Search

Recent large reconstruction models have made notable progress in generating high-quality 3D objects from single images. However, these methods often struggle with controllability, as they lack information from multiple views, leading to incomplete or inconsistent 3D reconstructions. To address this limitation, we introduce LucidFusion, a flexible end-to-end feed-forward framework that leverages the Relative Coordinate Map (RCM). Unlike traditional methods linking images to 3D world thorough pose, LucidFusion utilizes RCM to align geometric features coherently across different views, making it highly adaptable for 3D generation from arbitrary, unposed images. Furthermore, LucidFusion seamlessly integrates with the original single-image-to-3D pipeline, producing detailed 3D Gaussians at a resolution of $512 \times 512$, making it well-suited for a wide range of applications., Comment: 17 pages, 12 figures, [project page](https://heye0507.github.io/LucidFusion_page/)

Diffusion transformers (DiT) have already achieved appealing synthesis and scaling properties in content recreation, e.g., image and video generation. However, scaling laws of DiT are less explored, which usually offer precise predictions regarding optimal model size and data requirements given a specific compute budget. Therefore, experiments across a broad range of compute budgets, from 1e17 to 6e18 FLOPs are conducted to confirm the existence of scaling laws in DiT for the first time. Concretely, the loss of pretraining DiT also follows a power-law relationship with the involved compute. Based on the scaling law, we can not only determine the optimal model size and required data but also accurately predict the text-to-image generation loss given a model with 1B parameters and a compute budget of 1e21 FLOPs. Additionally, we also demonstrate that the trend of pre-training loss matches the generation performances (e.g., FID), even across various datasets, which complements the mapping from compute to synthesis quality and thus provides a predictable benchmark that assesses model performance and data quality at a reduced cost.

Combining Atacama Large Millimeter/sub-millimeter Array CO(2-1) mapping and JWST near- and mid-infrared imaging, we characterize the relationship between CO(2-1) and polycyclic aromatic hydrocarbon (PAH) emission at ~100 pc resolution in 66 nearby star-forming galaxies, expanding the sample size from previous ~100 pc resolution studies by more than an order of magnitude. Focusing on regions of galaxies where most of the gas is likely to be molecular, we find strong correlations between CO(2-1) and 3.3 micron, 7.7 micron, and 11.3 micron PAH emission, estimated from JWST's F335M, F770W, and F1130W filters. We derive power law relations between CO(2-1) and PAH emission, which have indices in the range 0.8-1.2, implying relatively weak variations in the observed CO-to-PAH ratios across the regions that we study. We find that CO-to-PAH ratios and scaling relationships near HII regions are similar to those in diffuse sight lines. The main difference between the two types of regions is that sight lines near HII regions show higher intensities in all tracers. Galaxy centers, on the other hand, show higher overall intensities and enhanced CO-to-PAH ratios compared to galaxy disks. Individual galaxies show 0.19 dex scatter in the normalization of CO at fixed I_PAH and this normalization anti-correlates with specific star formation rate (SFR/M*) and correlates with stellar mass. We provide a prescription that accounts for these galaxy-to-galaxy variations and represents our best current empirical predictor to estimate CO(2-1) intensity from PAH emission, which allows one to take advantage of JWST's excellent sensitivity and resolution to trace cold gas., Comment: 21 pages, 4 figures, 3 tables. Submitted to ApJ

The quantum spin Hall effect has been predicted in twisted homobilayer transition metal dichalcogenides (TMDs) owing to the layer-pseudospin magnetic field. Recently, experimental observations have also confirmed such topological states of matter. However, the topological electronic properties in multilayer moir\'e superlattices remain to be further explored. In twisted TMDs heterotrilayers, the realization of moir\'e potential with various symmetries becomes feasible. Here, we demonstrate that twisted trilayer TMDs can enter a topological insulator phase under the influence of moir\'e potential with ${C_6}$ symmetry. Specifically, we built two types of trilayer heterostructures, where the low-energy valence band electrons are contributed by the middle layer. In the AA-stacked moir\'e WS$_2$/WSe$_2$/MoS$_2$ heterotrilayers where only the middle layer is twisted, the maxima of the moir\'e potential exhibits an approximate ${C_6}$ symmetry. The $C_6$ symmetry effectively compensates for the spatial inversion symmetry breaking in the WSe$_2$ layer, leading to a twist-angle-dependent topological phase transition. Leveraging a Green's function approach, we calculate the local state density of edge states at topological minigaps, confirming their nature as moir\'e edge states. In the helical twisted AA-stacked moir\'e MoS$_2$/WSe$_2$/MoS$_2$ heterotrilayers, we observed a mosaic pattern of topological and trivial insulators. The emergence of topological mosaic is attributed to the maxima of the local moir\'e potential possessing $C_6$ symmetry. The results provide a new way for the experimental realization of topological phases in TMDs heterojunctions., Comment: 8 pages, 6 figures

We model emissivities of the HCN and CO $J=1-0$ transitions using measured properties of clouds found in normal star forming galaxies and more extreme systems. These models are compared with observations of HCN and CO $J=1-0$ transitions. We combine these model emissivities with predictions of gravoturbulent models of star formation, explore the impact of excitation and optical depth on CO and HCN emission, and assess if observed HCN/CO ratios track the fraction of gravitationally-bound dense gas, $f_\mathrm{grav}$, in molecular clouds. Our modeled HCN/CO ratios and emissivities are consistent with measurements from observations. CO emission shows a range of optical depths across different environments, from optically thick in normal galaxies to moderately optically thin in extreme systems. HCN is only moderately optically thick, with significant subthermal excitation in both normal and extreme galaxies. We find an anticorrelation between HCN/CO and $f_\mathrm{grav}$ as predicted by gravoturbulent models of star formation. Instead this ratio tracks gas at moderate densities ($n>10^{3.5}\ \mathrm{cm}^{-3}$), which is below the standard dense gas threshold of $n>10^{4.5}\ \mathrm{cm}^{-3}$. Variations in CO emissivity depend strongly on optical depth, due to variations in the dynamics of the cloud gas. HCN emissivity depends more strongly on excitation, and thus does not directly track variations in CO emissivity. We conclude that a single line ratio, such as HCN/CO, will not consistently track the fraction of gravitationally-bound, star-forming gas if the critical density for star formation varies in molecular clouds. This work highlights important uncertainties that need to be considered when observationally applying an HCN conversion factor in order to estimate the dense (i.e. $n>10^{4.5}\ \mathrm{cm}^{-3}$) gas content in nearby galaxies., Comment: Accepted for publication in A&A

Recent advances in Large Language Models (LLMs) have revolutionized code generation, leading to widespread adoption of AI coding tools by developers. However, LLMs can generate license-protected code without providing the necessary license information, leading to potential intellectual property violations during software production. This paper addresses the critical, yet underexplored, issue of license compliance in LLM-generated code by establishing a benchmark to evaluate the ability of LLMs to provide accurate license information for their generated code. To establish this benchmark, we conduct an empirical study to identify a reasonable standard for "striking similarity" that excludes the possibility of independent creation, indicating a copy relationship between the LLM output and certain open-source code. Based on this standard, we propose an evaluation benchmark LiCoEval, to evaluate the license compliance capabilities of LLMs. Using LiCoEval, we evaluate 14 popular LLMs, finding that even top-performing LLMs produce a non-negligible proportion (0.88% to 2.01%) of code strikingly similar to existing open-source implementations. Notably, most LLMs fail to provide accurate license information, particularly for code under copyleft licenses. These findings underscore the urgent need to enhance LLM compliance capabilities in code generation tasks. Our study provides a foundation for future research and development to improve license compliance in AI-assisted software development, contributing to both the protection of open-source software copyrights and the mitigation of legal risks for LLM users.

The molecular-to-atomic gas ratio is crucial to the evolution of the interstellar medium in galaxies. We investigate the balance between the atomic ($\Sigma_{\rm HI}$) and molecular gas ($\Sigma_{\rm H2}$) surface densities in eight nearby star-forming galaxies using new high-quality observations from MeerKAT and ALMA (for HI and CO, respectively). We define the molecular gas ratio as $R_{\rm mol} = \Sigma_{\rm H2} / \Sigma_{\rm HI}$ and measure how it depends on local conditions in the galaxy disks using multi-wavelength observations. We find that, depending on the galaxy, HI is detected at $>3\sigma$ out to 20-120 kpc in galactocentric radius ($r_{\rm gal}$). The typical radius at which $\Sigma_{\rm HI}$ reaches 1~$\rm M_\odot~pc^{-2}$ is $r_{\rm HI}\approx22$~kpc, which corresponds to 1-3 times the optical radius ($r_{25}$). $R_{\rm mol}$ correlates best with the dynamical equilibrium pressure, P$_{\rm DE}$, among potential drivers studied, with a median correlation coefficient of $<\rho>=0.89$. Correlations between $R_{\rm mol}$ and star formation rate, total gas and stellar surface density, metallicity, and $\Sigma_{\rm SFR}$/P$_{\rm DE}$ are present but somewhat weaker. Our results also show a direct correlation between P$_{\rm DE}$ and $\Sigma_{\rm SFR}$, supporting self-regulation models. Quantitatively, we measure similar scalings as previous works and attribute the modest differences that we find to the effect of varying resolution and sensitivity. At $r_{\rm gal} {\gtrsim}0.4~r_{25}$, atomic gas dominates over molecular gas, and at the balance of these two gas phases, we find that the baryon mass is dominated by stars, with $\Sigma_{*} > 5~\Sigma_{\rm gas}$. Our study constitutes an important step in the statistical investigation of how local galaxy properties impact the conversion from atomic to molecular gas in nearby galaxies., Comment: accepted for publication in A&A; 20 pages, 12 Figures (+4 appendix pages)

Research on video generation has recently made tremendous progress, enabling high-quality videos to be generated from text prompts or images. Adding control to the video generation process is an important goal moving forward and recent approaches that condition video generation models on camera trajectories make strides towards it. Yet, it remains challenging to generate a video of the same scene from multiple different camera trajectories. Solutions to this multi-video generation problem could enable large-scale 3D scene generation with editable camera trajectories, among other applications. We introduce collaborative video diffusion (CVD) as an important step towards this vision. The CVD framework includes a novel cross-video synchronization module that promotes consistency between corresponding frames of the same video rendered from different camera poses using an epipolar attention mechanism. Trained on top of a state-of-the-art camera-control module for video generation, CVD generates multiple videos rendered from different camera trajectories with significantly better consistency than baselines, as shown in extensive experiments. Project page: https://collaborativevideodiffusion.github.io/.

The ability to assess sleep at home, capture sleep stages, and detect the occurrence of apnea (without on-body sensors) simply by analyzing the radio waves bouncing off people's bodies while they sleep is quite powerful. Such a capability would allow for longitudinal data collection in patients' homes, informing our understanding of sleep and its interaction with various diseases and their therapeutic responses, both in clinical trials and routine care. In this article, we develop an advanced machine learning algorithm for passively monitoring sleep and nocturnal breathing from radio waves reflected off people while asleep. Validation results in comparison with the gold standard (i.e., polysomnography) (n=849) demonstrate that the model captures the sleep hypnogram (with an accuracy of 81% for 30-second epochs categorized into Wake, Light Sleep, Deep Sleep, or REM), detects sleep apnea (AUROC = 0.88), and measures the patient's Apnea-Hypopnea Index (ICC=0.95; 95% CI = [0.93, 0.97]). Notably, the model exhibits equitable performance across race, sex, and age. Moreover, the model uncovers informative interactions between sleep stages and a range of diseases including neurological, psychiatric, cardiovascular, and immunological disorders. These findings not only hold promise for clinical practice and interventional trials but also underscore the significance of sleep as a fundamental component in understanding and managing various diseases., Comment: The first two authors contributed equally to this work

Gamma-ray bursts (GRBs) have long been considered potential sources of ultra-high-energy cosmic rays (UHECRs; with energy $\gtrsim 10^{18} {\rm~eV}$). In this work, we propose a novel model generating MeV emission lines in GRB, which can constrain the properties of heavy nuclei that potentially exist in GRB jets. Specifically, we find that relativistic hydrogen-like high-atomic-number ions originating from the $\beta$ decay of unstable nuclei and/or the recombination entrained in the GRB jet can generate narrow MeV emission lines through the de-excitation of excited-electrons. This model can successfully explain the MeV emission line observed in the most luminous GRB ever recorded, GRB~221009A, with suitable parameters including a Lorentz factor $\gamma \sim 820-1700$ and a total mass of heavy nuclei $M_{\rm tot} \sim 10^{23} - 10^{26}$~g. Especially, the emission line broadening can be reasonably attributed to both the expansion of the jet shell and the thermal motion of nuclei, naturally resulting in a narrow width ($\sigma_{\rm line} / E_{\rm line} \lesssim 0.2$) consistent with the observation. Furthermore, we predict that different GRBs can exhibit lines in different bands with various evolving behaviors, which might be confirmed with further observations. Finally, our model provides indirect evidence that GRBs may be one of the sources of UHECRs., Comment: 13 pages, 4 figures; Published in ApJL, https://doi.org/10.3847/2041-8213/ad4ce1

Combining different forms of prompts with pre-trained large language models has yielded remarkable results on reasoning tasks (e.g. Chain-of-Thought prompting). However, along with testing on more complex reasoning, these methods also expose problems such as invalid reasoning and fictional reasoning paths. In this paper, we develop \textit{Hypothesis Testing Prompting}, which adds conclusion assumptions, backward reasoning, and fact verification during intermediate reasoning steps. \textit{Hypothesis Testing prompting} involves multiple assumptions and reverses validation of conclusions leading to its unique correct answer. Experiments on two challenging deductive reasoning datasets ProofWriter and RuleTaker show that hypothesis testing prompting not only significantly improves the effect, but also generates a more reasonable and standardized reasoning process.

Prompt-based methods have gained increasing attention on NLP and shown validity on many downstream tasks. Many works have focused on mining these methods' potential for knowledge extraction, but few explore their ability to make logical reasoning. In this work, we focus on the effectiveness of the prompt-based methods on first-order logical reasoning and find that the bottleneck lies in logical negation. Based on our analysis, logical negation tends to result in spurious correlations to negative answers, while propositions without logical negation correlate to positive answers. To solve the problem, we propose a simple but effective method, Negation Augmenting and Negation Debiasing (NAND), which introduces negative propositions to prompt-based methods without updating parameters. Specifically, these negative propositions can counteract spurious correlations by providing "not" for all instances so that models cannot make decisions only by whether expressions contain a logical negation. Experiments on three datasets show that NAND not only solves the problem of calibrating logical negation but also significantly enhances prompt-based methods of logical reasoning without model retraining.

The problem of missing Galactic supernova remnants (SNRs) refers to the issue that the currently known Galactic SNRs are significantly incomplete compared to the theoretical prediction. To expand the sample of Galactic SNRs, we use GLEAM and THOR+VGPS data across four wavebands ranging from 118 to 1420 MHz to drive a spectral index map covering the region within 26.6{\deg} < l < 30.6{\deg}, $\vert b \vert \leq$ 1.25{\deg}, where numerous SNR candidates were recently found. By using the spectral index map of the sky region and detailed analysis of the spectral indices of individual sources, we confirmed four SNR candidates, namely G26.75+0.73, G27.06+0.04, G28.36+0.21, and G28.78$-$0.44, as SNRs. Additionally, we discovered an expanding molecular superbubble located in this region, discussed pulsars associated with SNR candidates, and discovered a long H$\alpha$ filament that spatially overlaps with the candidate G29.38+0.10. We suggest that the problem of missing Galactic SNRs not only arises from observation limitations, but also could be due to the low-density environments of some SNRs, and the different SN explosion properties., Comment: 13 pages, 6 figures; Published in AJ

We present observations of the central 9 kpc of the Antennae merger (NGC 4038/9) at 55 pc resolution in the CO 2-1 line obtained with the Atacama Large Millimeter/submillimeter Array (ALMA). We use a pixel-based analysis to compare the gas properties in the Antennae to those in 70 nearby spiral galaxies from the PHANGS-ALMA survey, as well as the merger and nearest luminous infrared galaxy NGC 3256. Compared to PHANGS galaxies at matched spatial resolution, the molecular gas in the Antennae exhibits some of the highest surface densities, velocity dispersions, peak brightness temperatures, and turbulent pressures. However, the virial parameters in the Antennae are consistent with many of the PHANGS galaxies. NGC 3256 has similar gas surface densities but higher nuclear velocity dispersions than the Antennae, as well as higher system-wide peak brightness temperatures and virial parameters. NGC 3256 is at a later stage in the merging process than the Antennae, which may result in more intense merger-driven gas flows that could drive up the turbulence in the gas. The high virial parameters in NGC 3256 may indicate that this increased turbulence is suppressing future star formation as NGC 3256 moves out of the starburst phase. In comparison, the relatively normal virial parameters in the Antennae may imply that it is about to undergo a new burst of star formation., Comment: 16 pages, 8 figures, accepted to MNRAS

Controllability plays a crucial role in video generation since it allows users to create desired content. However, existing models largely overlooked the precise control of camera pose that serves as a cinematic language to express deeper narrative nuances. To alleviate this issue, we introduce CameraCtrl, enabling accurate camera pose control for text-to-video(T2V) models. After precisely parameterizing the camera trajectory, a plug-and-play camera module is then trained on a T2V model, leaving others untouched. Additionally, a comprehensive study on the effect of various datasets is also conducted, suggesting that videos with diverse camera distribution and similar appearances indeed enhance controllability and generalization. Experimental results demonstrate the effectiveness of CameraCtrl in achieving precise and domain-adaptive camera control, marking a step forward in the pursuit of dynamic and customized video storytelling from textual and camera pose inputs. Our project website is at: https://hehao13.github.io/projects-CameraCtrl/., Comment: Project page: https://hehao13.github.io/projects-CameraCtrl/ Code: https://github.com/hehao13/CameraCtrl

The quality of images captured outdoors is often affected by the weather. One factor that interferes with sight is rain, which can obstruct the view of observers and computer vision applications that rely on those images. The work aims to recover rain images by removing rain streaks via Self-supervised Reinforcement Learning (RL) for image deraining (SRL-Derain). We locate rain streak pixels from the input rain image via dictionary learning and use pixel-wise RL agents to take multiple inpainting actions to remove rain progressively. To our knowledge, this work is the first attempt where self-supervised RL is applied to image deraining. Experimental results on several benchmark image-deraining datasets show that the proposed SRL-Derain performs favorably against state-of-the-art few-shot and self-supervised deraining and denoising methods.

Large language models (LLMs) still grapple with complex tasks like mathematical reasoning. Despite significant efforts invested in improving prefix prompts or reasoning process, the crucial role of problem context might have been neglected. Accurate recognition of inputs is fundamental for solving mathematical tasks, as ill-formed problems could potentially mislead LLM's reasoning. In this study, we propose a new approach named Problem Elaboration Prompting (PEP) to enhance the mathematical capacities of LLMs. Specifically, PEP decomposes and elucidates the problem context before reasoning, therefore enhancing the context modeling and parsing efficiency. Experiments across datasets and models demonstrate promising performances: (1) PEP demonstrates an overall enhancement in various mathematical tasks. For instance, with the GPT-3.5 model, PEP exhibits improvements of 9.93% and 8.80% on GSM8k through greedy decoding and self-consistency, respectively. (2) PEP can be easily implemented and integrated with other prompting methods. (3) PEP shows particular strength in handling distraction problems.

CO emission has been widely used as a tracer of molecular gas mass. However, it is a long-standing issue to accurately constrain the CO-to-H$_2$ conversion factor ($\alpha_{\mathrm{CO}}$) that converts CO luminosity to molecular gas mass, especially in starburst galaxies. We present the first resolved $\alpha_{\mathrm{CO}}$ modeling results with multiple ALMA CO and $^{13}$CO transition observations at both giant molecular cloud (GMC) scale at 150 pc and kpc scale for one of the closest starburst mergers, the Antennae. By combining our CO modeling results and measurements of 350 GHz dust continuum, we find that most GMCs in the Antennae have $\alpha_{\mathrm{CO}}$ values $\sim$4 times smaller than the commonly adopted Milky Way value (4.3). We find $\alpha_{\mathrm{CO}}$ at GMC scales shows a strong dependence on CO intensity, $^{13}$CO/CO ratio and GMC velocity dispersion, which is consistent with various theoretical and simulation predictions. Specifically, we suggest that the $^{13}$CO/CO line ratio and the velocity dispersion can be used to infer $\alpha_{\mathrm{CO}}$ in starburst regions. By applying our modeled $\alpha_{\mathrm{CO}}$ in GMC analyses, we find that GMCs in the Antennae are less gravitationally bound than in normal spiral galaxies, which is more consistent with what is predicted by merger simulations. At kpc scale, we find that our modeled $\alpha_{\mathrm{CO}}$ values are smaller than the modeled $\alpha_{\mathrm{CO}}$ at GMC scale by 40%, which can be due to inclusion of a diffuse gas component with lower $\alpha_{\mathrm{CO}}$ values. We find a similar correlation of $\alpha_{\mathrm{CO}}$ and CO intensity at kpc scales to that at GMC scales., Comment: 22 pages and 16 figures in the main text; accepted to ApJ

We study young massive clusters (YMCs) in their embedded "infant" phase with $\sim0.\!^{\prime\prime}1$ ALMA, HST, and JWST observations targeting the central starburst ring in NGC 3351, a nearby Milky Way analog galaxy. Our new ALMA data reveal 18 bright and compact (sub-)millimeter continuum sources, of which 8 have counterparts in JWST images and only 6 have counterparts in HST images. Based on the ALMA continuum and molecular line data, as well as ancillary measurements for the HST and JWST counterparts, we identify 14 sources as infant star clusters with high stellar and/or gas masses (${\sim}10^5\;\mathrm{M_\odot}$), small radii (${\lesssim}\,5\;\mathrm{pc}$), large escape velocities ($6{-}10\;\mathrm{km/s}$), and short free-fall times ($0.5{-}1\;\mathrm{Myr}$). Their multiwavelength properties motivate us to divide them into four categories, likely corresponding to four evolutionary stages from starless clumps to exposed HII region-cluster complexes. Leveraging age estimates for HST-identified clusters in the same region, we infer an evolutionary timeline going from $\sim$1-2 Myr before cluster formation as starless clumps, to $\sim$4-6 Myr after as exposed HII region-cluster complexes. Finally, we show that the YMCs make up a substantial fraction of recent star formation across the ring, exhibit an non-uniform azimuthal distribution without a very coherent evolutionary trend along the ring, and are capable of driving large-scale gas outflows., Comment: 27 pages, 12 figures; ApJ accepted

Gamma-ray bursts (GRBs) have been proposed as one of promising sources of ultra-high-energy cosmic rays (UHECRs), but observational evidence is still lacking. The nearby B.O.A.T. (brightest of all time) GRB 221009A, an once-in-1000-year event, is able to accelerate protons to $\sim 10^{3}$ EeV. Protons arriving at the Milky Way are dominated by neutron-decay-induced protons. The inter-galactic magnetic fields would not yield a sizable delay of the $\geq 10{\rm~EeV}$ cosmic rays if its strength is $\lesssim 10^{-13}{\rm~G}$, while Galactic magnetic fields would cause a significant time delay. We predict that, an UHECR burst from GRB 221009A would be detectable by the Pierre Auger Observatory and the TA$\times$4, within $\sim$ 10 years. The detection of such an UHECR outburst will provide the direct evidence for UHECR acceleration in GRBs.

The introduction of ChatGPT has led to a significant increase in the utilization of Large Language Models (LLMs) for addressing downstream tasks. There's an increasing focus on cost-efficient training and deployment within this context. Low-cost training and deployment of LLMs represent the future development trend. This paper reviews the evolution of large language model training techniques and inference deployment technologies aligned with this emerging trend. The discussion on training includes various aspects, including data preprocessing, training architecture, pre-training tasks, parallel training, and relevant content related to model fine-tuning. On the inference side, the paper covers topics such as model compression, parallel computation, memory scheduling, and structural optimization. It also explores LLMs' utilization and provides insights into their future development., Comment: 30 pages,6 figures

Large language models (LLMs) face challenges in solving complex mathematical problems that require comprehensive capacities to parse the statements, associate domain knowledge, perform compound logical reasoning, and integrate the intermediate rationales. Tackling all these problems once could be arduous for LLMs, thus leading to confusion in generation. In this work, we explore the potential of enhancing LLMs with agents by meticulous decomposition and modeling of mathematical reasoning process. Specifically, we propose a formal description of the mathematical solving and extend LLMs with an agent-based zero-shot framework named $\bf{P}$lanner-$\bf{R}$easoner-$\bf{E}$xecutor-$\bf{R}$eflector (PRER). We further provide and implement two MathAgents that define the logical forms and inherent relations via a pool of actions in different grains and orientations: MathAgent-M adapts its actions to LLMs, while MathAgent-H aligns with humankind. Experiments on miniF2F and MATH have demonstrated the effectiveness of PRER and proposed MathAgents, achieving an increase of $12.3\%$($53.9\%\xrightarrow{}66.2\%$) on the MiniF2F, $9.2\%$ ($49.8\%\xrightarrow{}59.0\%$) on MATH, and $13.2\%$($23.2\%\xrightarrow{}35.4\%$) for level-5 problems of MATH against GPT-4. Further analytical results provide more insightful perspectives on exploiting the behaviors of LLMs as agents., Comment: There are unfair comparisons on miniF2F. This will be fixed in the future

In-Context Learning (ICL) is an important paradigm for adapting Large Language Models (LLMs) to downstream tasks through a few demonstrations. Despite the great success of ICL, the limitation of the demonstration number may lead to demonstration bias, i.e. the input-label mapping induced by LLMs misunderstands the task's essence. Inspired by human experience, we attempt to mitigate such bias through the perspective of the inter-demonstration relationship. Specifically, we construct Comparable Demonstrations (CDs) by minimally editing the texts to flip the corresponding labels, in order to highlight the task's essence and eliminate potential spurious correlations through the inter-demonstration comparison. Through a series of experiments on CDs, we find that (1) demonstration bias does exist in LLMs, and CDs can significantly reduce such bias; (2) CDs exhibit good performance in ICL, especially in out-of-distribution scenarios. In summary, this study explores the ICL mechanisms from a novel perspective, providing a deeper insight into the demonstration selection strategy for ICL., Comment: ICASSP 2024

Game theory, as an analytical tool, is frequently utilized to analyze human behavior in social science research. With the high alignment between the behavior of Large Language Models (LLMs) and humans, a promising research direction is to employ LLMs as substitutes for humans in game experiments, enabling social science research. However, despite numerous empirical researches on the combination of LLMs and game theory, the capability boundaries of LLMs in game theory remain unclear. In this research, we endeavor to systematically analyze LLMs in the context of game theory. Specifically, rationality, as the fundamental principle of game theory, serves as the metric for evaluating players' behavior -- building a clear desire, refining belief about uncertainty, and taking optimal actions. Accordingly, we select three classical games (dictator game, Rock-Paper-Scissors, and ring-network game) to analyze to what extent LLMs can achieve rationality in these three aspects. The experimental results indicate that even the current state-of-the-art LLM (GPT-4) exhibits substantial disparities compared to humans in game theory. For instance, LLMs struggle to build desires based on uncommon preferences, fail to refine belief from many simple patterns, and may overlook or modify refined belief when taking actions. Therefore, we consider that introducing LLMs into game experiments in the field of social science should be approached with greater caution., Comment: AAAI 2024