Search

Sharpness-Aware Minimization (SAM) has substantially improved the generalization of neural networks under various settings. Despite the success, its effectiveness remains poorly understood. In this work, we discover an intriguing phenomenon in the training dynamics of SAM, shedding lights on understanding its implicit bias towards flatter minima over Stochastic Gradient Descent (SGD). Specifically, we find that SAM efficiently selects flatter minima late in training. Remarkably, even a few epochs of SAM applied at the end of training yield nearly the same generalization and solution sharpness as full SAM training. Subsequently, we delve deeper into the underlying mechanism behind this phenomenon. Theoretically, we identify two phases in the learning dynamics after applying SAM late in training: i) SAM first escapes the minimum found by SGD exponentially fast; and ii) then rapidly converges to a flatter minimum within the same valley. Furthermore, we empirically investigate the role of SAM during the early training phase. We conjecture that the optimization method chosen in the late phase is more crucial in shaping the final solution's properties. Based on this viewpoint, we extend our findings from SAM to Adversarial Training., Comment: 24 pages, 9 figures

Language is not monolithic. While many benchmarks are used as proxies to systematically estimate Large Language Models' (LLM) performance in real-life tasks, they tend to ignore the nuances of within-language variation and thus fail to model the experience of speakers of minority dialects. Focusing on African American Vernacular English (AAVE), we present the first study on LLMs' fairness and robustness to a dialect in canonical reasoning tasks (algorithm, math, logic, and comprehensive reasoning). We hire AAVE speakers, including experts with computer science backgrounds, to rewrite seven popular benchmarks, such as HumanEval and GSM8K. The result of this effort is ReDial, a dialectal benchmark comprising $1.2K+$ parallel query pairs in Standardized English and AAVE. We use ReDial to evaluate state-of-the-art LLMs, including GPT-4o/4/3.5-turbo, LLaMA-3.1/3, Mistral, and Phi-3. We find that, compared to Standardized English, almost all of these widely used models show significant brittleness and unfairness to queries in AAVE. Furthermore, AAVE queries can degrade performance more substantially than misspelled texts in Standardized English, even when LLMs are more familiar with the AAVE queries. Finally, asking models to rephrase questions in Standardized English does not close the performance gap but generally introduces higher costs. Overall, our findings indicate that LLMs provide unfair service to dialect users in complex reasoning tasks. Code can be found at https://github.com/fangru-lin/redial_dialect_robustness_fairness.git.

With the rapid advances in diffusion models, generating decent images from text prompts is no longer challenging. The key to text-to-image generation is how to optimize the results of a text-to-image generation model so that they can be better aligned with human intentions or prompts. Existing optimization methods commonly treat the entire image uniformly and conduct global optimization. These methods overlook the fact that when viewing an image, the human visual system naturally prioritizes attention toward salient areas, often neglecting less or non-salient regions. That is, humans are likely to neglect optimizations in non-salient areas. Consequently, although model retaining is conducted under the guidance of additional large and multimodality models, existing methods, which perform uniform optimizations, yield sub-optimal results. To address this alignment challenge effectively and efficiently, we propose Saliency Guided Optimization Of Diffusion Latents (SGOOL). We first employ a saliency detector to mimic the human visual attention system and mark out the salient regions. To avoid retraining an additional model, our method directly optimizes the diffusion latents. Besides, SGOOL utilizes an invertible diffusion process and endows it with the merits of constant memory implementation. Hence, our method becomes a parameter-efficient and plug-and-play fine-tuning method. Extensive experiments have been done with several metrics and human evaluation. Experimental results demonstrate the superiority of SGOOL in image quality and prompt alignment.

As large language models rapidly evolve to support longer context, there is a notable disparity in their capability to generate output at greater lengths. Recent study suggests that the primary cause for this imbalance may arise from the lack of data with long-output during alignment training. In light of this observation, attempts are made to re-align foundation models with data that fills the gap, which result in models capable of generating lengthy output when instructed. In this paper, we explore the impact of data-quality in tuning a model for long output, and the possibility of doing so from the starting points of human-aligned (instruct or chat) models. With careful data curation, we show that it possible to achieve similar performance improvement in our tuned models, with only a small fraction of training data instances and compute. In addition, we assess the generalizability of such approaches by applying our tuning-recipes to several models. our findings suggest that, while capacities for generating long output vary across different models out-of-the-box, our approach to tune them with high-quality data using lite compute, consistently yields notable improvement across all models we experimented on. We have made public our curated dataset for tuning long-writing capability, the implementations of model tuning and evaluation, as well as the fine-tuned models, all of which can be openly-accessed.

Fast radio bursts (FRBs) are transient radio bursts of extragalactic origin characterized by millisecond durations and high luminosities. We report on observations of FRB 20240114A conducted with the Robert C. Byrd Green Bank Telescope (GBT) at frequencies ranging from 720 to 920 MHz. A total of 429 bursts were detected, with a single observation recording 359 bursts over 1.38 hours, corresponding to a burst rate of 260 bursts per hour. The average rotation measures (RMs) were $349.2 \pm 1.0$ rad m$^{-2}$ on February 23, 2024, and $360.4 \pm 0.4$ rad m$^{-2}$ on March 1, 2024. Of the 297 bursts with detected RMs, 72% have a linear polarization fraction greater than 90%, and 14% exhibit circular polarization with a signal-to-noise ratio $> 5$. Our sample also displayed polarization angle swings. We compare the linear polarization of FRB 20240114A with that of FRB 20201124A, FRB 20220912A, and non-repeating FRBs. The mean linear polarization fraction for non-repeating FRBs is 58%. In contrast, the mean linear polarization fraction for the three repeating FRBs is 94%, which is significantly higher than that of the non-repeating FRBs. Under the T-test, the three repeating FRBs have similar linear polarization distributions, but these distributions differ from those of the non-repeating FRBs. This suggests that non-repeating FRBs may have different emission mechanisms or are subject to depolarization., Comment: 42 pages, 5 figures. arXiv admin note: text overlap with arXiv:2304.14671

Whole Slide Image (WSI) classification has very significant applications in clinical pathology, e.g., tumor identification and cancer diagnosis. Currently, most research attention is focused on Multiple Instance Learning (MIL) using static datasets. One of the most obvious weaknesses of these methods is that they cannot efficiently preserve and utilize previously learned knowledge. With any new data arriving, classification models are required to be re-trained on both previous and current new data. To overcome this shortcoming and break through traditional vision modality, this paper proposes the first Vision-Language-based framework with Queryable Prototype Multiple Instance Learning (QPMIL-VL) specially designed for incremental WSI classification. This framework mainly consists of two information processing branches. One is for generating the bag-level feature by prototype-guided aggregating on the instance features. While the other is for enhancing the class feature through class ensemble, tunable vector and class similarity loss. The experiments on four TCGA datasets demonstrate that our QPMIL-VL framework is effective for incremental WSI classification and often significantly outperforms other compared methods, achieving state-of-the-art (SOTA) performance., Comment: 16 pages, 10 tables, 11 figures

Previous research has explored the computational expressivity of Transformer models in simulating Boolean circuits or Turing machines. However, the learnability of these simulators from observational data has remained an open question. Our study addresses this gap by providing the first polynomial-time learnability results (specifically strong, agnostic PAC learning) for single-layer Transformers with linear attention. We show that linear attention may be viewed as a linear predictor in a suitably defined RKHS. As a consequence, the problem of learning any linear transformer may be converted into the problem of learning an ordinary linear predictor in an expanded feature space, and any such predictor may be converted back into a multiheaded linear transformer. Moving to generalization, we show how to efficiently identify training datasets for which every empirical risk minimizer is equivalent (up to trivial symmetries) to the linear Transformer that generated the data, thereby guaranteeing the learned model will correctly generalize across all inputs. Finally, we provide examples of computations expressible via linear attention and therefore polynomial-time learnable, including associative memories, finite automata, and a class of Universal Turing Machine (UTMs) with polynomially bounded computation histories. We empirically validate our theoretical findings on three tasks: learning random linear attention networks, key--value associations, and learning to execute finite automata. Our findings bridge a critical gap between theoretical expressivity and learnability of Transformers, and show that flexible and general models of computation are efficiently learnable.

The form of the entanglement Hamiltonian varies with the parameters of the original system. Whether there is a singularity is the key problem for demonstrating/negating the universality of the relation between the entanglement spectrum and edge energy spectrum. We carefully study the phase diagram of a 1D Ising entanglement Hamiltonian as an example to clarify the long-standing controversy of the general relation between the entanglement Hamiltonian and original Hamiltonian. Interestingly, even if the singularities indeed exist, the Li-Haldane-Poilblanc conjecture, i.e., the general relation between the entanglement spectrum and edge energy spectrum, seemingly still holds., Comment: 13 pages, 14 figures

Liver cancer is a leading cause of mortality worldwide, and accurate CT-based tumor segmentation is essential for diagnosis and treatment. Manual delineation is time-intensive, prone to variability, and highlights the need for reliable automation. While deep learning has shown promise for automated liver segmentation, precise liver tumor segmentation remains challenging due to the heterogeneous nature of tumors, imprecise tumor margins, and limited labeled data. We present a novel holistic weakly supervised framework that integrates clinical knowledge to address these challenges with (1) A knowledge-informed label smoothing technique that leverages clinical data to generate smooth labels, which regularizes model training reducing the risk of overfitting and enhancing model performance; (2) A global and local-view segmentation framework, breaking down the task into two simpler sub-tasks, allowing optimized preprocessing and training for each; and (3) Pre- and post-processing pipelines customized to the challenges of each subtask, which enhances tumor visibility and refines tumor boundaries. We evaluated the proposed method on the HCC-TACE-Seg dataset and showed that these three key components complementarily contribute to the improved performance. Lastly, we prototyped a tool for automated liver tumor segmentation and diagnosis summary generation called MedAssistLiver. The app and code are published at https://github.com/lingchm/medassist-liver-cancer.

As telecommunications systems progress to support diverse use cases with heterogeneous and dynamic Quality of Service (QoS) requirements, it becomes an increasingly complex task to automatically manage various resources involved -- from radio, compute, to X-haul network, which are distributed from the edge to the cloud. Intent-driven network automation can play an important role in NextG networks to meet this need. Towards this, we have developed INA-Infra, an open, extensible, and end-to-end 5G/beyond 5G network infrastructure with intent-driven network automation and end-to-end network slicing capability. INA-Infra is designed using open-source components and is based on O-RAN architecture. INA-Infra manages the network infrastructure, various resources, and (virtualized / containerized) network functions using Nephio -- a cloud-native intent automation solution. It also incorporates intent-driven intelligent control using a Resource Management rApp and a Network Slicing xApp. We demonstrate that INA-Infra can manage the 5G network in a highly automatic and optimized manner, allowing the mobile network operators to focus on specifying the intents of different traffic classes., Comment: Preprint version, published at workshop OpenRIT-6G, part of IEEE GLOBECOM 2024

Event extraction has gained extensive research attention due to its broad range of applications. However, the current mainstream evaluation method for event extraction relies on token-level exact match, which misjudges numerous semantic-level correct cases. This reliance leads to a significant discrepancy between the evaluated performance of models under exact match criteria and their real performance. To address this problem, we propose RAEE, an automatic evaluation framework that accurately assesses event extraction results at semantic-level instead of token-level. Specifically, RAEE leverages Large Language Models (LLMs) as automatic evaluation agents, incorporating chain-of-thought prompting and an adaptive mechanism to achieve interpretable and adaptive evaluations for precision and recall of triggers and arguments. Extensive experimental results demonstrate that: (1) RAEE achieves a very high correlation with the human average; (2) after reassessing 14 models, including advanced LLMs, on 10 datasets, there is a significant performance gap between exact match and RAEE. The exact match evaluation significantly underestimates the performance of existing event extraction models, particularly underestimating the capabilities of LLMs; (3) fine-grained analysis under RAEE evaluation reveals insightful phenomena worth further exploration. The evaluation toolkit of our proposed RAEE will be publicly released.

We generalize several known results on small Simpson correspondence for smooth formal schemes over $\calO_C$ to the case for semi-stable formal schemes. More precisely, for a liftable semi-stable formal scheme $\frakX$ over $\calO_C$ with generic fiber $X$, we establish (1) an equivalence between the category of Hitchin-small integral $v$-bundles on $X_{v}$ and the category of Hitchin-small Higgs bundles on $\frakX_{\et}$, generalizing the previous work of Min--Wang, and (2) an equivalence between the moduli stack of $v$-bundles on $X_{v}$ and the moduli stack of rational Higgs bundles on $\frakX_{\et}$ (equivalently, moduli stack of Higgs bundles on $X_{\et}$), generalizing the previous work of Ansch\"utz--Heuer--Le Bras., Comment: Comments are welcome!

We present a general theory of potentials that support bound states at positive energies (bound states in the continuum). On the theoretical side, we prove that, for systems described by nonlocal potentials of the form $V(r,r')$, bound states at positive energies are as common as those at negative energies. At the same time, we show that a local potential of the form $V(r)$ rarely supports a positive energy bound state. On the practical side, we show how to construct a (naturally nonlocal) potential which supports an arbitrary normalizable state at an arbitrary positive energy. We demonstrate our theory with numerical examples both in momentum and coordinate spaces with emphasis on the important role played by nonlocal potentials. Finally, we discuss how to observe bound states at positive energies, and where to search for nonlocal potentials which may support them., Comment: 32 pages, 22 figures

Embodied artificial intelligence emphasizes the role of an agent's body in generating human-like behaviors. The recent efforts on EmbodiedAI pay a lot of attention to building up machine learning models to possess perceiving, planning, and acting abilities, thereby enabling real-time interaction with the world. However, most works focus on bounded indoor environments, such as navigation in a room or manipulating a device, with limited exploration of embodying the agents in open-world scenarios. That is, embodied intelligence in the open and outdoor environment is less explored, for which one potential reason is the lack of high-quality simulators, benchmarks, and datasets. To address it, in this paper, we construct a benchmark platform for embodied intelligence evaluation in real-world city environments. Specifically, we first construct a highly realistic 3D simulation environment based on the real buildings, roads, and other elements in a real city. In this environment, we combine historically collected data and simulation algorithms to conduct simulations of pedestrian and vehicle flows with high fidelity. Further, we designed a set of evaluation tasks covering different EmbodiedAI abilities. Moreover, we provide a complete set of input and output interfaces for access, enabling embodied agents to easily take task requirements and current environmental observations as input and then make decisions and obtain performance evaluations. On the one hand, it expands the capability of existing embodied intelligence to higher levels. On the other hand, it has a higher practical value in the real world and can support more potential applications for artificial general intelligence. Based on this platform, we evaluate some popular large language models for embodied intelligence capabilities of different dimensions and difficulties., Comment: All of the software, Python library, codes, datasets, tutorials, and real-time online service are available on this website: https://embodied-city.fiblab.net

Predicting the change in binding free energy ($\Delta \Delta G$) is crucial for understanding and modulating protein-protein interactions, which are critical in drug design. Due to the scarcity of experimental $\Delta \Delta G$ data, existing methods focus on pre-training, while neglecting the importance of alignment. In this work, we propose the Boltzmann Alignment technique to transfer knowledge from pre-trained inverse folding models to $\Delta \Delta G$ prediction. We begin by analyzing the thermodynamic definition of $\Delta \Delta G$ and introducing the Boltzmann distribution to connect energy with protein conformational distribution. However, the protein conformational distribution is intractable; therefore, we employ Bayes' theorem to circumvent direct estimation and instead utilize the log-likelihood provided by protein inverse folding models for $\Delta \Delta G$ estimation. Compared to previous inverse folding-based methods, our method explicitly accounts for the unbound state of protein complex in the $\Delta \Delta G$ thermodynamic cycle, introducing a physical inductive bias and achieving both supervised and unsupervised state-of-the-art (SoTA) performance. Experimental results on SKEMPI v2 indicate that our method achieves Spearman coefficients of 0.3201 (unsupervised) and 0.5134 (supervised), significantly surpassing the previously reported SoTA values of 0.2632 and 0.4324, respectively. Futhermore, we demonstrate the capability of our method on binding energy prediction, protein-protein docking and antibody optimization tasks.

Large language models (LLMs) have gained human trust due to their capabilities and helpfulness. However, this in turn may allow LLMs to affect users' mindsets by manipulating language. It is termed as gaslighting, a psychological effect. In this work, we aim to investigate the vulnerability of LLMs under prompt-based and fine-tuning-based gaslighting attacks. Therefore, we propose a two-stage framework DeepCoG designed to: 1) elicit gaslighting plans from LLMs with the proposed DeepGaslighting prompting template, and 2) acquire gaslighting conversations from LLMs through our Chain-of-Gaslighting method. The gaslighting conversation dataset along with a corresponding safe dataset is applied to fine-tuning-based attacks on open-source LLMs and anti-gaslighting safety alignment on these LLMs. Experiments demonstrate that both prompt-based and fine-tuning-based attacks transform three open-source LLMs into gaslighters. In contrast, we advanced three safety alignment strategies to strengthen (by 12.05%) the safety guardrail of LLMs. Our safety alignment strategies have minimal impacts on the utility of LLMs. Empirical studies indicate that an LLM may be a potential gaslighter, even if it passed the harmfulness test on general dangerous queries., Comment: 10/26 (Main Body/Total), 8 figures

Task-oriented grasping, which involves grasping specific parts of objects based on their functions, is crucial for developing advanced robotic systems capable of performing complex tasks in dynamic environments. In this paper, we propose a training-free framework that incorporates both semantic and geometric priors for zero-shot task-oriented grasp generation. The proposed framework, SegGrasp, first leverages the vision-language models like GLIP for coarse segmentation. It then uses detailed geometric information from convex decomposition to improve segmentation quality through a fusion policy named GeoFusion. An effective grasp pose can be generated by a grasping network with improved segmentation. We conducted the experiments on both segmentation benchmark and real-world robot grasping. The experimental results show that SegGrasp surpasses the baseline by more than 15\% in grasp and segmentation performance., Comment: 7pages,6 figures

For the High-Luminosity (HL-LHC) phase, the upgrade of the Compact Muon Solenoid (CMS) experiment at CERN will include a novel MIP Timing Detector (MTD). The central part of MTD, the barrel timing layer (BTL), is designed to provide a measurement of the time of arrival of charged particles with a precision of 30 ps at the beginning of HL-LHC, progressively degrading to 60 ps while operating in an extremely harsh radiation environment for over a decade. In this paper we present a comparative analysis of the time resolution of BTL module prototypes made of LYSO:Ce crystal bars read out by silicon photo-multipliers (SiPMs). The timing performance measured in beam test campaigns is presented for prototypes with different construction and operation parameters, such as different SiPM cell sizes (15, 20, 25 and 30 $\rm \mu m$), SiPM manufacturers and crystal bar thicknesses. The evolution of time resolution as a function of the irradiation level has been studied using non-irradiated SiPMs as well as SiPMs exposed up to $2\times 10^{14}~n_{eq}/cm^2$ fluence. The key parameters defining the module time resolution such as SiPM characteristics (gain, photon detection efficiency, radiation induced dark count rate) and crystal properties (light output and dimensions) are discussed. These results have informed the final choice of the MTD barrel sensor configuration and offer a unique starting point for the design of future large-area scintillator-based timing detectors in either low or high radiation environments.

Research on autonomous driving in unstructured outdoor environments is less advanced than in structured urban settings due to challenges like environmental diversities and scene complexity. These environments-such as rural areas and rugged terrains-pose unique obstacles that are not common in structured urban areas. Despite these difficulties, autonomous driving in unstructured outdoor environments is crucial for applications in agriculture, mining, and military operations. Our survey reviews over 250 papers for autonomous driving in unstructured outdoor environments, covering offline mapping, pose estimation, environmental perception, path planning, end-to-end autonomous driving, datasets, and relevant challenges. We also discuss emerging trends and future research directions. This review aims to consolidate knowledge and encourage further research for autonomous driving in unstructured environments. To support ongoing work, we maintain an active repository with up-to-date literature and open-source projects at: https://github.com/chaytonmin/Survey-Autonomous-Driving-in-Unstructured-Environments., Comment: Survey paper; 38 pages

We investigate the gravitational wave spectrum induced by first-order QCD phase transitions including the deconfinement phase transition in the pure gluon system and Friedberg-Lee model, and chiral phase transition in the quark-meson model and Polyakov quark-meson model. The gravitational wave power spectra are sensitive to the phase transition rate $\beta/H$. All QCD models predict a rather large phase transition rate in the order of $\beta/H\sim10^4$ at high temperature region, and the produced gravitational waves lie in the peak frequency region of $10^{-4}-0.01 {\rm Hz}$, corresponding to an energy spectrum in the range of $10^{-8}-10^{-7}$, which can be detected by LISA and Taiji. If a high baryon density is generated through Affleck-Dine baryogenesis or other mechanisms, the baryon chemical potential significantly reduces the phase transition rate, potentially dropping it to the order of $\beta/H\sim 10^1$, leading to the production of nanohertz gravitational waves. Furthermore, a critical quark chemical potential exists with a zero phase transition rate $\beta/H=0$, indicating that the false vacuum will not decay, thus supporting the formation of primordial quark nuggets in the early universe., Comment: 22 pages

The first very long baseline interferometry (VLBI) detections at 870$\mu$m wavelength (345$\,$GHz frequency) are reported, achieving the highest diffraction-limited angular resolution yet obtained from the surface of the Earth, and the highest-frequency example of the VLBI technique to date. These include strong detections for multiple sources observed on inter-continental baselines between telescopes in Chile, Hawaii, and Spain, obtained during observations in October 2018. The longest-baseline detections approach 11$\,$G$\lambda$ corresponding to an angular resolution, or fringe spacing, of 19$\mu$as. The Allan deviation of the visibility phase at 870$\mu$m is comparable to that at 1.3$\,$mm on the relevant integration time scales between 2 and 100$\,$s. The detections confirm that the sensitivity and signal chain stability of stations in the Event Horizon Telescope (EHT) array are suitable for VLBI observations at 870$\mu$m. Operation at this short wavelength, combined with anticipated enhancements of the EHT, will lead to a unique high angular resolution instrument for black hole studies, capable of resolving the event horizons of supermassive black holes in both space and time., Comment: Corresponding author: S. Doeleman

We aim to evaluate Large Language Models (LLMs) for embodied decision making. While a significant body of work has been leveraging LLMs for decision making in embodied environments, we still lack a systematic understanding of their performance because they are usually applied in different domains, for different purposes, and built based on different inputs and outputs. Furthermore, existing evaluations tend to rely solely on a final success rate, making it difficult to pinpoint what ability is missing in LLMs and where the problem lies, which in turn blocks embodied agents from leveraging LLMs effectively and selectively. To address these limitations, we propose a generalized interface (Embodied Agent Interface) that supports the formalization of various types of tasks and input-output specifications of LLM-based modules. Specifically, it allows us to unify 1) a broad set of embodied decision-making tasks involving both state and temporally extended goals, 2) four commonly-used LLM-based modules for decision making: goal interpretation, subgoal decomposition, action sequencing, and transition modeling, and 3) a collection of fine-grained metrics which break down evaluation into various types of errors, such as hallucination errors, affordance errors, various types of planning errors, etc. Overall, our benchmark offers a comprehensive assessment of LLMs' performance for different subtasks, pinpointing the strengths and weaknesses in LLM-powered embodied AI systems, and providing insights for effective and selective use of LLMs in embodied decision making., Comment: Accepted for oral presentation at NeurIPS 2024 in the Datasets and Benchmarks track. Camera-ready version

The girth of a graph is defined as the length of a shortest cycle in the graph. A $(k; g)$-cage is a graph of minimum order among all $k$-regular graphs with girth $g$. A cycle $C$ in a graph $G$ is termed nonseparating if the graph $G-V(C)$ remains connected. A conjecture, proposed in [T. Jiang, D. Mubayi. Connectivity and Separating Sets of Cages. J. Graph Theory 29(1)(1998) 35--44], posits that every cycle of length $g$ within a $(k; g)$-cage is nonseparating. While the conjecture has been proven for even $g$ in the aforementioned work, this paper presents a proof demonstrating that the conjecture holds true for odd $g$ as well. Thus, the previously mentioned conjecture was proven to be true., Comment: This paper was published in Chinese on April 25, 2001, in Volume 14, Issue 2 of the "MATHEMATICA APPLICATA" on pages 99 to 102. This is the English version of the paper, incorporating additional details and rectifying certain typographical errors

Randomized smoothing is a popular approach for providing certified robustness guarantees against adversarial attacks, and has become a very active area of research. Over the past years, the average certified radius (ACR) has emerged as the single most important metric for comparing methods and tracking progress in the field. However, in this work, we show that ACR is an exceptionally poor metric for evaluating robustness guarantees provided by randomized smoothing. We theoretically show not only that a trivial classifier can have arbitrarily large ACR, but also that ACR is much more sensitive to improvements on easy samples than on hard ones. Empirically, we confirm that existing training strategies that improve ACR reduce the model's robustness on hard samples. Further, we show that by focusing on easy samples, we can effectively replicate the increase in ACR. We develop strategies, including explicitly discarding hard samples, reweighing the dataset with certified radius, and extreme optimization for easy samples, to achieve state-of-the-art ACR, although these strategies ignore robustness for the general data distribution. Overall, our results suggest that ACR has introduced a strong undesired bias to the field, and better metrics are required to holistically evaluate randomized smoothing.

The code review comment (CRC) is pivotal in the process of modern code review. It provides reviewers with the opportunity to identify potential bugs, offer constructive feedback, and suggest improvements. Clear and concise code review comments (CRCs) facilitate the communication between developers and is crucial to the correct understanding of the issues identified and proposed solutions. Despite the importance of CRCs' clarity, there is still a lack of guidelines on what constitutes a good clarity and how to evaluate it. In this paper, we conduct a comprehensive study on understanding and evaluating the clarity of CRCs. We first derive a set of attributes related to the clarity of CRCs, namely RIE attributes (i.e., Relevance, Informativeness, and Expression), as well as their corresponding evaluation criteria based on our literature review and survey with practitioners. We then investigate the clarity of CRCs in open-source projects written in nine programming languages and find that a large portion (i.e., 28.8%) of the CRCs lack the clarity in at least one of the attributes. Finally, we propose ClearCRC, an automated framework that evaluates the clarity of CRCs. Experimental results show that ClearCRC can effectively evaluate the clarity of CRCs and outperform the baselines.

We introduce MotionRL, the first approach to utilize Multi-Reward Reinforcement Learning (RL) for optimizing text-to-motion generation tasks and aligning them with human preferences. Previous works focused on improving numerical performance metrics on the given datasets, often neglecting the variability and subjectivity of human feedback. In contrast, our novel approach uses reinforcement learning to fine-tune the motion generator based on human preferences prior knowledge of the human perception model, allowing it to generate motions that better align human preferences. In addition, MotionRL introduces a novel multi-objective optimization strategy to approximate Pareto optimality between text adherence, motion quality, and human preferences. Extensive experiments and user studies demonstrate that MotionRL not only allows control over the generated results across different objectives but also significantly enhances performance across these metrics compared to other algorithms.

In this work, we propose a novel method (GLOV) enabling Large Language Models (LLMs) to act as implicit Optimizers for Vision-Langugage Models (VLMs) to enhance downstream vision tasks. Our GLOV meta-prompts an LLM with the downstream task description, querying it for suitable VLM prompts (e.g., for zero-shot classification with CLIP). These prompts are ranked according to a purity measure obtained through a fitness function. In each respective optimization step, the ranked prompts are fed as in-context examples (with their accuracies) to equip the LLM with the knowledge of the type of text prompts preferred by the downstream VLM. Furthermore, we also explicitly steer the LLM generation process in each optimization step by specifically adding an offset difference vector of the embeddings from the positive and negative solutions found by the LLM, in previous optimization steps, to the intermediate layer of the network for the next generation step. This offset vector steers the LLM generation toward the type of language preferred by the downstream VLM, resulting in enhanced performance on the downstream vision tasks. We comprehensively evaluate our GLOV on 16 diverse datasets using two families of VLMs, i.e., dual-encoder (e.g., CLIP) and encoder-decoder (e.g., LLaVa) models -- showing that the discovered solutions can enhance the recognition performance by up to 15.0% and 57.5% (3.8% and 21.6% on average) for these models., Comment: Code: https://github.com/jmiemirza/GLOV

In this paper we give an algebraic construction of the fused model for ABJM spin chain and find the corresponding boost operator. We also investigate the open spin Hamiltonian for fused model and point out the general common structures of the boundary terms.

Dual-type qubits have become a promising way to suppress the crosstalk error of auxiliary operations in large-scale ion trap quantum computation. Here we demonstrate a direct entangling gate between dual-type qubits encoded in the $S_{1/2}$ and $D_{5/2}$ hyperfine manifolds of $^{137}\mathrm{Ba}^{+}$ ions. Our scheme is economic in the hardware, requiring only a single $532\,$nm laser system to entangle both qubit types by driving their Raman transitions. We achieve a Bell state fidelity of $96.3(4)\%$ for the dual-type Molmer-Sorensen gate between an $S$-$D$ ion pair, comparable to that for the same-type $S$-$S$ or $D$-$D$ gates. This technique can reduce the overhead for back-and-forth conversions between dual-type qubits in the quantum circuit with wide applications in quantum error correction and ion-photon quantum networks.

In this paper, we propose two variants of the positivity-preserving schemes, namely the truncated Euler-Maruyama (EM) method and the truncated Milstein scheme, applied to stochastic differential equations (SDEs) with positive solutions and super-linear coefficients. Under some regularity and integrability assumptions we derive the optimal strong convergence rates of the two schemes. Moreover, we demonstrate flexibility of our approaches by applying the truncated methods to approximate SDEs with super-linear coefficients (3/2 and Ai{\i}t-Sahalia models) directly and also with sub-linear coefficients (CIR model) indirectly. Numerical experiments are provided to verify the effectiveness of the theoretical results.

Source-Free domain adaptive Object Detection (SFOD) aims to transfer a detector (pre-trained on source domain) to new unlabelled target domains. Current SFOD methods typically follow the Mean Teacher framework, where weak-to-strong augmentation provides diverse and sharp contrast for self-supervised learning. However, this augmentation strategy suffers from an inherent problem called crucial semantics loss: Due to random, strong disturbance, strong augmentation is prone to losing typical visual components, hindering cross-domain feature extraction. To address this thus-far ignored limitation, this paper introduces a novel Weak-to-Strong Contrastive Learning (WSCoL) approach. The core idea is to distill semantics lossless knowledge in the weak features (from the weak/teacher branch) to guide the representation learning upon the strong features (from the strong/student branch). To achieve this, we project the original features into a shared space using a mapping network, thereby reducing the bias between the weak and strong features. Meanwhile, a weak features-guided contrastive learning is performed in a weak-to-strong manner alternatively. Specifically, we first conduct an adaptation-aware prototype-guided clustering on the weak features to generate pseudo labels for corresponding strong features matched through proposals. Sequentially, we identify positive-negative samples based on the pseudo labels and perform cross-category contrastive learning on the strong features where an uncertainty estimator encourages adaptive background contrast. Extensive experiments demonstrate that WSCoL yields new state-of-the-art performance, offering a built-in mechanism mitigating crucial semantics loss for traditional Mean Teacher framework. The code and data will be released soon.

Learning from observation (LfO) aims to imitate experts by learning from state-only demonstrations without requiring action labels. Existing adversarial imitation learning approaches learn a generator agent policy to produce state transitions that are indistinguishable to a discriminator that learns to classify agent and expert state transitions. Despite its simplicity in formulation, these methods are often sensitive to hyperparameters and brittle to train. Motivated by the recent success of diffusion models in generative modeling, we propose to integrate a diffusion model into the adversarial imitation learning from observation framework. Specifically, we employ a diffusion model to capture expert and agent transitions by generating the next state, given the current state. Then, we reformulate the learning objective to train the diffusion model as a binary classifier and use it to provide "realness" rewards for policy learning. Our proposed framework, Diffusion Imitation from Observation (DIFO), demonstrates superior performance in various continuous control domains, including navigation, locomotion, manipulation, and games. Project page: https://nturobotlearninglab.github.io/DIFO, Comment: NeurIPS 2024. Project page: https://nturobotlearninglab.github.io/DIFO

We report the detection of an extended very-high-energy (VHE) gamma-ray source coincident with the location of middle-aged (62.4~\rm kyr) pulsar PSR J0248+6021, by using the LHAASO-WCDA data of live 796 days and LHAASO-KM2A data of live 1216 days. A significant excess of \gray induced showers is observed both by WCDA in energy bands of 1-25~\rm TeV and KM2A in energy bands of $>$ 25~\rm TeV with 7.3 $\sigma$ and 13.5 $\sigma$, respectively. The best-fit position derived through WCDA data is R.A. = 42.06$^\circ \pm$ 0.12$^\circ$ and Dec. = 60.24$^\circ \pm $ 0.13$^\circ$ with an extension of 0.69$^\circ\pm$0.15$^\circ$ and that of the KM2A data is R.A.= 42.29$^\circ \pm $ 0.13$^\circ$ and Dec. = 60.38$^\circ \pm$ 0.07$^\circ$ with an extension of 0.37$^\circ\pm$0.07$^\circ$. No clear extended multiwavelength counterpart of this LHAASO source has been found from the radio band to the GeV band. The most plausible explanation of the VHE \gray emission is the inverse Compton process of highly relativistic electrons and positrons injected by the pulsar. These electrons/positrons are hypothesized to be either confined within the pulsar wind nebula or to have already escaped into the interstellar medium, forming a pulsar halo., Comment: 12 pages, 10 figures, Accepted by Sci. China-Phys. Mech. Astron

Open Radio Access Networks (RAN) aim to bring a paradigm shift to telecommunications industry, by enabling an open, intelligent, virtualized, and multi-vendor interoperable RAN ecosystem. At the center of this movement, O-RAN ALLIANCE defines the O-RAN architecture and standards, so that companies around the globe can use these specifications to create innovative and interoperable solutions. To accelerate the adoption of O-RAN products, rigorous testing of O-RAN Radio Unit (O-RU) and other O-RAN products plays a key role. O-RAN ALLIANCE has approved around 20 Open Testing and Integration Centres (OTICs) globally. OTICs serve as vendor-neutral platforms for providing the testing and integration services, with the vision that an O-RAN product certified in any OTIC is accepted in other parts of the world. To demonstrate the viability of such a certified-once-and-use-everywhere approach, one theme in the O-RAN Global PlugFest Spring 2024 is to demonstrate consistent and repeatable testing for the open fronthaul interface across multiple labs. Towards this, Japan OTIC and Asia Pacific OTIC in Singapore have teamed up together with an O-RU vendor and Keysight Technology. Our international team successfully completed all test cases defined by O-RAN ALLIANCE for O-RU conformance testing. In this paper, we share our journey in achieving this outcome, focusing on the challenges we have overcome and the lessons we have learned through this process., Comment: Published version at RitiRAN Workshop - co-located with IEEE VTC Fall 2024

Open Radio Access Networks (O-RAN) are expected to revolutionize the telecommunications industry with benefits like cost reduction, vendor diversity, and improved network performance through AI optimization. Supporting the O-RAN ALLIANCE's mission to achieve more intelligent, open, virtualized and fully interoperable mobile networks, O-RAN Open Testing and Integration Centers (OTICs) play a key role in accelerating the adoption of O-RAN specifications based on rigorous testing and validation. One theme in the recent O-RAN Global PlugFest Spring 2024 focused on demonstrating consistent and repeatable Open Fronthaul testing in multiple labs. To respond to this topic, in this paper, we present a detailed analysis of the testing methodologies and results for O-RAN Distributed Unit (O-DU) in O-RAN across two OTICs. We identify key differences in testing setups, share challenges encountered, and propose best practices for achieving repeatable and consistent testing results. Our findings highlight the impact of different deployment technologies and testing environments on performance and conformance testing outcomes, providing valuable insights for future O-RAN implementations., Comment: Published version at RitiRAN Workshop - co-located with IEEE VTC Fall 2024

We introduce a task and dataset for referring expression generation and comprehension in multi-agent embodied environments. In this task, two agents in a shared scene must take into account one another's visual perspective, which may be different from their own, to both produce and understand references to objects in a scene and the spatial relations between them. We collect a dataset of 2,970 human-written referring expressions, each paired with human comprehension judgments, and evaluate the performance of automated models as speakers and listeners paired with human partners, finding that model performance in both reference generation and comprehension lags behind that of pairs of human agents. Finally, we experiment training an open-weight speaker model with evidence of communicative success when paired with a listener, resulting in an improvement from 58.9 to 69.3% in communicative success and even outperforming the strongest proprietary model., Comment: Accepted to EMNLP2024 Main

Large language models (LLMs) have exhibited remarkable fluency across various tasks. However, their unethical applications, such as disseminating disinformation, have become a growing concern. Although recent works have proposed a number of LLM detection methods, their robustness and reliability remain unclear. In this paper, we present RAFT: a grammar error-free black-box attack against existing LLM detectors. In contrast to previous attacks for language models, our method exploits the transferability of LLM embeddings at the word-level while preserving the original text quality. We leverage an auxiliary embedding to greedily select candidate words to perturb against the target detector. Experiments reveal that our attack effectively compromises all detectors in the study across various domains by up to 99%, and are transferable across source models. Manual human evaluation studies show our attacks are realistic and indistinguishable from original human-written text. We also show that examples generated by RAFT can be used to train adversarially robust detectors. Our work shows that current LLM detectors are not adversarially robust, underscoring the urgent need for more resilient detection mechanisms., Comment: Accepted by EMNLP 2024

In this paper, we propose AutoDAN-Turbo, a black-box jailbreak method that can automatically discover as many jailbreak strategies as possible from scratch, without any human intervention or predefined scopes (e.g., specified candidate strategies), and use them for red-teaming. As a result, AutoDAN-Turbo can significantly outperform baseline methods, achieving a 74.3% higher average attack success rate on public benchmarks. Notably, AutoDAN-Turbo achieves an 88.5 attack success rate on GPT-4-1106-turbo. In addition, AutoDAN-Turbo is a unified framework that can incorporate existing human-designed jailbreak strategies in a plug-and-play manner. By integrating human-designed strategies, AutoDAN-Turbo can even achieve a higher attack success rate of 93.4 on GPT-4-1106-turbo., Comment: Pre-print. Project Page: https://autodans.github.io/AutoDAN-Turbo Code: https://github.com/SaFoLab-WISC/AutoDAN-Turbo

Although flares from late-type main-sequence stars have been frequently detected in multi-wavelength, the associated dynamical process has been rarely reported so far. Here, we report follow-up observations of an X-ray transient triggered by WXT onboard the Einstein Probe at UT08:45:08 in 2024, May 7. The photometry in multi-bands and time-resolved spectroscopy started at 3 and 7.5 hours after the trigger, respectively, which enables us to identify the transient as a flare of the M-dwarf 2MASS J12184187-0609123. The bolometric energy released in the flare is estimated to be $\sim10^{36}\ \mathrm{erg}$ from its X-ray light curve. The H$\alpha$ emission-line profile obtained at about 7 hours after the trigger shows an evident blue asymmetry with a maximum velocity of $200-250\ \mathrm{km\ s^{-1}}$. The blue wing can be likely explained by the chromospheric temperature (cool) upflow associated with chromospheric evaporation, in which the mass of the evaporating plasma is estimated to be $1.2\times10^{18}$g. In addition, a prominence eruption with an estimated mass of $7\times10^{15}\mathrm{g}

Recent advancements in Large Language Models (LLMs) have significantly increased context window sizes, enabling sophisticated applications but also introducing substantial computational overheads, particularly computing key-value (KV) cache in the prefill stage. Prefix caching has emerged to save GPU power in this scenario, which saves KV cache at disks and reuse them across multiple queries. However, traditional prefix caching mechanisms often suffer from substantial latency because the speed of loading KV cache from disks to GPU memory is bottlenecked by the throughput of I/O devices. To optimize the latency of long-context prefill, we propose Cake, a novel KV cache loader, which employs a bidirectional parallelized KV cache generation strategy. Upon receiving a prefill task, Cake simultaneously and dynamically loads saved KV cache from prefix cache locations and computes KV cache on local GPUs, maximizing the utilization of available computation and I/O bandwidth resources. Additionally, Cake automatically adapts to diverse system statuses without manual parameter. tuning. In experiments on various prompt datasets, GPUs, and I/O devices, Cake offers up to 68.1% Time To First Token (TTFT) reduction compare with compute-only method and 94.6% TTFT reduction compare with I/O-only method.

Safety is a paramount concern of large language models (LLMs) in their open deployment. To this end, safeguard methods aim to enforce the ethical and responsible use of LLMs through safety alignment or guardrail mechanisms. However, we found that the malicious attackers could exploit false positives of safeguards, i.e., fooling the safeguard model to block safe content mistakenly, leading to a new denial-of-service (DoS) attack on LLMs. Specifically, by software or phishing attacks on user client software, attackers insert a short, seemingly innocuous adversarial prompt into to user prompt templates in configuration files; thus, this prompt appears in final user requests without visibility in the user interface and is not trivial to identify. By designing an optimization process that utilizes gradient and attention information, our attack can automatically generate seemingly safe adversarial prompts, approximately only 30 characters long, that universally block over 97\% of user requests on Llama Guard 3. The attack presents a new dimension of evaluating LLM safeguards focusing on false positives, fundamentally different from the classic jailbreak.

Large Language Models (LLMs) boosts human efficiency but also poses misuse risks, with watermarking serving as a reliable method to differentiate AI-generated content from human-created text. In this work, we propose a novel theoretical framework for watermarking LLMs. Particularly, we jointly optimize both the watermarking scheme and detector to maximize detection performance, while controlling the worst-case Type-I error and distortion in the watermarked text. Within our framework, we characterize the universally minimum Type-II error, showing a fundamental trade-off between detection performance and distortion. More importantly, we identify the optimal type of detectors and watermarking schemes. Building upon our theoretical analysis, we introduce a practical, model-agnostic and computationally efficient token-level watermarking algorithm that invokes a surrogate model and the Gumbel-max trick. Empirical results on Llama-13B and Mistral-8$\times$7B demonstrate the effectiveness of our method. Furthermore, we also explore how robustness can be integrated into our theoretical framework, which provides a foundation for designing future watermarking systems with improved resilience to adversarial attacks.

With expansive state-action spaces, efficient multi-agent exploration remains a longstanding challenge in reinforcement learning. Although pursuing novelty, diversity, or uncertainty attracts increasing attention, redundant efforts brought by exploration without proper guidance choices poses a practical issue for the community. This paper introduces a systematic approach, termed LEMAE, choosing to channel informative task-relevant guidance from a knowledgeable Large Language Model (LLM) for Efficient Multi-Agent Exploration. Specifically, we ground linguistic knowledge from LLM into symbolic key states, that are critical for task fulfillment, in a discriminative manner at low LLM inference costs. To unleash the power of key states, we design Subspace-based Hindsight Intrinsic Reward (SHIR) to guide agents toward key states by increasing reward density. Additionally, we build the Key State Memory Tree (KSMT) to track transitions between key states in a specific task for organized exploration. Benefiting from diminishing redundant explorations, LEMAE outperforms existing SOTA approaches on the challenging benchmarks (e.g., SMAC and MPE) by a large margin, achieving a 10x acceleration in certain scenarios.

Reinforcement learning encounters challenges in various environments related to robustness and explainability. Traditional Q-learning algorithms cannot effectively make decisions and utilize the historical learning experience. To overcome these limitations, we propose Cognitive Belief-Driven Q-Learning (CBDQ), which integrates subjective belief modeling into the Q-learning framework, enhancing decision-making accuracy by endowing agents with human-like learning and reasoning capabilities. Drawing inspiration from cognitive science, our method maintains a subjective belief distribution over the expectation of actions, leveraging a cluster-based subjective belief model that enables agents to reason about the potential probability associated with each decision. CBDQ effectively mitigates overestimated phenomena and optimizes decision-making policies by integrating historical experiences with current contextual information, mimicking the dynamics of human decision-making. We evaluate the proposed method on discrete control benchmark tasks in various complicate environments. The results demonstrate that CBDQ exhibits stronger adaptability, robustness, and human-like characteristics in handling these environments, outperforming other baselines. We hope this work will give researchers a fresh perspective on understanding and explaining Q-learning.

The expected regret and target semi-variance are two of the most important risk measures for downside risk. When the distribution of a loss is uncertain, and only partial information of the loss is known, their worst-case values play important roles in robust risk management for finance, insurance, and many other fields. Jagannathan (1977) derived the worst-case expected regrets when only the mean and variance of a loss are known and the loss is arbitrary, symmetric, or non-negative. While Chen et al. (2011) obtained the worst-case target semi-variances under similar conditions but focusing on arbitrary losses. In this paper, we first complement the study of Chen et al. (2011) on the worst-case target semi-variances and derive the closed-form expressions for the worst-case target semi-variance when only the mean and variance of a loss are known and the loss is symmetric or non-negative. Then, we investigate worst-case target semi-variances over uncertainty sets that represent undesirable scenarios faced by an investors. Our methods for deriving these worst-case values are different from those used in Jagannathan (1977) and Chen et al. (2011). As applications of the results derived in this paper, we propose robust portfolio selection methods that minimize the worst-case target semi-variance of a portfolio loss over different uncertainty sets. To explore the insights of our robust portfolio selection methods, we conduct numerical experiments with real financial data and compare our portfolio selection methods with several existing portfolio selection models related to the models proposed in this paper., Comment: 36 pages, 5 figures

Research into the external behaviors and internal mechanisms of large language models (LLMs) has shown promise in addressing complex tasks in the physical world. Studies suggest that powerful LLMs, like GPT-4, are beginning to exhibit human-like cognitive abilities, including planning, reasoning, and reflection. In this paper, we introduce a research line and methodology called LLM Psychology, leveraging human psychology experiments to investigate the cognitive behaviors and mechanisms of LLMs. We migrate the Typoglycemia phenomenon from psychology to explore the "mind" of LLMs. Unlike human brains, which rely on context and word patterns to comprehend scrambled text, LLMs use distinct encoding and decoding processes. Through Typoglycemia experiments at the character, word, and sentence levels, we observe: (I) LLMs demonstrate human-like behaviors on a macro scale, such as lower task accuracy and higher token/time consumption; (II) LLMs exhibit varying robustness to scrambled input, making Typoglycemia a benchmark for model evaluation without new datasets; (III) Different task types have varying impacts, with complex logical tasks (e.g., math) being more challenging in scrambled form; (IV) Each LLM has a unique and consistent "cognitive pattern" across tasks, revealing general mechanisms in its psychology process. We provide an in-depth analysis of hidden layers to explain these phenomena, paving the way for future research in LLM Psychology and deeper interpretability.

The multimodal model has demonstrated promise in histopathology. However, most multimodal models are based on H\&E and genomics, adopting increasingly complex yet black-box designs. In our paper, we propose a novel interpretable multimodal framework named SHAP-CAT, which uses a Shapley-value-based dimension reduction technique for effective multimodal fusion. Starting with two paired modalities -- H\&E and IHC images, we employ virtual staining techniques to enhance limited input data by generating a new clinical-related modality. Lightweight bag-level representations are extracted from image modalities and a Shapley-value-based mechanism is used for dimension reduction. For each dimension of the bag-level representation, attribution values are calculated to indicate how changes in the specific dimensions of the input affect the model output. In this way, we select a few top important dimensions of bag-level representation for each image modality to late fusion. Our experimental results demonstrate that the proposed SHAP-CAT framework incorporating synthetic modalities significantly enhances model performance, yielding a 5\% increase in accuracy for the BCI, an 8\% increase for IHC4BC-ER, and an 11\% increase for the IHC4BC-PR dataset.

Origami principles are used to create strong, lightweight structures with complex mechanical response. However, identifying the fundamental physical principles that determine a sheet's behavior remains a challenge. We introduce a new analytic theory in which commonly studied origami sheets fall into distinct topological classes that predict sharply varying mechanical behavior, including effective stiffness and smoothness of mechanical response under external loads. Origami sheets with negative Poisson's ratios, such as the Miura ori, have conventional, smooth mechanical response amenable to continuum-based approaches. In contrast, positive Poisson's ratio, as in the Eggbox ori, generates a topological transition to lines of doubly degenerate zero modes that lead to dramatically softer structures with uneven, complex patterns of spatial response. These patterns interact in complicated ways with origami boundary conditions and source terms, leading to rich physical phenomena in experimentally accessible systems. This approach highlights topological mechanics, with deep connections to topologically protected quantum-mechanical systems, as a design principle for controlling the mechanical response of thin, complex sheets.

Recent breakthroughs in preference alignment have significantly improved Large Language Models' ability to generate texts that align with human preferences and values. However, current alignment metrics typically emphasize the post-hoc overall improvement, while overlooking a critical aspect: regression, which refers to the backsliding on previously correctly-handled data after updates. This potential pitfall may arise from excessive fine-tuning on already well-aligned data, which subsequently leads to over-alignment and degeneration. To address this challenge, we propose FlipGuard, a constrained optimization approach to detect and mitigate update regression with focal attention. Specifically, FlipGuard identifies performance degradation using a customized reward characterization and strategically enforces a constraint to encourage conditional congruence with the pre-aligned model during training. Comprehensive experiments demonstrate that FlipGuard effectively alleviates update regression while demonstrating excellent overall performance, with the added benefit of knowledge preservation while aligning preferences., Comment: Accepted by EMNLP 2024 Main track

Synchrotron polarization of relativistic nonthermal electrons in gamma-ray bursts (GRBs) has been widely studied. However, recent numerical simulations of relativistic shocks and magnetic reconnection have found that a more realistic electron distribution consists of a power-law component plus a thermal component, which requires observational validation. In this paper, we investigate synchrotron polarization using a hybrid energy distribution of relativistic thermal and nonthermal electrons within a globally toroidal magnetic field in GRB prompt emission. Our results show that, compared to the case of solely non-thermal electrons, the synchrotron polarization degrees (PDs) in these hybrid electrons can vary widely depending on different parameters and that the PD decreases progressively with frequency in the $\gamma$-ray, X-ray, and optical bands. The time-averaged PD spectrum displays a significant bump in the $\gamma$-ray and X-ray bands with the PDs higher than $\sim60\%$ if the thermal peak energy of electrons is much smaller than the conjunctive energy of electrons between the thermal and non-thermal distribution. The high synchrotron PD ($\gtrsim 60\%$) in the $\gamma$-ray and X-ray bands, which generally can not be produced by solely non-thermal electrons with typical power-law slopes, can be achieved by the hybrid electrons and primarily originates from the exponential decay part of the thermal component. Moreover, this model can roughly explain the PDs and spectral properties of some GRBs, where GRB 110301A with a high PD ($70_{-22}^{+22} \%$) may be potential evidence for the existence of relativistic thermal electrons., Comment: 13 pages, 8 figures, 2 tables. Accepted for publication in ApJ