically compromise this trust with severe overconfidence. In contrast, we observe that their corresponding base LLMs often remain well-calibrated. This naturally motivates us to calibrate PoLLM confidence using the base LLM as a reference. This work proposes two ways to achieve this. A straightforward solution, BaseCal-ReEval, evaluates PoLLM's responses by feeding them into the base LLM to get average probabilities as confidence. While effective, this approach introduces additional inference overhead. To address this, we propose BaseCal-Proj, which trains a lightweight projection to map the final-layer hidden states of PoLLMs back to those of their base LLMs. These projected states are then processed by the base LLM's output layer to derive base-calibrated confidence for PoLLM's responses. Notably, BaseCal is an unsupervised, plug-and-play solution that operates without human labels or LLM modifications. Experiments across five datasets and three LLM families demonstrate the effectiveness of BaseCal, reducing Expected Calibration Error (ECE) by an average of 42.90\% compared to the best unsupervised baselines.

cs.CL↗

arXiv2025

Fine-tuning Done Right in Model Editing

Fine-tuning, a foundational method for adapting large language models, has long been considered ineffective for model editing. Here, we challenge this belief, arguing that the reported failure arises not from the inherent limitation of fine-tuning itself, but from adapting it to the sequential nature of the editing task, a single-pass depth-first pipeline that optimizes each sample to convergence before moving on. While intuitive, this depth-first pipeline coupled with sample-wise updating over-optimizes each edit and induces interference across edits. Our controlled experiments reveal that simply restoring fine-tuning to the standard breadth-first (i.e., epoch-based) pipeline with mini-batch optimization substantially improves its effectiveness for model editing. Moreover, fine-tuning in editing also suffers from suboptimal tuning parameter locations inherited from prior methods. Through systematic analysis of tuning locations, we derive LocFT-BF, a simple and effective localized editing method built on the restored fine-tuning framework. Extensive experiments across diverse LLMs and datasets demonstrate that LocFT-BF outperforms state-of-the-art methods by large margins. Notably, to our knowledge, it is the first to sustain 100K edits and 72B-parameter models,10 x beyond prior practice, without sacrificing general capabilities. By clarifying a long-standing misconception and introducing a principled localized tuning strategy, we advance fine-tuning from an underestimated baseline to a leading method for model editing, establishing a solid foundation for future research.

cs.CL↗

arXiv2025

Negative Charge Transfer: Ground State Precursor towards High Energy Batteries

Modern energy applications, especially electric vehicles, demand high energy batteries. However, despite decades of intensive efforts, the highest energy density and commercially viable batteries are still based on LiCoO2, the very first generation of cathode materials. The technical bottleneck is the stability of oxide-based cathodes at high operating voltages. The fundamental puzzle is that we actually never understood the redox mechanism of LiCoO2. Conventional wisdom generally defines redox to be centered on cations at low voltages, and on anions, i.e. oxygen, at high voltages by forming oxidized chemical states like O2 or peroxo-species. Here, through in-situ and ex-situ spectroscopy coupled with theoretical calculations, we show that high-energy layered cathodes, represented by LiCoO2 and LiNiO2, operate through enhancement of negative charge transfer (NCT) ground states upon charging throughout the whole voltage range - i.e., NCT evolution itself is the intrinsic redox mechanism regardless of voltage ranges. NCT inherently engages high covalency and oxygen holes, leading to optimized performance without conventional redox centers in LiCoO2. The level of NCT, i.e., number of ligand holes, naturally explains many seemingly controversial results. The redefinition of redox mechanism reveals the pathway toward viable high energy battery electrodes.

cond-mat.mtrl-sci↗

arXiv2025

IEFS-GMB: Gradient Memory Bank-Guided Feature Selection Based on Information Entropy for EEG Classification of Neurological Disorders

Deep learning-based EEG classification is crucial for the automated detection of neurological disorders, improving diagnostic accuracy and enabling early intervention. However, the low signal-to-noise ratio of EEG signals limits model performance, making feature selection (FS) vital for optimizing representations learned by neural network encoders. Existing FS methods are seldom designed specifically for EEG diagnosis; many are architecture-dependent and lack interpretability, limiting their applicability. Moreover, most rely on single-iteration data, resulting in limited robustness to variability. To address these issues, we propose IEFS-GMB, an Information Entropy-based Feature Selection method guided by a Gradient Memory Bank. This approach constructs a dynamic memory bank storing historical gradients, computes feature importance via information entropy, and applies entropy-based weighting to select informative EEG features. Experiments on four public neurological disease datasets show that encoders enhanced with IEFS-GMB achieve accuracy improvements of 0.64% to 6.45% over baseline models. The method also outperforms four competing FS techniques and improves model interpretability, supporting its practical use in clinical settings.

cs.LG↗

arXiv2025

The Mirage of Model Editing: Revisiting Evaluation in the Wild

Despite near-perfect results reported in the literature, the effectiveness of model editing in real-world applications remains unclear. To bridge this gap, we introduce QAEdit, a new benchmark aligned with widely used question answering (QA) datasets, and WILD, a task-agnostic evaluation framework designed to better reflect real-world usage of model editing. Our single editing experiments show that current editing methods perform substantially worse than previously reported (38.5% vs. 96.8%). We demonstrate that it stems from issues in the synthetic evaluation practices of prior work. Among them, the most severe is the use of teacher forcing during testing, which leaks both content and length of the ground truth, leading to overestimated performance. Furthermore, we simulate practical deployment by sequential editing, revealing that current approaches fail drastically with only 1000 edits. This work calls for a shift in model editing research toward rigorous evaluation and the development of robust, scalable methods that can reliably update knowledge in LLMs for real-world use.

cs.CL↗

arXiv2025

Revealing Local Structures through Machine-Learning- Fused Multimodal Spectroscopy

Atomistic structures of materials offer valuable insights into their functionality. Determining these structures remains a fundamental challenge in materials science, especially for systems with defects. While both experimental and computational methods exist, each has limitations in resolving nanoscale structures. Core-level spectroscopies, such as x-ray absorption (XAS) or electron energy-loss spectroscopies (EELS), have been used to determine the local bonding environment and structure of materials. Recently, machine learning (ML) methods have been applied to extract structural and bonding information from XAS/EELS, but most of these frameworks rely on a single data stream, which is often insufficient. In this work, we address this challenge by integrating multimodal ab initio simulations, experimental data acquisition, and ML techniques for structure characterization. Our goal is to determine local structures and properties using EELS and XAS data from multiple elements and edges. To showcase our approach, we use various lithium nickel manganese cobalt (NMC) oxide compounds which are used for lithium ion batteries, including those with oxygen vacancies and antisite defects, as the sample material system. We successfully inferred local element content, ranging from lithium to transition metals, with quantitative agreement with experimental data. Beyond improving prediction accuracy, we find that ML model based on multimodal spectroscopic data is able to determine whether local defects such as oxygen vacancy and antisites are present, a task which is impossible for single mode spectra or other experimental techniques. Furthermore, our framework is able to provide physical interpretability, bridging spectroscopy with the local atomic and electronic structures.

cond-mat.mtrl-sci↗

arXiv2025

Automated Detection of Epileptic Spikes and Seizures Incorporating a Novel Spatial Clustering Prior

A Magnetoencephalography (MEG) time-series recording consists of multi-channel signals collected by superconducting sensors, with each signal's intensity reflecting magnetic field changes over time at the sensor location. Automating epileptic MEG spike detection significantly reduces manual assessment time and effort, yielding substantial clinical benefits. Existing research addresses MEG spike detection by encoding neural network inputs with signals from all channel within a time segment, followed by classification. However, these methods overlook simultaneous spiking occurred from nearby sensors. We introduce a simple yet effective paradigm that first clusters MEG channels based on their sensor's spatial position. Next, a novel convolutional input module is designed to integrate the spatial clustering and temporal changes of the signals. This module is fed into a custom MEEG-ResNet3D developed by the authors, which learns to extract relevant features and classify the input as a spike clip or not. Our method achieves an F1 score of 94.73% on a large real-world MEG dataset Sanbo-CMR collected from two centers, outperforming state-of-the-art approaches by 1.85%. Moreover, it demonstrates efficacy and stability in the Electroencephalographic (EEG) seizure detection task, yielding an improved weighted F1 score of 1.4% compared to current state-of-the-art techniques evaluated on TUSZ, whch is the largest EEG seizure dataset.

eess.SP↗

arXiv2024

LV-CadeNet: A Long-View Feature Convolution-Attention Fusion Encoder-Decoder Network for EEG/MEG Spike Analysis

The analysis of interictal epileptiform discharges (IEDs) in magnetoencephalography (MEG) or electroencephalogram (EEG) recordings represents a critical component in the diagnosis of epilepsy. However, manual analysis of these IEDs, which appear as epileptic spikes, from the large amount of MEG/EEG data is labor intensive and requires high expertise. Although automated methods have been developed to address this challenge, current approaches fail to fully emulate clinical experts' diagnostic intelligence in two key aspects: (1) their analysis on the input signals is limited to short temporal windows matching individual spike durations, missing the extended contextual patterns clinicians use to assess significance; and (2) they fail to adequately capture the dipole patterns with simultaneous positive-negative potential distributions across adjacent sensors that serve as clinicians' key diagnostic criterion for IED identification. To bridge this artificial-human intelligence gap, we propose a novel deep learning framework LV-CadeNet that integrates two key innovations: (1) a Long-View morphological feature representation that mimics expert clinicians' comprehensive assessment of both local spike characteristics and long-view contextual information, and (2) a hierarchical Encoder-Decoder NETwork that employs Convolution-Attention blocks for multi-scale spatiotemporal feature learning with progressive abstraction. Extensive evaluations confirm the superior performance of LV-CadeNet, which outperforms six state-of-the-art methods in EEG spike classification on TUEV, the largest public EEG spike dataset. Additionally, LV-CadeNet attains a significant improvement of 13.58% in balanced accuracy over the leading baseline for MEG spike detection on a clinical MEG dataset from Sanbo Brain Hospital, Capital Medical University.

cs.CV↗

arXiv2024

Understanding the Collapse of LLMs in Model Editing

Despite significant progress in model editing methods, their application in real-world scenarios remains challenging as they often cause large language models (LLMs) to collapse. Among them, ROME is particularly concerning, as it could disrupt LLMs with only a single edit. In this paper, we study the root causes of such collapse. Through extensive analysis, we identify two primary factors that contribute to the collapse: i) inconsistent handling of prefixed and unprefixed keys in the parameter update equation may result in very small denominators, causing excessively large parameter updates; ii) the subject of collapse cases is usually the first token, whose unprefixed key distribution significantly differs from the prefixed key distribution in autoregressive transformers, causing the aforementioned issue to materialize. To validate our findings, we propose a simple yet effective approach: uniformly using prefixed keys during editing phase and adding prefixes during testing phase to ensure the consistency between training and testing. The experimental results show that the proposed solution can prevent model collapse while maintaining the effectiveness of the edits.

cs.CL↗

arXiv2024

The Butterfly Effect of Model Editing: Few Edits Can Trigger Large Language Models Collapse

Although model editing has shown promise in revising knowledge in Large Language Models (LLMs), its impact on the inherent capabilities of LLMs is often overlooked. In this work, we reveal a critical phenomenon: even a single edit can trigger model collapse, manifesting as significant performance degradation in various benchmark tasks. However, benchmarking LLMs after each edit, while necessary to prevent such collapses, is impractically time-consuming and resource-intensive. To mitigate this, we propose using perplexity as a surrogate metric, validated by extensive experiments demonstrating changes in an edited model's perplexity are strongly correlated with its downstream task performances. We further conduct an in-depth study on sequential editing, a practical setting for real-world scenarios, across various editing methods and LLMs, focusing on hard cases from our previous single edit studies. The results indicate that nearly all examined editing methods result in model collapse after only few edits. To facilitate further research, we have utilized GPT-3.5 to develop a new dataset, HardEdit, based on those hard cases. This dataset aims to establish the foundation for pioneering research in reliable model editing and the mechanisms underlying editing-induced model collapse. We hope this work can draw the community's attention to the potential risks inherent in model editing practices.

cs.AI↗

arXiv2024

Blinded by Generated Contexts: How Language Models Merge Generated and Retrieved Contexts When Knowledge Conflicts?

While auxiliary information has become a key to enhancing Large Language Models (LLMs), relatively little is known about how LLMs merge these contexts, specifically contexts generated by LLMs and those retrieved from external sources. To investigate this, we formulate a systematic framework to identify whether LLMs' responses are attributed to either generated or retrieved contexts. To easily trace the origin of the response, we construct datasets with conflicting contexts, i.e., each question is paired with both generated and retrieved contexts, yet only one of them contains the correct answer. Our experiments reveal a significant bias in several LLMs (GPT-4/3.5 and Llama2) to favor generated contexts, even when they provide incorrect information. We further identify two key factors contributing to this bias: i) contexts generated by LLMs typically show greater similarity to the questions, increasing their likelihood of being selected; ii) the segmentation process used in retrieved contexts disrupts their completeness, thereby hindering their full utilization in LLMs. Our analysis enhances the understanding of how LLMs merge diverse contexts, offers valuable insights for advancing current LLM augmentation methods, and highlights the risk of generated misinformation for retrieval-augmented LLMs.

cs.CL↗

arXiv2023

Robust Machine Learning Inference from X-ray Absorption Near Edge Spectra through Featurization

X-ray absorption spectroscopy (XAS) is a commonly-employed technique for characterizing functional materials. In particular, x-ray absorption near edge spectra (XANES) encodes local coordination and electronic information and machine learning approaches to extract this information is of significant interest. To date, most ML approaches for XANES have primarily focused on using the raw spectral intensities as input, overlooking the potential benefits of incorporating spectral transformations and dimensionality reduction techniques into ML predictions. In this work, we focused on systematically comparing the impact of different featurization methods on the performance of ML models for XAS analysis. We evaluated the classification and regression capabilities of these models on computed datasets and validated their performance on previously unseen experimental datasets. Our analysis revealed an intriguing discovery: the cumulative distribution function (CDF) feature achieves both high prediction accuracy and exceptional transferability. This remarkably robust performance can be attributed to its tolerance to horizontal shifts in spectra, which is crucial when validating models using experimental data. While this work exclusively focuses on XANES analysis, we anticipate that the methodology presented here will hold promise as a versatile asset to the broader spectroscopy community.

physics.comp-ph↗

arXiv2021

Synthetic control over polymorph formation in the d-band semiconductor system FeS$_2$

Pyrite, also known as fool's gold is the thermodynamic stable polymorph of FeS$_2$. It is widely considered as a promising d-band semiconductor for various applications due to its intriguing physical properties. Marcasite is the other naturally occurring polymorph of FeS$_2$. Measurements on natural crystals have shown that it has similarly promising electronic, mechanical, and optical properties as pyrite. However, it has been only scarcely investigated so far, because the laboratory-based synthesis of phase-pure samples or high-quality marcasite single crystal has been a challenge until now. Here, we report the targeted phase formation via hydrothermal synthesis of marcasite and pyrite. The formation condition and phase purity of the FeS$_2$ polymorphs are systematically studied in the form of a comprehensive synthesis map. We, furthermore, report on a detailed analysis of marcasite single crystal growth by a space-separated hydrothermal synthesis. We observe that single phase product of marcasite forms only on the surface under the involvement of H$_2$S and sulphur vapor. The availability of high-quality crystals of marcasite allows us to measure the fundamental physical properties, including an allowed direct optical bandgap of 0.76 eV, temperature independent diamagnetism, an electronic transport gap of 0.11 eV, and a room-temperature carrier concentration of 4.14 $\times$ 10$^{18}$ cm$^{-3}$. X-ray absorption/emission spectroscopy are employed to measure the band gap of the two FeS$_2$ phases. We find marcasite has a band gap of 0.73 eV, while pyrite has a band gap of 0.87 eV. Our results indicate that marcasite -- that is now synthetically available in a straightforward fashion -- is as equally promising as pyrite as candidate for various semiconductor applications based on earth abundant elements.

cond-mat.mtrl-sci↗