Search

There is a large and growing number of non-degree credential offerings between a high school diploma and a bachelor's degree, as well as degree programs beyond a bachelor's degree. Nevertheless, research on the financial returns to non-degree credentials and degree-granting programs is often narrow and siloed. To fill this gap, we leverage a national sample of individuals across nine MSAs and four industries to examine the relative financial returns to a variety of non-degree credentials and degree programs. Leveraging fixed-effect models, we explore the relationship between completing a credential or degree and earnings premiums. We find that an associate's, bachelor's, master's and doctorate degree follows a similar model of returns in which the number of schooling years is linearly related to proportional earnings premiums. However, students completing sub-baccalaureate certificates, post-baccalaureate certificates, and non-school credentials appear to get larger financial returns for less time. Furthermore, while the returns to both non-degree credentials and degree granting programs generally favored males over females and non-binary persons, this was not the case for race/ethnicity. Although individuals from Asian and White racial/ethnic groups often maintained an advantage in traditional education settings, Black individuals earned greater premiums from non-school credentials than White individuals, which may represent an opportunity to close racial/ethnic gaps in earnings.

Zero-shot object counting aims to count instances of arbitrary object categories specified by text descriptions. Existing methods typically rely on vision-language models like CLIP, but often exhibit limited sensitivity to text prompts. We present T2ICount, a diffusion-based framework that leverages rich prior knowledge and fine-grained visual understanding from pretrained diffusion models. While one-step denoising ensures efficiency, it leads to weakened text sensitivity. To address this challenge, we propose a Hierarchical Semantic Correction Module that progressively refines text-image feature alignment, and a Representational Regional Coherence Loss that provides reliable supervision signals by leveraging the cross-attention maps extracted from the denosing U-Net. Furthermore, we observe that current benchmarks mainly focus on majority objects in images, potentially masking models' text sensitivity. To address this, we contribute a challenging re-annotated subset of FSC147 for better evaluation of text-guided counting ability. Extensive experiments demonstrate that our method achieves superior performance across different benchmarks. Code is available at https://github.com/cha15yq/T2ICount., Comment: Accepted by CVPR2025

Landau levels in certain models are known to protrude into the zero-field energy gap. These are known as anomalous Landau levels (ALLs). We study whether ALLs can lead to in-gap quantum oscillation in the absence of a zero-field Fermi surface. Focusing on two-dimensional multi-band low-energy models of electrons with continuous rotation symmetry, we show that an effective-band description, akin to the semiclassical treatment of Landau level problems in metals, can be used to predict the Landau level spectrum, including possible ALLs. This description then predicts quantum oscillation for certain insulating models, which we demonstrate through numerical calculations., Comment: 10 pages, 4 figures

A critical requirement for deep learning models is ensuring their robustness against adversarial attacks. These attacks commonly introduce noticeable perturbations, compromising the visual fidelity of adversarial examples. Another key challenge is that while white-box algorithms can generate effective adversarial perturbations, they require access to the model gradients, limiting their practicality in many real-world scenarios. Existing attack mechanisms struggle to achieve similar efficacy without access to these gradients. In this paper, we introduce GreedyPixel, a novel pixel-wise greedy algorithm designed to generate high-quality adversarial examples using only query-based feedback from the target model. GreedyPixel improves computational efficiency in what is typically a brute-force process by perturbing individual pixels in sequence, guided by a pixel-wise priority map. This priority map is constructed by ranking gradients obtained from a surrogate model, providing a structured path for perturbation. Our results demonstrate that GreedyPixel achieves attack success rates comparable to white-box methods without the need for gradient information, and surpasses existing algorithms in black-box settings, offering higher success rates, reduced computational time, and imperceptible perturbations. These findings underscore the advantages of GreedyPixel in terms of attack efficacy, time efficiency, and visual quality.

In the Hong Kong Observatory, the Analogue Forecast System (AFS) for precipitation has been providing useful reference in predicting possible daily rainfall scenarios for the next 9 days, by identifying historical cases with similar weather patterns to the latest output from the deterministic model of the European Centre for Medium-Range Weather Forecasts (ECMWF). Recent advances in machine learning allow more sophisticated models to be trained using historical data and the patterns of high-impact weather events to be represented more effectively. As such, an enhanced AFS has been developed using the deep learning technique autoencoder. The datasets of the fifth generation of the ECMWF Reanalysis (ERA5) are utilised where more meteorological elements in higher horizontal, vertical and temporal resolutions are available as compared to the previous ECMWF reanalysis products used in the existing AFS. The enhanced AFS features four major steps in generating the daily rain class forecasts: (1) preprocessing of gridded ERA5 and ECMWF model forecast, (2) feature extraction by the pretrained autoencoder, (3) application of optimised feature weightings based on historical cases, and (4) calculation of the final rain class from a weighted ensemble of top analogues. The enhanced AFS demonstrates a consistent and superior performance over the existing AFS, especially in capturing heavy rain cases, during the verification period from 2019 to 2022. This paper presents the detailed formulation of the enhanced AFS and discusses its advantages and limitations in supporting precipitation forecasting in Hong Kong., Comment: 16 pages, 10 figures

We present polarization pulse profiles for 56 millisecond pulsars (MSPs) monitored by the Chinese Pulsar Timing Array (CPTA) collaboration using the Five-hundred-meter Aperture Spherical radio Telescope (FAST). The observations centered at 1.25 GHz with a raw bandwidth of 500 MHz. Due to the high sensitivity ($\sim$16 K/Jy) of the FAST telescope and our long integration time, the high signal-to-noise ratio polarization profiles show features hardly detected before. Among 56 pulsars, the polarization profiles of PSRs J0406$+$3039, J1327$+$3423, and J2022$+$2534 were not previously reported. 80\% of MSPs in the sample show weak components below 3\% of peak flux, 25\% of pulsars show interpulse-like structures, and most pulsars show linear polarization position angle jumps. Six pulsars seem to be emitting for full rotation phase, with another thirteen pulsars being good candidates for such a 360$^\circ$ radiator. We find that the distribution of the polarization percentage in our sample is compatible with the normal pulsar distribution. Our detailed evaluation of the MSP polarization properties suggests that the wave propagation effects in the pulsar magnetosphere are important in shaping the MSP polarization pulse profiles., Comment: 17 pages, 10 figures, 2 tables, accepted by A&A

Machine Translation (MT) has been predominantly designed for sentence-level translation using transformer-based architectures. While next-token prediction based Large Language Models (LLMs) demonstrate strong capabilities in long-text translation, non-extensive language models often suffer from omissions and semantic inconsistencies when processing paragraphs. Existing preference alignment methods improve sentence-level translation but fail to ensure coherence over extended contexts due to the myopic nature of next-token generation. We introduce Plan2Align, a test-time alignment framework that treats translation as a predictive planning problem, adapting Model Predictive Control to iteratively refine translation outputs. Experiments on WMT24 Discourse-Level Literary Translation show that Plan2Align significantly improves paragraph-level translation, achieving performance surpassing or on par with the existing training-time and test-time alignment methods on LLaMA-3.1 8B., Comment: Preprint. Code will be released at Plan2Align GitHub link: https://github.com/NYCU-RL-Bandits-Lab/Plan2Align

Jets and outflows are commonly observed in young stellar objects (YSOs), yet their origins remain debated. Using 3D non-ideal magnetohydrodynamic (MHD) simulations of a circumstellar disk threaded by a large-scale open poloidal magnetic field, we identify three components in the disk-driven outflow: (1) a fast, collimated jet, (2) a less collimated, slower laminar disk wind, and (3) a magneto-rotational instability (MRI)-active disk wind that separates the former two. At high altitudes, the MRI-active wind merges with the laminar disk wind, leaving only the jet and disk wind as distinct components. The jet is powered by a novel mechanism: a lightly mass-loaded outflow driven primarily by toroidal magnetic pressure in the low-density polar funnel near the system's rotation axis. A geometric analysis of the magnetic field structure confirms that magnetic tension does not contribute to the outflow acceleration, with magnetic pressure acting as the dominant driver. While the outflow in our model shares similarities with the classical magneto-centrifugal model-such as angular momentum extraction from the accreting disk-centrifugal forces play a negligible role in jet acceleration. In particular, the flow near the jet base does not satisfy the conditions for magneto-centrifugal wind launching. Additionally, the jet in our simulation exhibits strong spatial and temporal variability. These differences challenge the applicability of rotation-outflow velocity relations derived from steady-state, axisymmetric magneto-centrifugal jet models for estimating the jet's launching radius. For the slower disk wind, vertical motion is primarily driven by toroidal magnetic pressure, while centrifugal forces widen the wind's opening angle near its base.

Large language models (LLMs) demonstrate strong task-specific capabilities through fine-tuning, but merging multiple fine-tuned models often leads to degraded performance due to overlapping instruction-following components. Task Arithmetic (TA), which combines task vectors derived from fine-tuning, enables multi-task learning and task forgetting but struggles to isolate task-specific knowledge from general instruction-following behavior. To address this, we propose Layer-Aware Task Arithmetic (LATA), a novel approach that assigns layer-specific weights to task vectors based on their alignment with instruction-following or task-specific components. By amplifying task-relevant layers and attenuating instruction-following layers, LATA improves task learning and forgetting performance while preserving overall model utility. Experiments on multiple benchmarks, including WikiText-2, GSM8K, and HumanEval, demonstrate that LATA outperforms existing methods in both multi-task learning and selective task forgetting, achieving higher task accuracy and alignment with minimal degradation in output quality. Our findings highlight the importance of layer-wise analysis in disentangling task-specific and general-purpose knowledge, offering a robust framework for efficient model merging and editing.

Membership Inference Attacks (MIAs) aim to predict whether a data sample belongs to the model's training set or not. Although prior research has extensively explored MIAs in Large Language Models (LLMs), they typically require accessing to complete output logits (\ie, \textit{logits-based attacks}), which are usually not available in practice. In this paper, we study the vulnerability of pre-trained LLMs to MIAs in the \textit{label-only setting}, where the adversary can only access generated tokens (text). We first reveal that existing label-only MIAs have minor effects in attacking pre-trained LLMs, although they are highly effective in inferring fine-tuning datasets used for personalized LLMs. We find that their failure stems from two main reasons, including better generalization and overly coarse perturbation. Specifically, due to the extensive pre-training corpora and exposing each sample only a few times, LLMs exhibit minimal robustness differences between members and non-members. This makes token-level perturbations too coarse to capture such differences. To alleviate these problems, we propose \textbf{PETAL}: a label-only membership inference attack based on \textbf{PE}r-\textbf{T}oken sem\textbf{A}ntic simi\textbf{L}arity. Specifically, PETAL leverages token-level semantic similarity to approximate output probabilities and subsequently calculate the perplexity. It finally exposes membership based on the common assumption that members are `better' memorized and have smaller perplexity. We conduct extensive experiments on the WikiMIA benchmark and the more challenging MIMIR benchmark. Empirically, our PETAL performs better than the extensions of existing label-only attacks against personalized LLMs and even on par with other advanced logit-based attacks across all metrics on five prevalent open-source LLMs., Comment: Accepted by USENIX Security 2025

Building articulated objects is a key challenge in computer vision. Existing methods often fail to effectively integrate information across different object states, limiting the accuracy of part-mesh reconstruction and part dynamics modeling, particularly for complex multi-part articulated objects. We introduce ArtGS, a novel approach that leverages 3D Gaussians as a flexible and efficient representation to address these issues. Our method incorporates canonical Gaussians with coarse-to-fine initialization and updates for aligning articulated part information across different object states, and employs a skinning-inspired part dynamics modeling module to improve both part-mesh reconstruction and articulation learning. Extensive experiments on both synthetic and real-world datasets, including a new benchmark for complex multi-part objects, demonstrate that ArtGS achieves state-of-the-art performance in joint parameter estimation and part mesh reconstruction. Our approach significantly improves reconstruction quality and efficiency, especially for multi-part articulated objects. Additionally, we provide comprehensive analyses of our design choices, validating the effectiveness of each component to highlight potential areas for future improvement.

In recent years, machine learning potentials (MLPs) have become indispensable tools in physics, chemistry, and materials science, driving the development of software packages for molecular dynamics (MD) simulations and related applications. These packages, typically built on specific machine learning frameworks such as TensorFlow, PyTorch, or JAX, face integration challenges when advanced applications demand communication across different frameworks. The previous TensorFlow-based implementation of DeePMD-kit exemplified these limitations. In this work, we introduce DeePMD-kit version 3, a significant update featuring a multi-backend framework that supports TensorFlow, PyTorch, JAX, and PaddlePaddle backends, and demonstrate the versatility of this architecture through the integration of other MLPs packages and of Differentiable Molecular Force Field. This architecture allows seamless backend switching with minimal modifications, enabling users and developers to integrate DeePMD-kit with other packages using different machine learning frameworks. This innovation facilitates the development of more complex and interoperable workflows, paving the way for broader applications of MLPs in scientific research.

In recent years, kagome lattices have garnered significant attention for their diverse properties in topology, magnetism, and electron correlations. However, the exploration of breathing kagome lattices, which exhibit dynamic breathing behavior, remains relatively scarce. Structural breathing introduces an additional degree of freedom that is anticipated to fine-tune the electronic structure, potentially leading to exotic properties within the system. In this study, we employ a combination of the kp model and first-principles calculations to explore how breathing ferroelectricity can modulate valley states within a monolayer of niobium halide with breathing kagome lattice. Through the interplay of magnetoelectric coupling and the lock-in between breathing and ferroelectric states, we demonstrate that a dynamically breathing process, when controlled by an appropriately applied electric field, can achieve valley polarization reversal and generate multiple valley states, including those that are topologically nontrivial. These state transformations may couple to distinctive properties in circularly-polarized optical responses and various valley Hall effects. Consequently, our results suggest that materials featuring breathing kagome lattices represent promising platforms for studying the interplay among structure, charge, spin, and valley degrees of freedom, a crucial step toward developing multifunctional devices.

Steganography is the art and science of covert writing, with a broad range of applications interwoven within the realm of cybersecurity. As artificial intelligence continues to evolve, its ability to synthesise realistic content emerges as a threat in the hands of cybercriminals who seek to manipulate and misrepresent the truth. Such synthetic content introduces a non-trivial risk of overwriting the subtle changes made for the purpose of steganography. When the signals in both the spatial and temporal domains are vulnerable to unforeseen overwriting, it calls for reflection on what can remain invariant after all. This study proposes a paradigm in steganography for audiovisual media, where messages are concealed beyond both spatial and temporal domains. A chain of multimodal agents is developed to deconstruct audiovisual content into a cover text, embed a message within the linguistic domain, and then reconstruct the audiovisual content through synchronising both aural and visual modalities with the resultant stego text. The message is encoded by biasing the word sampling process of a language generation model and decoded by analysing the probability distribution of word choices. The accuracy of message transmission is evaluated under both zero-bit and multi-bit capacity settings. Fidelity is assessed through both biometric and semantic similarities, capturing the identities of the recorded face and voice, as well as the core ideas conveyed through the media. Secrecy is examined through statistical comparisons between cover and stego texts. Robustness is tested across various scenarios, including audiovisual compression, face-swapping, voice-cloning and their combinations.

This paper argues that generating output tokens is more effective than using pooled representations for prediction tasks because token-level generation retains more mutual information. Since LLMs are trained on massive text corpora using next-token prediction, generation aligns naturally with their learned behavior. Using the Data Processing Inequality (DPI), we provide both theoretical and empirical evidence supporting this claim. However, autoregressive models face two key challenges when used for prediction: (1) exposure bias, where the model sees ground truth tokens during training but relies on its own predictions during inference, leading to errors, and (2) format mismatch, where discrete tokens do not always align with the tasks required output structure. To address these challenges, we introduce PredGen(Predicting Through Generating), an end to end framework that (i) uses scheduled sampling to reduce exposure bias, and (ii) introduces a task adapter to convert the generated tokens into structured outputs. Additionally, we introduce Writer-Director Alignment Loss (WDAL), which ensures consistency between token generation and final task predictions, improving both text coherence and numerical accuracy. We evaluate PredGen on multiple classification and regression benchmarks. Our results show that PredGen consistently outperforms standard baselines, demonstrating its effectiveness in structured prediction tasks., Comment: Preprint paper

In recent years, with the rapid development of machine learning, physicists have been exploring its new applications in solving or alleviating the curse of dimensionality in many-body problems. In order to accurately reflect the underlying physics of the problem, domain knowledge must be encoded into the machine learning algorithms. In this work, inspired by field theory, we propose a new set of machine learning models called effective field neural networks (EFNNs) that can automatically and efficiently capture important many-body interactions through multiple self-refining processes. Taking the classical $3$-spin infinite-range model and the quantum double exchange model as case studies, we explicitly demonstrate that EFNNs significantly outperform fully-connected deep neural networks (DNNs) and the effective model. Furthermore, with the help of convolution operations, the EFNNs learned in a small system can be seamlessly used in a larger system without additional training and the relative errors even decrease, which further demonstrates the efficacy of EFNNs in representing core physical behaviors.

Strength estimation and adjustment are crucial in designing human-AI interactions, particularly in games where AI surpasses human players. This paper introduces a novel strength system, including a strength estimator (SE) and an SE-based Monte Carlo tree search, denoted as SE-MCTS, which predicts strengths from games and offers different playing strengths with human styles. The strength estimator calculates strength scores and predicts ranks from games without direct human interaction. SE-MCTS utilizes the strength scores in a Monte Carlo tree search to adjust playing strength and style. We first conduct experiments in Go, a challenging board game with a wide range of ranks. Our strength estimator significantly achieves over 80% accuracy in predicting ranks by observing 15 games only, whereas the previous method reached 49% accuracy for 100 games. For strength adjustment, SE-MCTS successfully adjusts to designated ranks while achieving a 51.33% accuracy in aligning to human actions, outperforming a previous state-of-the-art, with only 42.56% accuracy. To demonstrate the generality of our strength system, we further apply SE and SE-MCTS to chess and obtain consistent results. These results show a promising approach to strength estimation and adjustment, enhancing human-AI interactions in games. Our code is available at https://rlg.iis.sinica.edu.tw/papers/strength-estimator., Comment: Accepted by the Thirteenth International Conference on Learning Representations (ICLR 2025)

Recent studies show that the visual place recognition (VPR) method using pre-trained visual foundation models can achieve promising performance. In our previous work, we propose a novel method to realize seamless adaptation of foundation models to VPR (SelaVPR). This method can produce both global and local features that focus on discriminative landmarks to recognize places for two-stage VPR by a parameter-efficient adaptation approach. Although SelaVPR has achieved competitive results, we argue that the previous adaptation is inefficient in training time and GPU memory usage, and the re-ranking paradigm is also costly in retrieval latency and storage usage. In pursuit of higher efficiency and better performance, we propose an extension of the SelaVPR, called SelaVPR++. Concretely, we first design a parameter-, time-, and memory-efficient adaptation method that uses lightweight multi-scale convolution (MultiConv) adapters to refine intermediate features from the frozen foundation backbone. This adaptation method does not back-propagate gradients through the backbone during training, and the MultiConv adapter facilitates feature interactions along the spatial axes and introduces proper local priors, thus achieving higher efficiency and better performance. Moreover, we propose an innovative re-ranking paradigm for more efficient VPR. Instead of relying on local features for re-ranking, which incurs huge overhead in latency and storage, we employ compact binary features for initial retrieval and robust floating-point (global) features for re-ranking. To obtain such binary features, we propose a similarity-constrained deep hashing method, which can be easily integrated into the VPR pipeline. Finally, we improve our training strategy and unify the training protocol of several common training datasets to merge them for better training of VPR models. Extensive experiments show that ......

Altermagnetism, as an unconventional antiferromagnetism, exhibits collinear-compensated magnetic order in real space and spin-splitting band structure in reciprocal space. In this work, we propose a general approach to generating multi-component structures with two-dimensional altermagnetism, based on symmetry analysis. Specifically, by analyzing the space group of the crystal structures and their subgroups, we systematically categorize equivalent atomic positions and arrange them into orbits based on symmetry operations. Chemical elements are then allowed to occupy all atomic positions on these orbits, generating candidate structures with specific symmetries. We present a general technique for generating collinear-compensated magnetic order, characterized by the symmetrical interconnection between opposite-spin sublattices, and employ first-principles calculations to determine magnetic ground states of multi-component materials. This approach integrates symmetry analysis with the screening of altermagnetic configurations to evaluate the likelihood of candidates possessing altermagnetism. To verify the methodology, we provide examples of previously unreported 2D altermagnets, such as Cr2Si2S3Se3, Fe2P2S3Se3, and V2O2BrI3, and evaluate their dynamical stability by calculating the phonon spectrum. The results demonstrate the feasibility of our approach in generating stable multi-component structures with two-dimensional altermagnetism. Our research has significantly enriched the candidate materials for 2D altermagnet and provided a reference for experimental synthesis.

Recently, it was proposed that an off-center dipole magnetic configuration, together with a non-trivial temperature profile, may be the best model to explain the X-ray light curve of PSR J0030+0451 observed by the Neutron Star Interior Composition Explorer (\emph{NICER}). Using a theoretical model for the electric current density in a force-free pulsar magnetosphere, we compute from first principles the distribution of electric current over the polar cap associated with an off-center magnetic dipole. We then use a simple prescription to compute the resulting temperature distribution, which allows us to derive the observed X-ray light curve. We investigate the role of the volumetric return current region in the polar cap and find that although it does not make a big difference in an aligned dipole case, the difference can be bigger in the case of an off-center dipole. Finally, we apply Markov Chain Monte Carlo (MCMC) fitting to the X-ray light curves of pulsars PSR J0030+0451 and PSR J0437--4715 with and without the volumetric return current, and find that our model can reasonably recover the observed X-ray light curves., Comment: Submitted to ApJ

Soft robots, made from compliant materials, exhibit complex dynamics due to their flexibility and high degrees of freedom. Controlling soft robots presents significant challenges, particularly underactuation, where the number of inputs is fewer than the degrees of freedom. This research aims to develop a digital twin for multi-fingered soft grippers to advance the development of underactuation algorithms. The digital twin is designed to capture key effects observed in soft robots, such as nonlinearity, hysteresis, uncertainty, and time-varying phenomena, ensuring it closely replicates the behavior of a real-world soft gripper. Uncertainty is simulated using the Monte Carlo method. With the digital twin, a Q-learning algorithm is preliminarily applied to identify the optimal motion speed that minimizes uncertainty caused by the soft robots. Underactuated motions are successfully simulated within this environment. This digital twin paves the way for advanced machine learning algorithm training., Comment: 6 pages, 5 figures

Large language models (LLMs) exhibit excellent performance in various tasks. However, the memory requirements of LLMs present a great challenge when deploying on memory-limited devices, even for quantized LLMs. This paper introduces a framework to compress LLM after quantization further, achieving about 2.2x compression ratio. A compression-aware quantization is first proposed to enhance model weight compressibility by re-scaling the model parameters before quantization, followed by a pruning method to improve further. Upon this, we notice that decompression can be a bottleneck during practical scenarios. We then give a detailed analysis of the trade-off between memory usage and latency brought by the proposed method. A speed-adaptive method is proposed to overcome it. The experimental results show inference with the compressed model can achieve a 40% reduction in memory size with negligible loss in accuracy and inference speed.

In both artificial and biological systems, the centered kernel alignment (CKA) has become a widely used tool for quantifying neural representation similarity. While current CKA estimators typically correct for the effects of finite stimuli sampling, the effects of sampling a subset of neurons are overlooked, introducing notable bias in standard experimental scenarios. Here, we provide a theoretical analysis showing how this bias is affected by the representation geometry. We then introduce a novel estimator that corrects for both input and feature sampling. We use our method for evaluating both brain-to-brain and model-to-brain alignments and show that it delivers reliable comparisons even with very sparsely sampled neurons. We perform within-animal and across-animal comparisons on electrophysiological data from visual cortical areas V1, V4, and IT data, and use these as benchmarks to evaluate model-to-brain alignment. We also apply our method to reveal how object representations become progressively disentangled across layers in both biological and artificial systems. These findings underscore the importance of correcting feature-sampling biases in CKA and demonstrate that our bias-corrected estimator provides a more faithful measure of representation alignment. The improved estimates increase our understanding of how neural activity is structured across both biological and artificial systems.

Large language models (LLMs) have demonstrated remarkable proficiency in mainstream academic disciplines such as mathematics, physics, and computer science. However, human knowledge encompasses over 200 specialized disciplines, far exceeding the scope of existing benchmarks. The capabilities of LLMs in many of these specialized fields-particularly in light industry, agriculture, and service-oriented disciplines-remain inadequately evaluated. To address this gap, we present SuperGPQA, a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines. Our benchmark employs a novel Human-LLM collaborative filtering mechanism to eliminate trivial or ambiguous questions through iterative refinement based on both LLM responses and expert feedback. Our experimental results reveal significant room for improvement in the performance of current state-of-the-art LLMs across diverse knowledge domains (e.g., the reasoning-focused model DeepSeek-R1 achieved the highest accuracy of 61.82% on SuperGPQA), highlighting the considerable gap between current model capabilities and artificial general intelligence. Additionally, we present comprehensive insights from our management of a large-scale annotation process, involving over 80 expert annotators and an interactive Human-LLM collaborative system, offering valuable methodological guidance for future research initiatives of comparable scope.

We report results of magnetotransport measurements on KV$_3$Sb$_5$ thin flakes under pressure. Our zero-field electrical resistance reveals an additional anomaly emerging under pressure ($p$), marking a previously unidentified phase boundary $T^{\rm \ast}$($p$). Together with the established $T_{\rm CDW}(p)$ and $T_c(p)$, denoting the charge-density-wave transition and a superconducting transition, respectively, the temperature-pressure phase diagram of KV$_3$Sb$_5$ features a rich interplay among multiple phases. The Hall coefficient evolves reasonably smoothly when crossing the $T^{\rm \ast}$ phase boundary compared with the variation when crossing $T_{\rm CDW}$, indicating the preservation of the pristine electronic structure. The mobility spectrum analysis provides further insights into distinguishing different phases. Finally, our high-pressure quantum oscillation studies up to 31 T combined with density functional theory calculations further demonstrate that the new phase does not reconstruct the Fermi surface, confirming that the translational symmetry of the pristine metallic state is preserved., Comment: 10 pages, 5 figures. Advanced Science (2025)

Large Language Models (LLMs) have achieved remarkable success in natural language processing tasks, but their massive size and computational demands hinder their deployment in resource-constrained environments. Existing structured pruning methods address this issue by removing redundant structures (e.g., elements, channels, layers) from the model. However, these methods employ a heuristic pruning strategy, which leads to suboptimal performance. Besides, they also ignore the data characteristics when pruning the model. To overcome these limitations, we propose EvoP, an evolutionary pruning framework for robust LLM inference. EvoP first presents a cluster-based calibration dataset sampling (CCDS) strategy for creating a more diverse calibration dataset. EvoP then introduces an evolutionary pruning pattern searching (EPPS) method to find the optimal pruning pattern. Compared to existing structured pruning techniques, EvoP achieves the best performance while maintaining the best efficiency. Experiments across different LLMs and different downstream tasks validate the effectiveness of the proposed EvoP, making it a practical and scalable solution for deploying LLMs in real-world applications.

Quantum memory, serving as a crucial device for storing and releasing quantum states, holds significant importance in long-distance quantum communications. Up to date, quantum memories have been realized in many different systems. However, most of them have complex structures and high costs. Besides, it is not easy to simultaneously achieve both high storage efficiency and fidelity. In this paper, we experimentally demonstrate a low-cost optical quantum memory with high efficiency and high fidelity, by utilizing a butterfly shaped cavity consisting of one polarization beam splitter, two reflecting mirrors and one pockels cell crystal. In order to quantify the quality of the quantum memory, we carry out tomography measurements on the time-bin qubits encoded with weak coherent states after storage for N rounds. The storage efficiency per round can reach up to 95.0\%, and the overall state fidelity can exceed 99.1\%. It thus seems very promising for practical implementation in quantum communication and networks.

The superconducting order parameter at the KTaO3 interfaces and its dependence on interface orientation remains a subject of debate. The superconductivity at these interfaces exhibits strong resilience against in-plane magnetic field and violates Pauli limit. The interface orientation dependence of critical field and violation of Pauli limit, however, have not been investigated. To address this problem, we grew epitaxial LaMnO3/KTaO3 heterostructures using molecular beam epitaxy. We show that superconductivity is extremely robust against the in-plane magnetic field. Our results indicate that the interface orientation, despite impacting the critical temperature, does not affect the ratio of critical field to the Pauli limiting field. These results offer opportunities to engineer superconductors which are resilient against magnetic field.

Growing application memory demands and concurrent usage are making mobile device memory scarce. When memory pressure is high, current mobile systems use a RAM-based compressed swap scheme (called ZRAM) to compress unused execution-related data (called anonymous data in Linux) in main memory. We observe that the state-of-the-art ZRAM scheme prolongs relaunch latency and wastes CPU time because it does not differentiate between hot and cold data or leverage different compression chunk sizes and data locality. We make three new observations. 1) anonymous data has different levels of hotness. Hot data, used during application relaunch, is usually similar between consecutive relaunches. 2) when compressing the same amount of anonymous data, small-size compression is very fast, while large-size compression achieves a better compression ratio. 3) there is locality in data access during application relaunch. We propose Ariadne, a compressed swap scheme for mobile devices that reduces relaunch latency and CPU usage with three key techniques. 1) a low-overhead hotness-aware data organization scheme aims to quickly identify the hotness of anonymous data without significant overhead. 2) a size-adaptive compression scheme uses different compression chunk sizes based on the data's hotness level to ensure fast decompression of hot and warm data. 3) a proactive decompression scheme predicts the next set of data to be used and decompresses it in advance, reducing the impact of data swapping back into main memory during application relaunch. Our experimental evaluation results on Google Pixel 7 show that, on average, Ariadne reduces application relaunch latency by 50% and decreases the CPU usage of compression and decompression procedures by 15% compared to the state-of-the-art ZRAM scheme., Comment: This is an extended version of a paper that will appear in HPCA 2025

This paper discovers that the magnetic tunnel junction (MTJ) structure is successfully magnetized with hybrid microwave annealing, confirmed by the tunneling magnetoresistance (TMR) and Coercivity (Hc) results. Hybrid microwave annealing can transform CoFeB into a single crystal and form the Fe-O bond at the interface between CoFeB and MgO without adding an extra magnet. The annealing time is significantly reduced from the original 120 minutes to just 1 minute, allowing for rapid low-temperature annealing of the MTJ structure. The TEM results are used to determine the change in the lattice structure of CoFeB from amorphous to a single crystal, and the EELS result indicates the diffusion distribution of atoms in the MTJ structure. This hybrid annealing process can save a significant amount of fabrication time and is an energy-efficient alternative to the current fabrication process of MRAM., Comment: 3 pages, 5 figures

Twisted moir\'e supercells, which can be approximated as a combination of sliding bilayers and constitute various topologically nontrivial polarization patterns, attract extensive attention recently. However, because of the excessive size of the moir\'e supercell, most studies are based on effective models and lack the results of first-principles calculation. In this work, we use machine learning to determine the topological structure of the polarization pattern in twisted and strained bilayer of hexagonal boron nitride (h-BN). We further confirm that the topological pattern can be effectively modulated by the vertical electric field and lattice mismatch. Finally, local polarization also exists in the antiparallel stacked h-BN twisted and strained bilayers. Our work provides a detailed study of the polarization pattern in the moir\'e superlattice, which we believe can facilitate more research in moir\'e ferroelectricity, topological physics, and related fields.

Long context large language models (LLMs) pose significant challenges for efficient serving due to the large memory footprint and high access overhead of KV cache. Retrieval-based KV cache reduction methods can mitigate these challenges, typically by offloading the complete KV cache to CPU and retrieving necessary tokens on demand during inference. However, these methods still suffer from unsatisfactory accuracy degradation and extra retrieval overhead. To address these limitations, this paper proposes A$^2$ATS, a novel retrieval-based KV cache reduction method. A$^2$ATS aims to obtain an accurate approximation of attention scores by applying the vector quantization technique to key states, thereby enabling efficient and precise retrieval of the top-K tokens. First, we propose Windowed Rotary Position Embedding, which decouples the positional dependency from query and key states after position embedding. Then, we propose query-aware vector quantization that optimizes the objective of attention score approximation directly. Finally, we design the heterogeneous inference architecture for KV cache offloading, enabling long context serving with larger batch sizes. Experimental results demonstrate that A$^2$ATS can achieve a lower performance degradation with similar or lower overhead compared to existing methods, thereby increasing long context serving throughput by up to $2.7 \times$.

Graph Neural Networks (GNNs) have become a prominent approach for learning from graph-structured data. However, their effectiveness can be significantly compromised when the graph structure is suboptimal. To address this issue, Graph Structure Learning (GSL) has emerged as a promising technique that refines node connections adaptively. Nevertheless, we identify two key limitations in existing GSL methods: 1) Most methods primarily focus on node similarity to construct relationships, while overlooking the quality of node information. Blindly connecting low-quality nodes and aggregating their ambiguous information can degrade the performance of other nodes. 2) The constructed graph structures are often constrained to be symmetric, which may limit the model's flexibility and effectiveness. To overcome these limitations, we propose an Uncertainty-aware Graph Structure Learning (UnGSL) strategy. UnGSL estimates the uncertainty of node information and utilizes it to adjust the strength of directional connections, where the influence of nodes with high uncertainty is adaptively reduced. Importantly, UnGSL serves as a plug-in module that can be seamlessly integrated into existing GSL methods with minimal additional computational cost. In our experiments, we implement UnGSL into six representative GSL methods, demonstrating consistent performance improvements., Comment: This paper has been accepted by TheWebConf 2025

We report direct measurement of the dispersion relation of polaritons in GaAs/AlGaAs microcavity structures at room temperature, which clearly shows that the polaritons are in the strong coupling limit. The Rabi splitting of the polariton states decreases as the polariton gas increases in density, but even when the polariton gas becomes a coherent, Bose-condensate-like state, the polaritons retain a strong exciton component, as seen in the nonlinear energy shift of the light emission. This opens up the possibility of polaritonic devices at room temperature in a material system which can be grown with very high quality and uniformity.

We point out that specifying the finite modular group does not uniquely fix a modular flavor symmetry. We illustrate this using the finite modular group $T'$. Otherwise equivalent models based on different $T'$ lead to modular forms with different properties and, hence, produce different phenomenological features. We exemplify this in various scenarios, and show that the ability of a given model to accommodate mass and other observed hierarchies depends sensitively on the way the $T'$ is implemented., Comment: 26 pages

Understanding the existence of exotic matter phases and phase transitions within the core of neutron stars is crucial to advancing our knowledge of cold-dense matter physics. Recent multi-messenger observations, including gravitational waves from neutron star mergers and precise X-ray data from NASA's Neutron Star Interior Composition Explorer (NICER) mission, have significantly constrained the neutron star equation of state (EOS). This study investigates the effects of phase transitions in neutron stars, focusing on NICER's latest observation of PSR J0437-4715. We employ Bayesian inference techniques to evaluate the presence of first-order phase transitions using a piecewise polytropic EOS model. Our analysis incorporates data from multiple NICER sources, to refine constraints on key phase transition parameters, including critical density and transition depth. We find that including data from PSR J0437-4715 improves the evidence of phase transitions and tightens the EOS constraints, especially at higher densities. However, Bayes factor analysis only indicates a slight preference for models without phase transitions and current observational precision is insufficient to draw definitive conclusions. In particular, this polytropic model identifies the critical phase transition mass of neutron stars as being close to 1.4 solar masses, which coincides with the approximate mass range of PSR J0437-4715. This work emphasizes the importance of precise measurements of PSR J0437-4715 for deepening our understanding of neutron star interiors and exploring potential new physics at extreme densities., Comment: Submitted to ApJ

Recent multimodal large language models (MLLMs) have demonstrated significant potential in open-ended conversation, generating more accurate and personalized responses. However, their abilities to memorize, recall, and reason in sustained interactions within real-world scenarios remain underexplored. This paper introduces MMRC, a Multi-Modal Real-world Conversation benchmark for evaluating six core open-ended abilities of MLLMs: information extraction, multi-turn reasoning, information update, image management, memory recall, and answer refusal. With data collected from real-world scenarios, MMRC comprises 5,120 conversations and 28,720 corresponding manually labeled questions, posing a significant challenge to existing MLLMs. Evaluations on 20 MLLMs in MMRC indicate an accuracy drop during open-ended interactions. We identify four common failure patterns: long-term memory degradation, inadequacies in updating factual knowledge, accumulated assumption of error propagation, and reluctance to say no. To mitigate these issues, we propose a simple yet effective NOTE-TAKING strategy, which can record key information from the conversation and remind the model during its responses, enhancing conversational capabilities. Experiments across six MLLMs demonstrate significant performance improvements.

The anomalous Hall {\color{black} e}ffect (AHE) has emerged as a key indicator of time-reversal symmetry breaking (TRSB) and topological features in electronic band structures. Absent of a magnetic field, the AHE requires spontaneous TRSB but has proven hard to probe due to averaging over domains. The anomalous component of the Hall effect is thus frequently derived from extrapolating the magnetic field dependence of the Hall response. We show that discerning whether the AHE is an intrinsic property of the field free system becomes intricate in the presence of strong magnetic fluctuations. {\color{black}As a study case,} we use the Weyl semimetal PrAlGe, where TRSB can be toggled via a ferromagnetic transition, providing a transparent view of the AHE's topological origin. Through a combination of thermodynamic, transport and muon spin relaxation measurements, we contrast the behaviour below the ferromagnetic transition temperature to that of strong magnetic fluctuations above. Our results {\color{black}on PrAlGe provide general insights into the} interpretation of anomalous Hall signals in systems where TRSB is debated, such as families of Kagome metals or certain transition metal dichalcogenides., Comment: 8 pages, 5 figures

Offline model-based reinforcement learning (MBRL) serves as a competitive framework that can learn well-performing policies solely from pre-collected data with the help of learned dynamics models. To fully unleash the power of offline MBRL, model selection plays a pivotal role in determining the dynamics model utilized for downstream policy learning. However, offline MBRL conventionally relies on validation or off-policy evaluation, which are rather inaccurate due to the inherent distribution shift in offline RL. To tackle this, we propose BOMS, an active model selection framework that enhances model selection in offline MBRL with only a small online interaction budget, through the lens of Bayesian optimization (BO). Specifically, we recast model selection as BO and enable probabilistic inference in BOMS by proposing a novel model-induced kernel, which is theoretically grounded and computationally efficient. Through extensive experiments, we show that BOMS improves over the baseline methods with a small amount of online interaction comparable to only $1\%$-$2.5\%$ of offline training data on various RL tasks., Comment: 18 pages, 6 figures

Resource limitations often constrain the parameter counts of Large Language Models (LLMs), hindering their performance. While existing methods employ parameter sharing to reuse the same parameter set under fixed budgets, such approaches typically force each layer to assume multiple roles with a predetermined number of iterations, restricting efficiency and adaptability. In this work, we propose the Zero Token Transformer (ZTT), which features a head-tail decoupled parameter cycling method. We disentangle the first (head) and last (tail) layers from parameter cycling and iteratively refine only the intermediate layers. Furthermore, we introduce a Zero-Token Mechanism, an internal architectural component rather than an input token, to guide layer-specific computation. At each cycle, the model retrieves a zero token (with trainable key values) from a Zero-Token Pool, integrating it alongside regular tokens in the attention mechanism. The corresponding attention scores not only reflect each layer's computational importance but also enable dynamic early exits without sacrificing overall model accuracy. Our approach achieves superior performance under tight parameter budgets, effectively reduces computational overhead via early exits, and can be readily applied to fine-tune existing pre-trained models for enhanced efficiency and adaptability.

This work presents a novel Bayesian framework for unsupervised domain adaptation (UDA) in medical image segmentation. While prior works have explored this clinically significant task using various strategies of domain alignment, they often lack an explicit and explainable mechanism to ensure that target image features capture meaningful structural information. Besides, these methods are prone to the curse of dimensionality, inevitably leading to challenges in interpretability and computational efficiency. To address these limitations, we propose RemInD, a framework inspired by human adaptation. RemInD learns a domain-agnostic latent manifold, characterized by several anchors, to memorize anatomical variations. By mapping images onto this manifold as weighted anchor averages, our approach ensures realistic and reliable predictions. This design mirrors how humans develop representative components to understand images and then retrieve component combinations from memory to guide segmentation. Notably, model prediction is determined by two explainable factors: a low-dimensional anchor weight vector, and a spatial deformation. This design facilitates computationally efficient and geometry-adherent adaptation by aligning weight vectors between domains on a probability simplex. Experiments on two public datasets, encompassing cardiac and abdominal imaging, demonstrate the superiority of RemInD, which achieves state-of-the-art performance using a single alignment approach, outperforming existing methods that often rely on multiple complex alignment strategies., Comment: Accepted by IPMI 2025 (Information Processing in Medical Imaging)

Recurrent events, characterized by the repeated occurrence of the same event in an individual, are a common type of data in medical research. Motivated by cancer recurrences, we aim to estimate the optimal individualized treatment regime (ITR) that effectively mitigates such recurrent events. An ITR is a decision rule that assigns the optimal treatment to each patient, based on personalized information, with the aim of maximizing the overall therapeutic benefits. However, existing studies of estimating ITR mainly focus on first-time events rather than recurrent events. To address the issue of determining the optimal ITR for recurrent events, we propose the Recurrent C-learning (ReCL) method to identify the optimal ITR from two or multiple treatment options. The proposed method reformulates the optimization problem into a weighted classification problem. We introduce three estimators for the misclassification cost: the outcome regression estimator, the inverse probability weighting estimator, and the augmented inverse probability weighting estimator. The ReCL method leverages classification techniques to generate an interpretable optimal ITR tailored for recurrent event data. The advantages of the ReCL method are demonstrated through simulations under various scenarios. Furthermore, based on real data on colorectal cancer treatments, we employ this novel method to derive interpretable tree treatment regimes for colorectal cancer, thus providing a practical framework for enhancing treatment strategies., Comment: 22 pages, 4 figures

Learned indexes are promising to replace traditional tree-based indexes. They typically employ machine learning models to efficiently predict target positions in strictly sorted linear arrays. However, the strict sorted order 1) significantly increases insertion overhead, 2) makes it challenging to support lock-free concurrency, and 3) harms in-node lookup/insertion efficiency due to model inaccuracy.\ In this paper, we introduce a \textit{Bucket-based Learned Index (BLI)}, which is an updatable in-memory learned index that adopts a "globally sorted, locally unsorted" approach by replacing linear sorted arrays with \textit{Buckets}. BLI optimizes the insertion throughput by only sorting Buckets, not the key-value pairs within a Bucket. BLI strategically balances three critical performance metrics: tree fanouts, lookup/insert latency for inner nodes, lookup/insert latency for leaf nodes, and memory consumption. To minimize maintenance costs, BLI performs lightweight bulk loading, insert, node scaling, node split, model retraining, and node merging adaptively. BLI supports lock-free concurrency thanks to the unsorted design with Buckets. Our results show that BLI achieves up to 2.21x better throughput than state-of-the-art learned indexes, with up to 3.91x gains under multi-threaded conditions.

This paper presents a hybrid approach that integrates Large Language Models (LLMs) with a multi-scenario Stochastic Unit Commitment (SUC) framework, focusing on both efficiency and reliability under high wind generation uncertainties. Numerical experiments on small-to-medium-sized test systems show that while the traditional SUC approach yields a total cost of 99.05 million USD with 3.04 GWh of load curtailment, the LLM-assisted SUC (LLM-SUC) reduces costs to 98.87 million USD and lowers load curtailment to 2.32 GWh, an improvement of nearly 24%. Both methods maintain zero wind curtailment, confirming robust renewable integration. By employing an LLM agent that helps balance the energy system more effectively, the proposed framework enhances demand fulfillment at reduced costs, illustrating the potential of AI to inform generator commitments in uncertain operating conditions. Further gains may be realized by refining prompt design, incorporating historical operational data, and extending this approach to higher-dimensional uncertainties and energy storage systems, ultimately fostering greater resilience, efficiency, and adaptability in next-generation power system operations.

In this paper, we investigate the fundamental tradeoff between communication and sensing performance of \emph{ad hoc} integrated sensing and communication (ISAC) wireless networks. Specifically, we consider that $n$ nodes are randomly located in an extended network with area $n$ and transmit ISAC signals. Under the pure path loss channel gain model and the condition that the transmission power scales according to the communication distance, we fully characterize the optimal scaling law tradeoff between throughput and sensing distance by proposing an achievable scheme and proving its converse. Our results can be interpreted as follows: by reducing the throughput by a factor of a function of $n$, the sensing range order improves according to the same function of $n$, raised to the power of the ratio between the path loss factors in communication and sensing. We prove that the same result also holds true for ISAC networks with random fading, despite the uncertainty on the connectivity and power level created by random fading. In addition, we show that the scaling law tradeoff cannot be improved by allowing the transmission power and communication distance to scale freely. To the best of our knowledge, this is the first work formally formulating and characterizing the communication and sensing performance scaling law tradeoff of \emph{ad hoc} ISAC networks.

3D Gaussian Splatting (3DGS) has been widely used in 3D reconstruction and 3D generation. Training to get a 3DGS scene often takes a lot of time and resources and even valuable inspiration. The increasing amount of 3DGS digital asset have brought great challenges to the copyright protection. However, it still lacks profound exploration targeted at 3DGS. In this paper, we propose a new framework X-SG$^2$S which can simultaneously watermark 1 to 3D messages while keeping the original 3DGS scene almost unchanged. Generally, we have a X-SG$^2$S injector for adding multi-modal messages simultaneously and an extractor for extract them. Specifically, we first split the watermarks into message patches in a fixed manner and sort the 3DGS points. A self-adaption gate is used to pick out suitable location for watermarking. Then use a XD(multi-dimension)-injection heads to add multi-modal messages into sorted 3DGS points. A learnable gate can recognize the location with extra messages and XD-extraction heads can restore hidden messages from the location recommended by the learnable gate. Extensive experiments demonstrated that the proposed X-SG$^2$S can effectively conceal multi modal messages without changing pretrained 3DGS pipeline or the original form of 3DGS parameters. Meanwhile, with simple and efficient model structure and high practicality, X-SG$^2$S still shows good performance in hiding and extracting multi-modal inner structured or unstructured messages. X-SG$^2$S is the first to unify 1 to 3D watermarking model for 3DGS and the first framework to add multi-modal watermarks simultaneous in one 3DGS which pave the wave for later researches.

We explore the energy spectrum and eigenstates of two-component atomic Fermi superfluids with tunable pairing interactions in the presence of spin-dependent ultra long-range Rydberg molecule (ULRM) potentials, within the Bogoliubov-de Gennes formalism. The attractive ULRM potentials lead to local density accumulation, while their difference results in a local polarization potential and induces the in-gap Yu-Shiba-Rusinov (YSR) states whose energies lie below the bulk energy gap. A transition from equal-population to population-imbalance occurs as the pairing strength falls below a critical value, accompanied by the emergence of local Fulde-Ferrell-Larkin-Ovchinnikov (FFLO) like states characterized by out-of-phase wave functions and lower energies compared to the YSR states. The negative contribution emanating from the FFLO-like states also causes a sign change in the gap function within the ULRM potentials. Depending on the Rydberg state generating the ULRM potentials, the transition towards population-imbalance can be on either the BCS or the Bose-Einstein condensation side of the Fermi superfluid. Additionally, spin-polarized bound states arise along with oscillatory ``clumpy states" to compensate for the local density difference. Finally, we discuss possible experimental realizations and measurements of the composite Rydberg atom-Fermi superfluid system., Comment: 12 pages, 7 figures, submitted

We prove that the Fourier dimension of the graph of fractional Brownian motion with Hurst index greater than $1/2$ is almost surely 1. This extends the result of Fraser and Sahlsten (2018) for the Brownian motion and verifies partly the conjecture of Fraser, Orponen and Sahlsten (2014). We introduce a combinatorial integration by parts formula to compute the moments of the Fourier transform of the graph measure. The proof of our main result is based on this integration by parts formula together with Fa\`a di Bruno's formula and strong local nondeterminism of fractional Brownian motion. We also show that the Fourier dimension of the graph of a symmetric $\alpha$-stable process with $\alpha\in[1,2]$ is almost surely 1.

The field of music generation using Large Language Models (LLMs) is evolving rapidly, yet existing music notation systems, such as MIDI, ABC Notation, and MusicXML, remain too complex for effective fine-tuning of LLMs. These formats are difficult for both machines and humans to interpret due to their variability and intricate structure. To address these challenges, we introduce YNote, a simplified music notation system that uses only four characters to represent a note and its pitch. YNote's fixed format ensures consistency, making it easy to read and more suitable for fine-tuning LLMs. In our experiments, we fine-tuned GPT-2 (124M) on a YNote-encoded dataset and achieved BLEU and ROUGE scores of 0.883 and 0.766, respectively. With just two notes as prompts, the model was able to generate coherent and stylistically relevant music. We believe YNote offers a practical alternative to existing music notations for machine learning applications and has the potential to significantly enhance the quality of music generation using LLMs.