Search

With the ongoing advancement of autonomous driving technology and intelligent transportation systems, research into semantic segmentation has become increasingly pivotal. Accurate understanding and analysis of real-world scenarios are now essential for these emerging fields. However, traditional semantic segmentation methods often struggle to balance high model accuracy with computational efficiency, particularly in terms of parameter count. To address this challenge, we introduce SegRet, a novel approach that leverages the Retentive Network (RetNet) architecture and integrates a lightweight residual decoder featuring zero-initialization. SegRet exhibits three key characteristics: (1) Lightweight Residual Decoder: We incorporate a zero-initialization layer within the residual network framework, ensuring that the decoder remains computationally efficient while preserving critical information flow; (2) Robust Feature Extraction: Utilizing RetNet as the backbone, our model adeptly extracts hierarchical features from input images, thereby enhancing the depth and breadth of feature representation; (3) Parameter Efficiency: SegRet achieves state-of-the-art performance while significantly reducing the number of parameters, maintaining high accuracy without compromising on computational resources. Empirical evaluations on benchmark datasets such as ADE20K, Cityscapes, and COCO-Stuff10K demonstrate the efficacy of our approach. SegRet delivers impressive results, achieving an mIoU of 52.23\% on ADE20K with only 95.81M parameters, 83.36\% on Cityscapes, and 46.63\% on COCO-Stuff. The code is available at: https://github.com/ZhiyuanLi218/segret., Comment: 12 pages

Potentially Hazardous Asteroids (PHAs), a special subset of Near-Earth Objects, are both dangerous and scientifically valuable. PHAs that truly undergo close approaches with the Earth (dubbed CAPHAs) are of particular interest and extensively studied. The concept and study of CAPHA can be extended to other Solar system planets, which have significant implications for future planet-based observations and explorations. In this work, we conduct numerical simulations that incorporate the Yarkovsky effect to study the transformation of main belt asteroids into CAPHAs of terrestrial planets, using precise nominal timesteps, especially to ensure the reliability of the results for Mercury and Venus. Our simulations predict a total of 1893 Mercury-CAPHAs, 3014 Venus-CAPHAs, 3791 Earth-CAPHAs and 18066 Mars-CAPHAs, with an occurrence frequency of about 1, 9, 15 and 66 per year, respectively. The values for Mars-CAPHAs are consistent with our previous work, which were based on simulations with a larger nominal timestep. The predicted occurrence frequency and velocity distribution of Earth-CAPHAs are in reasonable agreement with the observed population of Earth-CAPHAs. We also find that certain asteroids can be caught in close approach with different planets at different times, raising an interesting possibility of using them as transportation between terrestrial planets in the future., Comment: 8 pages, 4 figures, accepted for publication in MNRAS. arXiv admin note: text overlap with arXiv:2405.02614

Despite much progress in training distributed artificial intelligence (AI), building cooperative multi-agent systems with multi-agent reinforcement learning (MARL) faces challenges in sample efficiency, interpretability, and transferability. Unlike traditional learning-based methods that require extensive interaction with the environment, large language models (LLMs) demonstrate remarkable capabilities in zero-shot planning and complex reasoning. However, existing LLM-based approaches heavily rely on text-based observations and struggle with the non-Markovian nature of multi-agent interactions under partial observability. We present COMPASS, a novel multi-agent architecture that integrates vision-language models (VLMs) with a dynamic skill library and structured communication for decentralized closed-loop decision-making. The skill library, bootstrapped from demonstrations, evolves via planner-guided tasks to enable adaptive strategies. COMPASS propagates entity information through multi-hop communication under partial observability. Evaluations on the improved StarCraft Multi-Agent Challenge (SMACv2) demonstrate COMPASS achieves up to 30\% higher win rates than state-of-the-art MARL algorithms in symmetric scenarios.

We report the position reconstruction methods and surface background model for the PandaX-4T dark matter direct search experiment. This work develops two position reconstruction algorithms: template matching (TM) method and photon acceptance function (PAF) method. Both methods determine the horizontal position of events based on the light pattern of secondary scintillation collected by the light sensors. After a comprehensive evaluation of resolution, uniformity, and robustness, the PAF method was selected for position reconstruction, while the TM method was employed for verification. The PAF method achieves a bulk event resolution of 1.0 mm and a surface event resolution of 4.4 mm for a typical $S2$ signal with a bottom charge of 1500 PE (about 14 keV). The uniformity is around 20\%. Robustness studies reveal average deviations of 5.1 mm and 8.8 mm for the commissioning run (Run0) and the first science run (Run1), respectively, due to the deactivation of certain PMTs. A data-driven surface background model is developed based on the PAF method. The surface background is estimated to be $0.09 \pm 0.06$ events for Run0 (0.54 tonne$\cdot$year) and $0.17 \pm 0.11$ events for Run1 (1.00 tonne$\cdot$year)., Comment: 22 pages, 15 figures, 2 tables

We perform a search of double beta decay of $^{136}$Xe to the excited state, $0^+_1$, of $^{136}$Ba (2$\nu \beta \beta$-0$_1^+$), using the dual-phase xenon detector of PandaX-4T with the first 94.9-day commissioning data. The multi-site events are reconstructed up to the MeV energy scale, which helps to improve the background model significantly. The background contribution from the stainless steel platform outside PandaX-4T cryostat is evaluated for the first time. No significant evidence for 2$\nu \beta \beta$-0$_1^+$ is observed, resulting in a lower limit of $T_{1/2}^{2 \nu \beta \beta-0_1^+} > 7.5 \times 10^{22}$ yr at the 90% confidence level. This is the first experimental limit on such a rare decay in a natural xenon-based detector.

This work investigates both direct and inverse problems of the variable-exponent sub-diffusion model, which attracts increasing attentions in both practical applications and theoretical aspects. Based on the perturbation method, which transfers the original model to an equivalent but more tractable form, the analytical extensibility of the solutions and the weak unique continuation principle are proved, which results in the uniqueness of the inverse space-dependent source problem from local internal observation. Then, based on the variational identity connecting the inversion input data with the unknown source function, we propose a weak norm and prove the conditional stability for the inverse problem in this norm. The iterative thresholding algorithm and Nesterov iteration scheme are employed to numerically reconstruct the smooth and non-smooth sources, respectively. Numerical experiments are performed to investigate their effectiveness.

Defacing is often applied to head magnetic resonance image (MRI) datasets prior to public release to address privacy concerns. The alteration of facial and nearby voxels has provoked discussions about the true capability of these techniques to ensure privacy as well as their impact on downstream tasks. With advancements in deep generative models, the extent to which defacing can protect privacy is uncertain. Additionally, while the altered voxels are known to contain valuable anatomical information, their potential to support research beyond the anatomical regions directly affected by defacing remains uncertain. To evaluate these considerations, we develop a refacing pipeline that recovers faces in defaced head MRIs using cascaded diffusion probabilistic models (DPMs). The DPMs are trained on images from 180 subjects and tested on images from 484 unseen subjects, 469 of whom are from a different dataset. To assess whether the altered voxels in defacing contain universally useful information, we also predict computed tomography (CT)-derived skeletal muscle radiodensity from facial voxels in both defaced and original MRIs. The results show that DPMs can generate high-fidelity faces that resemble the original faces from defaced images, with surface distances to the original faces significantly smaller than those of a population average face (p < 0.05). This performance also generalizes well to previously unseen datasets. For skeletal muscle radiodensity predictions, using defaced images results in significantly weaker Spearman's rank correlation coefficients compared to using original images (p < 10-4). For shin muscle, the correlation is statistically significant (p < 0.05) when using original images but not statistically significant (p > 0.05) when any defacing method is applied, suggesting that defacing might not only fail to protect privacy but also eliminate valuable information.

Accessing visual information is crucial yet challenging for people with low vision due to their visual conditions (e.g., low visual acuity, limited visual field). However, unlike blind people, low vision people have and prefer using their functional vision in daily tasks. Gaze patterns thus become an important indicator to uncover their visual challenges and intents, inspiring more adaptive visual support. We seek to deeply understand low vision users' gaze behaviors in different image viewing tasks, characterizing typical visual intents and the unique gaze patterns exhibited by people with different low vision conditions. We conducted a retrospective think-aloud study using eye tracking with 14 low vision participants and nine sighted controls. Participants completed various image viewing tasks and watched the playback of their gaze trajectories to reflect on their visual experiences. Based on the study, we derived a visual intent taxonomy with five intents characterized by participants' gaze behaviors and demonstrated how low vision conditions affect gaze patterns across visual intents. Our findings underscore the importance of combining visual ability information, image context, and eye tracking data in visual intent recognition, setting up a foundation for intent-aware assistive technologies for low vision.

Diffusion-weighted Magnetic Resonance Imaging (dMRI) is an essential tool in neuroimaging. It is arguably the sole noninvasive technique for examining the microstructural properties and structural connectivity of the brain. Recent years have seen the emergence of machine learning and data-driven approaches that enhance the speed, accuracy, and consistency of dMRI data analysis. However, traditional deep learning models often fell short, as they typically utilize pixel-level or volumetric patch-level embeddings similar to those used in structural MRI, and do not account for the unique distribution of various gradient encodings. In this paper, we propose a novel method called Polyhedra Encoding Transformer (PE-Transformer) for dMRI, designed specifically to handle spherical signals. Our approach involves projecting an icosahedral polygon onto a unit sphere to resample signals from predetermined directions. These resampled signals are then transformed into embeddings, which are processed by a transformer encoder that incorporates orientational information reflective of the icosahedral structure. Through experimental validation with various gradient encoding protocols, our method demonstrates superior accuracy in estimating multi-compartment models and Fiber Orientation Distributions (FOD), outperforming both conventional CNN architectures and standard transformers.

Many active galaxies harbor powerful relativistic jets, however, the detailed mechanisms of their formation and acceleration remain poorly understood. To investigate the area of jet acceleration and collimation with the highest available angular resolution, we study the innermost region of the bipolar jet in the nearby low-ionization nuclear emission-line region (LINER) galaxy NGC 1052. We combined observations of NGC 1052 taken with VLBA, GMVA, and EHT over one week in the spring of 2017. For the first time, NGC 1052 was detected with the EHT, providing a size of the central region in-between both jet bases of 250 RS (Schwarzschild radii) perpendicular to the jet axes. This size estimate supports previous studies of the jets expansion profile which suggest two breaks of the profile at around 300 RS and 10000 RS distances to the core. Furthermore, we estimated the magnetic field to be 1.25 Gauss at a distance of 22 {\mu}as from the central engine by fitting a synchrotron-self absorption spectrum to the innermost emission feature, which shows a spectral turn-over at about 130 GHz. Assuming a purely poloidal magnetic field, this implies an upper limit on the magnetic field strength at the event horizon of 26000 Gauss, which is consistent with previous measurements. The complex, low-brightness, double-sided jet structure in NGC 1052 makes it a challenge to detect the source at millimeter (mm) wavelengths. However, our first EHT observations have demonstrated that detection is possible up to at least 230 GHz. This study offers a glimpse through the dense surrounding torus and into the innermost central region, where the jets are formed. This has enabled us to finally resolve this region and provide improved constraints on its expansion and magnetic field strength., Comment: 22 pages, 10 figures, published in A&A

The 2017 observing campaign of the Event Horizon Telescope (EHT) delivered the first very long baseline interferometry (VLBI) images at the observing frequency of 230 GHz, leading to a number of unique studies on black holes and relativistic jets from active galactic nuclei (AGN). In total, eighteen sources were observed: the main science targets, Sgr A* and M87 along with various calibrators. We investigated the morphology of the sixteen AGN in the EHT 2017 data set, focusing on the properties of the VLBI cores: size, flux density, and brightness temperature. We studied their dependence on the observing frequency in order to compare it with the Blandford-K\"onigl (BK) jet model. We modeled the source structure of seven AGN in the EHT 2017 data set using linearly polarized circular Gaussian components and collected results for the other nine AGN from dedicated EHT publications, complemented by lower frequency data in the 2-86 GHz range. Then, we studied the dependences of the VLBI core flux density, size, and brightness temperature on the frequency measured in the AGN host frame. We compared the observations with the BK jet model and estimated the magnetic field strength dependence on the distance from the central black hole. Our results indicate a deviation from the standard BK model, particularly in the decrease of the brightness temperature with the observing frequency. Either bulk acceleration of the jet material, energy transfer from the magnetic field to the particles, or both are required to explain the observations.

We present a novel constraint on light dark matter utilizing $1.54$ tonne$\cdot$year of data acquired from the PandaX-4T dual-phase xenon time projection chamber. This constraint is derived through detecting electronic recoil signals resulting from the interaction with solar-enhanced dark matter flux. Low-mass dark matter particles, lighter than a few MeV/$c^2$, can scatter with the thermal electrons in the Sun. Consequently, with higher kinetic energy, the boosted dark matter component becomes detectable via contact scattering with xenon electrons, resulting in a few keV energy deposition that exceeds the threshold of PandaX-4T. We calculate the expected recoil energy in PandaX-4T considering the Sun's acceleration and the detection capabilities of the xenon detector. The first experimental search results using the xenon detector yield the most stringent cross-section of $3.51 \times 10^{-39}~\mathrm{cm}^2$ at $0.08~\mathrm{MeV}$/$c^2$ for a solar boosted dark matter mass ranging from $0.02$ to $10~ \mathrm{MeV}$/$c^2$, achieving a 23 fold improvement compared with earlier experimental studies.

The Receptance Weighted Key Value (RWKV) model offers a novel alternative to the Transformer architecture, merging the benefits of recurrent and attention-based systems. Unlike conventional Transformers, which depend heavily on self-attention, RWKV adeptly captures long-range dependencies with minimal computational demands. By utilizing a recurrent framework, RWKV addresses some computational inefficiencies found in Transformers, particularly in tasks with long sequences. RWKV has recently drawn considerable attention for its robust performance across multiple domains. Despite its growing popularity, no systematic review of the RWKV model exists. This paper seeks to fill this gap as the first comprehensive review of the RWKV architecture, its core principles, and its varied applications, such as natural language generation, natural language understanding, and computer vision. We assess how RWKV compares to traditional Transformer models, highlighting its capability to manage long sequences efficiently and lower computational costs. Furthermore, we explore the challenges RWKV encounters and propose potential directions for future research and advancement. We consistently maintain the related open-source materials at: https://github.com/MLGroupJLU/RWKV-Survey., Comment: 19 pages

We report the search for neutrinoless double-beta decay of $^{136}$Xe from the PandaX-4T experiment with a 3.7-tonne natural xenon target. The data reconstruction and the background modeling are optimized in the MeV energy region. A blind analysis is performed with data from the commissioning run and the first science run. No significant excess of signal over the background is observed. A lower limit on the half-life of $^{136}$Xe neutrinoless double-beta decay is established to be $2.1 \times 10^{24}$~yr at the 90\% confidence level, with a $^{136}$Xe exposure of 44.6~kg$\cdot$year. Our result represents the most stringent constraint from a natural xenon detector to date., Comment: 9 pages, 4 figures, 2 tables

We present OmniVLM, a sub-billion-parameter vision-language model for efficient on-device inference. OmniVLM introduces a token compression mechanism that reduces visual token sequence length from 729 to 81 tokens, significantly reducing computational overhead while preserving visual-semantic fidelity. Through a multi-stage training pipeline of pretraining, supervised fine-tuning, and minimal-edit Direct Preference Optimization (DPO), OmniVLM matches the performance of larger models. On multiple benchmarks including ScienceQA, POPE, and MMMU, OmniVLM outperforms existing baselines like nanoLLAVA within a 968M-parameter footprint. Empirical results on the same laptop demonstrate 9.1x faster time-to-first-token (0.75s vs 6.82s) and 1.5x higher decoding speed (29.41 vs 19.20 tokens/s) compared to nanoLLAVA, enabling efficient deployment on edge devices. The model weights can be accessed on huggingface: \url{https://huggingface.co/NexaAIDev/OmniVLM-968M}, and the inference examples can be find in Appendix B.

The broadcast model is widely used to describe the process of information dissemination from a single node to all nodes within an interconnected network. In this model, a graph represents the network, where vertices correspond to nodes and edges to communication links. The efficiency of this broadcasting process is evaluated by the broadcast time, the minimum discrete time units required to broadcast from a given vertex. Determining the broadcast time is referred to as the problem Broadcast Time. The set of vertices with the minimum broadcast time among the graph is called the broadcast center. Identifying this center or determining its size are both proven to be NP-hard. For a graph with n vertices, the minimum broadcast time is at least ceil(log2 n). The Broadcast Graph problem asks in a graph of n vertices, whether the broadcast time from any vertex equals ceil(log2 n). Extensive research over the past 50 years has focused on constructing broadcast graphs, which are optimal network topologies for one-to-all communication efficiency. However, the computational complexity of the Broadcast Graph problem has rarely been the subject of study. We believe that the difficulty lies in the mapping reduction for an NP-completeness proof. Consequently, we must construct broadcast graphs for yes-instances and non-broadcast graphs for no-instances. The most closely related result is the NP-completeness of Broadcast Time proved by Slater et al. in 1981. More recently, Fomin et al. has proved that Broadcast Time is fixed-parameter tractable. In this paper, we prove that Broadcast Graph is NP-complete by proving a reduction from Broadcast Time. We also improve the results on the complexity of the broadcast center problem. We show Broadcast Center Size is in delta^2_p, and is DP-hard, implying a complexity upper bound of delta^2_p-complete and a lower bound of DP-hard., Comment: 15 pages, 5 figures, for conference CALDAM 2025

The widespread application of large language models (LLMs) underscores the importance of deep learning (DL) technologies that rely on foundational DL libraries such as PyTorch and TensorFlow. Despite their robust features, these libraries face challenges with scalability and adaptation to rapid advancements in the LLM community. In response, tech giants like Apple and Huawei are developing their own DL libraries to enhance performance, increase scalability, and safeguard intellectual property. Ensuring the security of these libraries is crucial, with fuzzing being a vital solution. However, existing fuzzing frameworks struggle with target flexibility, effectively testing bug-prone API sequences, and leveraging the limited available information in new libraries. To address these limitations, we propose FUTURE, the first universal fuzzing framework tailored for newly introduced and prospective DL libraries. FUTURE leverages historical bug information from existing libraries and fine-tunes LLMs for specialized code generation. This strategy helps identify bugs in new libraries and uses insights from these libraries to enhance security in existing ones, creating a cycle from history to future and back. To evaluate FUTURE's effectiveness, we conduct comprehensive evaluations on three newly introduced DL libraries. Evaluation results demonstrate that FUTURE significantly outperforms existing fuzzers in bug detection, success rate of bug reproduction, validity rate of code generation, and API coverage. Notably, FUTURE has detected 148 bugs across 452 targeted APIs, including 142 previously unknown bugs. Among these, 10 have been assigned CVE IDs. Additionally, FUTURE detects 7 bugs in PyTorch, demonstrating its ability to enhance security in existing libraries in reverse., Comment: This paper has been accepted by 47th International Conference on Software Engineering (ICSE 2025)

Over the last several decades, extensive research has been conducted on the baryon cycles within cosmic structures, encompassing a broad mass range from dwarf galaxies to galaxy clusters. However, a notable gap in understanding the cosmic baryon cycle is the poor galaxy groups with halo masses around $10^{13}\ M_{\odot}$ (e.g., McGaugh et al. 2010). Poor galaxy groups, like our own Local Group, are prevalent throughout the universe, yet robust detection of their hot, X-ray emitting intragroup medium (IGrM) has remained elusive. The presence of this hot IGrM is crucial for addressing the long-standing "missing baryons" problem. Previous ROSAT-based studies were limited by a small number of X-ray bright samples, thus restricting the scope of their findings. Here we show a robust detection of this hot IGrM in a large, optically selected poor groups sample, based on the stacked X-ray images from the eROSITA Final Equatorial Depth Survey. These groups are identified in DESI LS with a mass range of log($M_\mathrm{halo}/h^{-1}M_{\odot}$) = 11.5-13.5 and a redshift range of z = 0.1-0.5. Additionally, our results indicate that despite its presence in virtually groups at all sizes, this gas component is still not sufficient to recover the universal baryon fraction, and hence the "missing baryons" problem still persists in poor galaxy groups., Comment: 12 pages, 6 figures, 1 table, published in ApJ Letters

The rapid advancement of lithium niobate on insulator (LNOI) photonics has spurred interest in approaches to develop ultra-narrow linewidth Brillouin microlasers. Here we demonstrate an integrated Brillouin microlaser with 118-Hz intrinsic linewidth and 3.15-mW threshold power in a dispersion engineered and suspended LNOI microdisk resonator of 116 um diameter. Benefited from the ultrahigh Q factor and sub-millimeter-scale microresonator, large Brillouin gain is attained via the backward intermodal SBS between the dual-resonant optical WGMs with a 10-GHz whispering-gallery mechanical mode, while satisfying the requirements of both energy and momentum conservations. Such strong optomechanical coupling up to 12.1 kHz is promising for a record narrow linewidth and a lowest stimulated Brillouin laser threshold value within sub-millimeter-scale integrated microresonators reported so far. This advancement in integrated ultra-narrow linewidth Brillouin lasers with compact cavity lengths paves the way for applications ranging from coherent information processing to precision metrology within the realm of high density photonic integration., Comment: 17 pages,4 figures

Detailed studies of two-neutrino double electron capture (2$\nu$DEC) is a crucial step towards searching for the neutrino-less mode to explore the Majorana nature of neutrinos. We have measured precisely the half-life of the 2$\nu$DEC process in $^{124}$Xe, utilizing a total exposure of 1.73 tonne$\cdot$year from the commissioning run and the first science run of the PandaX-4T experiment. A time-dependent background model in the $\mathcal{O}$(10 keV) energy is constructed for the first time in PandaX-4T data. With an unbinned maximum likelihood fit, we determine the half-life of the 2$\nu$DEC process to be $(1.03\pm0.15_{\rm stat}\pm0.08_{\rm sys})\times 10^{22}$$\,$yr. Furthermore, we have evaluated the branching ratio for both electrons captured from the $K$ shell ($KK$) to be $(65\pm5)\%$, which aligns with the $^{124}$Xe nuclear model calculations within 1.5$\,$$\sigma$., Comment: 19 pages, 5 figures, 4 tables

In this paper, we study the Picard group of the Baily-Borel compactification of orthogonal Shimura varieties. As a result, we determine the Picard group of the Baily-Borel compactification of the moduli space of quasi-polarized K3 surfaces. Interestingly, in contrast to the situation observed in the moduli space of curves, we find that the Picard group of the Baily-Borel compactification is isomorphic to $\mathbb{Z}$., Comment: 29 pages, any comments are welcome!

Estimated brain age from magnetic resonance image (MRI) and its deviation from chronological age can provide early insights into potential neurodegenerative diseases, supporting early detection and implementation of prevention strategies. Diffusion MRI (dMRI) presents an opportunity to build an earlier biomarker for neurodegenerative disease prediction because it captures subtle microstructural changes that precede more perceptible macrostructural changes. However, the coexistence of macro- and micro-structural information in dMRI raises the question of whether current dMRI-based brain age estimation models are leveraging the intended microstructural information or if they inadvertently rely on the macrostructural information. To develop a microstructure-specific brain age, we propose a method for brain age identification from dMRI that mitigates the model's use of macrostructural information by non-rigidly registering all images to a standard template. Imaging data from 13,398 participants across 12 datasets were used for the training and evaluation. We compare our brain age models, trained with and without macrostructural information mitigated, with an architecturally similar T1-weighted (T1w) MRI-based brain age model and two recent, popular, openly available T1w MRI-based brain age models that primarily use macrostructural information. We observe difference between our dMRI-based brain age and T1w MRI-based brain age across stages of neurodegeneration, with dMRI-based brain age being older than T1w MRI-based brain age in participants transitioning from cognitively normal (CN) to mild cognitive impairment (MCI), but younger in participants already diagnosed with Alzheimer's disease (AD). Furthermore, dMRI-based brain age may offer advantages over T1w MRI-based brain age in predicting the transition from CN to MCI up to five years before diagnosis.

This article is concerned with an inverse problem of simultaneously determining a spatially varying coefficient and a Robin coefficient for a one-dimensional fractional diffusion equation with a time-fractional derivative of order $\alpha\in(0,1)$. We prove the uniqueness for the inverse problem by observation data at one interior point over a finite time interval, provided that a coefficient is known on a subinterval. Our proof is based on the uniqueness in the inverse spectracl problem for a Sturm-Liouville problem by means of the Weyl $m$-function and the spectral representation of the solution to an initial-boundary value problem for the fractional diffusion equation.

The remarkable generalization ability of neural networks is usually attributed to the implicit bias of SGD, which often yields models with lower complexity using simpler (e.g. linear) and low-rank features. Recent works have provided empirical and theoretical evidence for the bias of particular variants of SGD (such as label noise SGD) toward flatter regions of the loss landscape. Despite the folklore intuition that flat solutions are 'simple', the connection with the simplicity of the final trained model (e.g. low-rank) is not well understood. In this work, we take a step toward bridging this gap by studying the simplicity structure that arises from minimizers of the sharpness for a class of two-layer neural networks. We show that, for any high dimensional training data and certain activations, with small enough step size, label noise SGD always converges to a network that replicates a single linear feature across all neurons; thereby, implying a simple rank one feature matrix. To obtain this result, our main technical contribution is to show that label noise SGD always minimizes the sharpness on the manifold of models with zero loss for two-layer networks. Along the way, we discover a novel property -- a local geodesic convexity -- of the trace of Hessian of the loss at approximate stationary points on the manifold of zero loss, which links sharpness to the geometry of the manifold. This tool may be of independent interest.

Adam outperforms SGD when training language models. Yet this advantage is not well-understood theoretically -- previous convergence analysis for Adam and SGD mainly focuses on the number of steps $T$ and is already minimax-optimal in non-convex cases, which are both $\widetilde{O}(T^{-1/4})$. In this work, we argue that the exploitation of nice $\ell_\infty$-geometry is the key advantage of Adam over SGD. More specifically, we give a new convergence analysis for Adam under novel assumptions that loss is smooth under $\ell_\infty$-geometry rather than the more common $\ell_2$-geometry, which yields a much better empirical smoothness constant for GPT-2 and ResNet models. Our experiments confirm that Adam performs much worse when the favorable $\ell_\infty$-geometry is changed while SGD provably remains unaffected. We also extend the convergence analysis to blockwise Adam under novel blockwise smoothness assumptions.

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

Training language models currently requires pre-determining a fixed compute budget because the typical cosine learning rate schedule depends on the total number of steps. In contrast, the Warmup-Stable-Decay (WSD) schedule uses a constant learning rate to produce a main branch of iterates that can in principle continue indefinitely without a pre-specified compute budget. Then, given any compute budget, one can branch out from the main branch at a proper time with a rapidly decaying learning rate to produce a strong model. Empirically, WSD generates a non-traditional loss curve: the loss remains elevated during the stable phase but sharply declines during the decay phase. Towards explaining this phenomenon, we conjecture that pretraining loss exhibits a river valley landscape, which resembles a deep valley with a river at its bottom. Under this assumption, we show that during the stable phase, the iterate undergoes large oscillations due to the high learning rate, yet it progresses swiftly along the river. During the decay phase, the rapidly dropping learning rate minimizes the iterate's oscillations, moving it closer to the river and revealing true optimization progress. Therefore, the sustained high learning rate phase and fast decaying phase are responsible for progress in the river and the mountain directions respectively, and are both critical. Our analysis predicts phenomenons consistent with empirical observations and shows that this landscape can emerge from pretraining on a simple bi-gram dataset. Inspired by the theory, we introduce WSD-S, a variant of WSD that reuses previous checkpoints' decay phases and keeps only one main branch, where we resume from a decayed checkpoint. WSD-S empirically outperforms WSD and Cyclic-Cosine in obtaining multiple language model checkpoints across various compute budgets in a single run for parameters scaling from 0.1B to 1.2B., Comment: 45 pages,13 figures

Diffusion-based extreme image compression methods have achieved impressive performance at extremely low bitrates. However, constrained by the iterative denoising process that starts from pure noise, these methods are limited in both fidelity and efficiency. To address these two issues, we present Relay Residual Diffusion Extreme Image Compression (RDEIC), which leverages compressed feature initialization and residual diffusion. Specifically, we first use the compressed latent features of the image with added noise, instead of pure noise, as the starting point to eliminate the unnecessary initial stages of the denoising process. Second, we directly derive a novel residual diffusion equation from Stable Diffusion's original diffusion equation that reconstructs the raw image by iteratively removing the added noise and the residual between the compressed and target latent features. In this way, we effectively combine the efficiency of residual diffusion with the powerful generative capability of Stable Diffusion. Third, we propose a fixed-step fine-tuning strategy to eliminate the discrepancy between the training and inference phases, thereby further improving the reconstruction quality. Extensive experiments demonstrate that the proposed RDEIC achieves state-of-the-art visual quality and outperforms existing diffusion-based extreme image compression methods in both fidelity and efficiency. The source code will be provided in https://github.com/huai-chang/RDEIC.

Proper quality control (QC) is time consuming when working with large-scale medical imaging datasets, yet necessary, as poor-quality data can lead to erroneous conclusions or poorly trained machine learning models. Most efforts to reduce data QC time rely on outlier detection, which cannot capture every instance of algorithm failure. Thus, there is a need to visually inspect every output of data processing pipelines in a scalable manner. We design a QC pipeline that allows for low time cost and effort across a team setting for a large database of diffusion weighted and structural magnetic resonance images. Our proposed method satisfies the following design criteria: 1.) a consistent way to perform and manage quality control across a team of researchers, 2.) quick visualization of preprocessed data that minimizes the effort and time spent on the QC process without compromising the condition or caliber of the QC, and 3.) a way to aggregate QC results across pipelines and datasets that can be easily shared. In addition to meeting these design criteria, we also provide information on what a successful output should be and common occurrences of algorithm failures for various processing pipelines. Our method reduces the time spent on QC by a factor of over 20 when compared to naively opening outputs in an image viewer and demonstrate how it can facilitate aggregation and sharing of QC results within a team. While researchers must spend time on robust visual QC of data, there are mechanisms by which the process can be streamlined and efficient., Comment: 22 pages, 12 figures, 1 table, 6 supplemental figures

Recent advancements in Vision-Language (VL) research have sparked new benchmarks for complex visual reasoning, challenging models' advanced reasoning ability. Traditional Vision-Language Models (VLMs) perform well in visual perception tasks while struggling with complex reasoning scenarios. Conversely, Large Language Models (LLMs) demonstrate robust text reasoning capabilities; however, they lack visual acuity. To bridge this gap, we propose Complex Visual Reasoning Large Language Models (CVR-LLM), capitalizing on VLMs' visual perception proficiency and LLMs' extensive reasoning capability. Unlike recent multimodal large language models (MLLMs) that require a projection layer, our approach transforms images into detailed, context-aware descriptions using an iterative self-refinement loop and leverages LLMs' text knowledge for accurate predictions without extra training. We also introduce a novel multi-modal in-context learning (ICL) methodology to enhance LLMs' contextual understanding and reasoning. Additionally, we introduce Chain-of-Comparison (CoC), a step-by-step comparison technique enabling contrasting various aspects of predictions. Our CVR-LLM presents the first comprehensive study across a wide array of complex visual reasoning tasks and achieves SOTA performance among all., Comment: EMNLP 2024 Main

An incomplete field-of-view (FOV) in diffusion magnetic resonance imaging (dMRI) can severely hinder the volumetric and bundle analyses of whole-brain white matter connectivity. Although existing works have investigated imputing the missing regions using deep generative models, it remains unclear how to specifically utilize additional information from paired multi-modality data and whether this can enhance the imputation quality and be useful for downstream tractography. To fill this gap, we propose a novel framework for imputing dMRI scans in the incomplete part of the FOV by integrating the learned diffusion features in the acquired part of the FOV to the complete brain anatomical structure. We hypothesize that by this design the proposed framework can enhance the imputation performance of the dMRI scans and therefore be useful for repairing whole-brain tractography in corrupted dMRI scans with incomplete FOV. We tested our framework on two cohorts from different sites with a total of 96 subjects and compared it with a baseline imputation method that treats the information from T1w and dMRI scans equally. The proposed framework achieved significant improvements in imputation performance, as demonstrated by angular correlation coefficient (p < 1E-5), and in downstream tractography accuracy, as demonstrated by Dice score (p < 0.01). Results suggest that the proposed framework improved imputation performance in dMRI scans by specifically utilizing additional information from paired multi-modality data, compared with the baseline method. The imputation achieved by the proposed framework enhances whole brain tractography, and therefore reduces the uncertainty when analyzing bundles associated with neurodegenerative., Comment: 20 pages; 8 figures

Faraday rotation is an important probe of the magnetic fields and magnetized plasma around active galactic nuclei (AGN) jets. We present a Faraday rotation measure image of the M87 jet between 85.2 GHz and 101.3 GHz with a resolution of ~2" with the Atacama Large Millimeter/submillimeter Array (ALMA). We found that the rotation measure (RM) of the M87 core is $\rm (4.5\pm 0.4)\times10^{4}\ rad\ m^{-2}$ with a low linear polarization fraction of $\rm (0.88\pm 0.08)\%$. The spatial RM gradient in the M87 jet spans a wide range from $\sim -2\times10^4\rm~rad\ m^{-2}$ to $\sim 3\times10^4\rm~rad\ m^{-2}$ with a typical uncertainty of $0.3\times10^4\rm~rad\ m^{-2}$. A comparison with previous RM measurements of the core suggests that the Faraday rotation of the core may originate very close to the super massive black hole (SMBH). Both an internal origin and an external screen with a rapidly varying emitting source could be possible. As for the jet, the RM gradient indicates a helical configuration of the magnetic field that persists up to kpc scale. Combined with the kpc-scale RM measurements at lower frequencies, we found that RM is frequency-dependent in the jet. One possible scenario to explain this dependence is that the kpc-scale jet has a trumpet-like shape and the jet coil unwinds near its end., Comment: 11pages, 5 figures. Accepted for publication in ApJ. Comments are welcome

Vision-language navigation (VLN) is a critical domain within embedded intelligence, requiring agents to navigate 3D environments based on natural language instructions. Traditional VLN research has focused on improving environmental understanding and decision accuracy. However, these approaches often exhibit a significant performance gap when agents are deployed in novel environments, mainly due to the limited diversity of training data. Expanding datasets to cover a broader range of environments is impractical and costly. We propose the Vision-Language Navigation with Continual Learning (VLNCL) paradigm to address this challenge. In this paradigm, agents incrementally learn new environments while retaining previously acquired knowledge. VLNCL enables agents to maintain an environmental memory and extract relevant knowledge, allowing rapid adaptation to new environments while preserving existing information. We introduce a novel dual-loop scenario replay method (Dual-SR) inspired by brain memory replay mechanisms integrated with VLN agents. This method facilitates consolidating past experiences and enhances generalization across new tasks. By utilizing a multi-scenario memory buffer, the agent efficiently organizes and replays task memories, thereby bolstering its ability to adapt quickly to new environments and mitigating catastrophic forgetting. Our work pioneers continual learning in VLN agents, introducing a novel experimental setup and evaluation metrics. We demonstrate the effectiveness of our approach through extensive evaluations and establish a benchmark for the VLNCL paradigm. Comparative experiments with existing continual learning and VLN methods show significant improvements, achieving state-of-the-art performance in continual learning ability and highlighting the potential of our approach in enabling rapid adaptation while preserving prior knowledge.

Vision Language Navigation in Continuous Environments (VLN-CE) represents a frontier in embodied AI, demanding agents to navigate freely in unbounded 3D spaces solely guided by natural language instructions. This task introduces distinct challenges in multimodal comprehension, spatial reasoning, and decision-making. To address these challenges, we introduce Cog-GA, a generative agent founded on large language models (LLMs) tailored for VLN-CE tasks. Cog-GA employs a dual-pronged strategy to emulate human-like cognitive processes. Firstly, it constructs a cognitive map, integrating temporal, spatial, and semantic elements, thereby facilitating the development of spatial memory within LLMs. Secondly, Cog-GA employs a predictive mechanism for waypoints, strategically optimizing the exploration trajectory to maximize navigational efficiency. Each waypoint is accompanied by a dual-channel scene description, categorizing environmental cues into 'what' and 'where' streams as the brain. This segregation enhances the agent's attentional focus, enabling it to discern pertinent spatial information for navigation. A reflective mechanism complements these strategies by capturing feedback from prior navigation experiences, facilitating continual learning and adaptive replanning. Extensive evaluations conducted on VLN-CE benchmarks validate Cog-GA's state-of-the-art performance and ability to simulate human-like navigation behaviors. This research significantly contributes to the development of strategic and interpretable VLN-CE agents.

Axion-like particles (ALPs) and dark photons (DPs) are viable dark matter particle candidates. We have searched for possible ALP/DP signals in the PandaX-4T liquid xenon detector using 94.8 days of data. A binned likelihood fit is constructed to search for possible mono-energetic peaks induced by the absorption processes between ALPs/DPs and atomic electrons of xenon. A detailed temporal model of decays associated with xenon isotopes is introduced to constrain the number of background events. No signal excess over background expectations is observed, and we have established the most stringent exclusion limits for most ALP/DP masses ranging from 150 keV/$c^2$ to 1 MeV/$c^2$.