搜索结果：Ever

Mixed Reality (MR)-aided operation overlays digital objects on the physical world to provide a more immersive and intuitive operation process. A primary challenge is the precise and fast auto-verification of whether the user follows MR guidance by comparing frames before and after each operation. The pre-operation frame includes virtual guiding objects, while the post-operation frame contains physical counterparts. Existing approaches fall short of accounting for the discrepancies between physical and virtual objects due to imperfect 3D modeling or lighting estimation. In this paper, we propose EVER: an edge-assisted auto-verification system for mobile MR-aided operations. Unlike traditional frame-based similarity comparisons, EVER leverages the segmentation model and rendering pipeline adapted to the unique attributes of frames with physical pieces and those with their virtual counterparts; it adopts a threshold-based strategy using Intersection over Union (IoU) metrics for accurate auto-verification. To ensure fast auto-verification and low energy consumption, EVER offloads compute-intensive tasks to an edge server. Through comprehensive evaluations of public datasets and custom

We report an instrumental observation of the very exceptional Geminid fireball which was observed in scope of the Czech part of the European Fireball Network (EN) on 13 December 2012 at 4h12m59.4s UT. The uniqueness of this Geminid fireball consists of the record depth of its penetration in the atmosphere (to the height of 32.5 km) and in the fact that most likely a very small fraction of its initial mass survived severe deceleration in the atmosphere and landed on the ground. Such deeply penetrating Geminid with so precise and reliable data has not yet been observed. From a comparison with a large number of Geminids observed by the European Fireball Network and all brightest Geminids from the Prairie Fireball Network in USA and the Canadian MORP Network, we have shown that for Geminids with an entry mass greater than approximately 10 grams, the terminal altitude limit does not decrease further as it does for smaller Geminids, but remains constant at around 38 km. In this comparison, we have shown that there is only one exception, and that is the Geminid presented here. This one penetrated nearly 6 km deeper with very low terminal speed for Geminids. During the atmospheric flight t

Solar radius measurements and their variations -- if any -- are a difficult problem that has vexed researchers for decades. In this paper, we have attempted to clarify the various ways of expressing the definition ''solar diameter'', from a physical point of view. The concept of diameter is taken here in its broadest sense, leaving aside the issue concerning the oblateness caused by surface and internal angular velocity variations, as deviations from sphericity are negligible in our context. Astrometric time-series observations are still needed, and we advocate strengthening long-term metrological measures to achieve greater consensus on the subject. To date, modern observations of the solar diameter provide a frame of reference, and we give a new glossary. By comparing the best values obtained to date, it is shown that the ''seismic radius'' obtained from the Solar and Heliospheric Observatory (SOHO) and the Solar Dynamics Observatory (SDO) provides the best determination, a finding supported by observations made at the Calern (F) and Pic du Midi (F) observatories. The latest results on the leptocline show that it is more important than ever to consider at which layers of the Sun

The training process of reinforcement learning often suffers from severe oscillations, leading to instability and degraded performance. In this paper, we propose a Constant in an Ever-Changing World (CIC) framework that enhances algorithmic stability to improve performance. CIC maintains both a representative policy and a current policy. Instead of updating the representative policy blindly, CIC selectively updates it only when the current policy demonstrates superiority. Furthermore, CIC employs an adaptive adjustment mechanism, enabling the representative and current policies to jointly facilitate critic training. We evaluate CIC on five MuJoCo environments, and the results show that CIC improves the performance of conventional algorithms without incurring additional computational cost.

Reasoning is the fundamental capability of large language models (LLMs). Due to the rapid progress of LLMs, there are two main issues of current benchmarks: i) these benchmarks can be crushed in a short time (less than 1 year), and ii) these benchmarks may be easily hacked. To handle these issues, we propose the ever-scalingness for building the benchmarks which are scaling over complexity against crushing, instance against hacking and exploitation, oversight for easy verification, and coverage for real-world relevance. This paper presents Nondeterministic Polynomial-time Problem Challenge (NPPC), an ever-scaling reasoning benchmark for LLMs. Specifically, the NPPC has three main modules: i) npgym, which provides a unified interface of 25 well-known NP-complete problems and can generate any number of instances with any levels of complexities, ii) npsolver, which provides a unified interface to evaluate the problem instances with both online and offline models via APIs and local deployments, respectively, and iii) npeval, which provides the comprehensive and ready-to-use tools to analyze the performances of LLMs over different problems, the number of tokens, the reasoning errors and

As foundation models grow rapidly in capability and deployment, evaluating their scientific understanding becomes increasingly critical. Existing science benchmarks have made progress towards broad Range, wide Reach, and high Rigor, yet they often face two major challenges: data leakage risks that compromise benchmarking validity, and evaluation inefficiency due to large-scale testing. To address these issues, we introduce the Ever-Evolving Science Exam (EESE), a dynamic benchmark designed to reliably assess scientific capabilities in foundation models. Our approach consists of two components: 1) a non-public EESE-Pool with over 100K expertly constructed science instances (question-answer pairs) across 5 disciplines and 500+ subfields, built through a multi-stage pipeline ensuring Range, Reach, and Rigor, 2) a periodically updated 500-instance subset EESE, sampled and validated to enable leakage-resilient, low-overhead evaluations. Experiments on 32 open- and closed-source models demonstrate that EESE effectively differentiates the strengths and weaknesses of models in scientific fields and cognitive dimensions. Overall, EESE provides a robust, scalable, and forward-compatible solu

In recent years, extensive work has explored the application of the Transformer architecture to reinforcement learning problems. Among these, Decision Transformer (DT) has gained particular attention in the context of offline reinforcement learning due to its ability to frame return-conditioned policy learning as a sequence modeling task. Most recently, Bhargava et al. (2024) provided a systematic comparison of DT with more conventional MLP-based offline RL algorithms, including Behavior Cloning (BC) and Conservative Q-Learning (CQL), and claimed that DT exhibits superior performance in sparse-reward and low-quality data settings. In this paper, through experimentation on robotic manipulation tasks (Robomimic) and locomotion benchmarks (D4RL), we show that MLP-based Filtered Behavior Cloning (FBC) achieves competitive or superior performance compared to DT in sparse-reward environments. FBC simply filters out low-performing trajectories from the dataset and then performs ordinary behavior cloning on the filtered dataset. FBC is not only very straightforward, but it also requires less training data and is computationally more efficient. The results therefore suggest that DT is not p

While people generally trust AI to make decisions in various aspects of their lives, concerns arise when AI is involved in decisions with significant moral implications. The absence of a precise mathematical framework for moral reasoning intensifies these concerns, as ethics often defies simplistic mathematical models. Unlike fields such as logical reasoning, reasoning under uncertainty, and strategic decision-making, which have well-defined mathematical frameworks, moral reasoning lacks a broadly accepted framework. This absence raises questions about the confidence we can place in AI's moral decision-making capabilities. The environments in which AI systems are typically trained today seem insufficiently rich for such a system to learn ethics from scratch, and even if we had an appropriate environment, it is unclear how we might bring about such learning. An alternative approach involves AI learning from human moral decisions. This learning process can involve aggregating curated human judgments or demonstrations in specific domains, or leveraging a foundation model fed with a wide range of data. Still, concerns persist, given the imperfections in human moral decision making. Giv

We introduce an innovative RAG-based framework with an ever-improving memory. Inspired by humans'pedagogical process, RAM utilizes recursively reasoning-based retrieval and experience reflections to continually update the memory and learn from users' communicative feedback, namely communicative learning. Extensive experiments with both simulated and real users demonstrate significant improvements over traditional RAG and self-knowledge methods, particularly excelling in handling false premise and multi-hop questions. Furthermore, RAM exhibits promising adaptability to various feedback and retrieval methods, showcasing its potential for advancing AI capabilities in dynamic knowledge acquisition and lifelong learning.

Multi-wavelength polarimetry and radio observations of Swift J1727.8-1613 at the beginning of its recent 2023 outburst suggested the presence of a bright compact jet aligned in the north-south direction, which could not be confirmed without high angular resolution images. Using the Very Long Baseline Array and the Long Baseline Array, we imaged Swift J1727.8-1613, during the hard/hard-intermediate state, revealing a bright core and a large, two-sided, asymmetrical, resolved jet. The jet extends in the north-south direction, at a position angle of $-0.60\pm0.07°$ East of North. At 8.4 GHz, the entire resolved jet structure is $\sim110 (d/2.7\,\text{kpc})/\sin i$ AU long, with the southern approaching jet extending $\sim80 (d/2.7\,\text{kpc})/\sin i$ AU from the core, where $d$ is the distance to the source and $i$ is the inclination of the jet axis to the line of sight. These images reveal the most resolved continuous X-ray binary jet, and possibly the most physically extended continuous X-ray binary jet ever observed. Based on the brightness ratio of the approaching and receding jets, we put a lower limit on the intrinsic jet speed of $β\geq0.27$ and an upper limit on the jet incli

We present Exact Volumetric Ellipsoid Rendering (EVER), a method for real-time differentiable emission-only volume rendering. Unlike recent rasterization based approach by 3D Gaussian Splatting (3DGS), our primitive based representation allows for exact volume rendering, rather than alpha compositing 3D Gaussian billboards. As such, unlike 3DGS our formulation does not suffer from popping artifacts and view dependent density, but still achieves frame rates of $\sim\!30$ FPS at 720p on an NVIDIA RTX4090. Since our approach is built upon ray tracing it enables effects such as defocus blur and camera distortion (e.g. such as from fisheye cameras), which are difficult to achieve by rasterization. We show that our method is more accurate with fewer blending issues than 3DGS and follow-up work on view-consistent rendering, especially on the challenging large-scale scenes from the Zip-NeRF dataset where it achieves sharpest results among real-time techniques.

Large Language Models (LLMs) have demonstrated remarkable proficiency in generating fluent text. However, they often encounter the challenge of generating inaccurate or hallucinated content. This issue is common in both non-retrieval-based generation and retrieval-augmented generation approaches, and existing post-hoc rectification methods may not address the accumulated hallucination errors that may be caused by the "snowballing" issue, especially in reasoning tasks. To tackle these challenges, we introduce a novel approach called Real-time Verification and Rectification (Ever). Instead of waiting until the end of the generation process to rectify hallucinations, Ever employs a real-time, step-wise generation and hallucination rectification strategy. The primary objective is to detect and rectify hallucinations as they occur during the text generation process. When compared to both retrieval-based and non-retrieval-based baselines, Ever demonstrates a significant improvement in generating trustworthy and factually accurate text across a diverse range of tasks, including short-form QA, biography generation, and multi-hop reasoning.

Surprising as it may seem, applications of statistical methods to physics were inspired by the social sciences, which in turn are linked to the humanities. So perhaps it is not as unlikely as it might first appear for a group of statistical physicists and humanists to come together to investigate one of the subjects of Thomas Jefferson's poetic interests from a scientific point of view. And that is the nature of this article: a collaborative interdisciplinary analysis of the works of a figure Jefferson described as a ''rude bard of the North'' and ''the greatest Poet that has ever existed.'' In 2012, a subset of this team embraced an increase in interdisciplinary methods to apply the new science of complex networks to longstanding questions in comparative mythology. Investigations of network structures embedded in epic narratives allowed universal properties to be identified and ancient texts to be compared to each other. The approach inspired new challenges in mathematics, physics and even processes in industry, thereby illustrating how collaborations of this nature can be mutually beneficial and can capture the attention of a public, often ill-served by academic communication and

The recently discovered stellar system Ursa Major III/UNIONS 1 (UMa3/U1) is the faintest known Milky Way satellite to date. With a stellar mass of $16^{+6}_{-5}\,\rm M_\odot$ and a half-light radius of $3\pm1$pc, it is either the darkest galaxy ever discovered or the faintest self-gravitating star cluster known to orbit the Galaxy. Its line-of-sight velocity dispersion suggests the presence of dark matter, although current measurements are inconclusive because of the unknown contribution to the dispersion of potential binary stars. We use $N$-body simulations to show that, if self-gravitating, the system could not survive in the Milky Way tidal field for much longer than a single orbit (roughly 0.4Gyr), which strongly suggests that the system is stabilized by the presence of large amounts of dark matter. If UMa3/U1 formed at the center of a ~$10^9\rm M_\odot$ cuspy LCDM halo, its velocity dispersion would be predicted to be of order ~1km/s. This is roughly consistent with the current estimate, which, neglecting binaries, places $σ_{\rm los}$ in the range 1 to 4km/s. Because of its dense cusp, such a halo should be able to survive the Milky Way tidal field, keeping UMa3/U1 relativel

In the life cycle of highly automated systems operating in an open and dynamic environment, the ability to adjust to emerging challenges is crucial. For systems integrating data-driven AI-based components, rapid responses to deployment issues require fast access to related data for testing and reconfiguration. In the context of automated driving, this especially applies to road obstacles that were not included in the training data, commonly referred to as out-of-distribution (OoD) road obstacles. Given the availability of large uncurated recordings of driving scenes, a pragmatic approach is to query a database to retrieve similar scenarios featuring the same safety concerns due to OoD road obstacles. In this work, we extend beyond identifying OoD road obstacles in video streams and offer a comprehensive approach to extract sequences of OoD road obstacles using text queries, thereby proposing a way of curating a collection of OoD data for subsequent analysis. Our proposed method leverages the recent advances in OoD segmentation and multi-modal foundation models to identify and efficiently extract safety-relevant scenes from unlabeled videos. We present a first approach for the novel

This paper aims to develop a rigorous asymptotic analysis of an approximate renormalization group recursion for inverse participation ratios $P_q$ of critical powerlaw random band matrices. The recursion goes back to the work by Mirlin and Evers [37] and earlier works by Levitov [32, 33] and is aimed to describe the ensuing multifractality of the eigenvectors of such matrices. We point out both similarities and dissimilarities of LME recursion to those appearing in the theory of multiplicative cascades and branching random walks and show that the methods developed in those fields can be adapted to the present case. In particular the LME recursion is shown to exhibit a phase transition, which we expect is a freezing transition, where the role of temperature is played by the exponent $q$. However, the LME recursion has features that make its rigorous analysis considerably harder and we point out several open problems for further study

We show the softest ever spectrum from Cyg~X-1, detected in 2013 with Suzaku. This has the weakest high energy Compton tail ever seen from this object, so should give the cleanest view of the underlying disk spectrum, and hence the best determination of black hole spin from disk continuum fitting. Using the standard model of a disk with simple non-thermal Comptonisation to produce the weak high energy tail gives a high spin black hole. However, we get a significantly better fit by including an additional, low temperature thermal Comptonisation component, which allows a much lower black hole spin. Corroboration of the existence of an additional Compton component comes from the frequency dependent hard lags seen in the rapid variability in archival high/soft state data. These can not be explained if the continuum is a single non-thermal Comptonisation component, but are instead consistent with a radially stratified, multi zone Comptonisation spectrum, where the spectrum is softer further from the black hole. A complex multi-zone Comptonisation continuum is required to explain both spectra and timing together, and this has an impact on the derived black hole spin.

The eXtreme Deep Field (XDF) combines data from ten years of observations with the HST Advanced Camera for Surveys (ACS) and the Wide-Field Camera 3 Infra-Red (WFC3/IR) into the deepest image of the sky ever in the optical/near-IR. Since the initial observations on the Hubble Ultra-Deep Field (HUDF) in 2003, numerous surveys and programs, including supernova followup, HUDF09, CANDELS, and HUDF12 have contributed additional imaging data across the HUDF region. Yet these have never been combined and made available as one complete ultra-deep optical and near-infrared image dataset. We do so now for the eXtreme Deep Field (XDF) program. Our new and improved processing techniques provide higher quality reductions of the total dataset. All WFC3 near-IR and optical ACS data sets have been fully combined and accurately matched, resulting in the deepest imaging ever taken at these wavelengths ranging from 29.1 to 30.3 AB mag (5sigma in a 0.35" diameter aperture) in 9 filters. The gains in the optical for the four filters done in the original ACS HUDF correspond to a typical improvement of 0.15 mag, with gains of 0.25 mag in the deepest areas. Such gains are equivalent to adding ~130 to ~240