623509 research outputs found
Sort by
No Argument Left Behind: Overlapping Chunks for Faster Processing of Arbitrarily Long Legal Texts
No full textIn a context where the Brazilian judiciary system, the largest in the world, faces a crisis due to the slow processing of millions of cases, it becomes imperative to develop efficient methods for analyzing legal texts. We introduce uBERT, a hybrid model that combines Transformer and Recurrent Neural Network architectures to effectively handle long legal texts. Our approach processes the full text regardless of its length while maintaining reasonable computational overhead. Our experiments demonstrate that uBERT achieves superior performance compared to BERT+LSTM when overlapping input is used and is significantly faster than ULMFiT for processing long legal documents.Presented at 15th Symposium in Information and Human Language Technology (STIL) @ BRACIS\u272
Stationary acoustic black hole solutions in Bose-Einstein condensates and their Borel analysis
No full textIn this article, we study the dynamics of a Bose-Einstein condensate (BEC) with the idea of finding solutions that could possibly correspond to a so-called acoustic (or Unruh) black/white holes. Those are flows with horizons where the speed of the flow goes from sub-sonic to super-sonic. This is because sound cannot go back from the supersonic to the subsonic region. The speed of sound plays the role of the speed of light in a gravitational black hole, an important difference being that there are excitations that can go faster than the speed of sound and therefore can escape the sonic black hole. Here, the motion of the BEC is described by the Gross-Pitaevskii Equation (GPE). More concretely, we discuss singular Stationary solutions of Gross-Pitaevskii equation in 2D (with Circular symmetry) and 3D (with Spherical symmetry). We use these solutions to study the local speed of sound and magnitude of flow velocity of the condensate to see whether they cross, indicating the potential existence of a sonic analog of a black/white hole. We discuss numerical techniques used and also study the semi-analytical Laplace-Borel resummation of asymptotic series solutions to see how well they agree with numerical solutions. We also study how the resurgent transseries plays a role in these solutions.Added and contextualized more references in Section 1, citing relevant previous literature. Updated acknowledgement
Spectro-temporal evolution of 4U 1702-429 using AstroSat-NICER
No full textWe present the broadband spectral and timing properties of the atoll source 4U 1702-429 using two observations of AstroSat with the second one having simultaneous NICER data. For both observations, the spectra can be represented by a Comptonizing medium with a black body seed photon source which can be identified with the surface of the neutron star. A disk emission along with a distant reflection is also required for both spectra. For the first observation, the coronal temperature ( keV) is smaller than the second ( keV), and the disk is truncated at a larger radius, km, compared to the second, km, for an assumed distance of 7 kpc. A kHz QPO at Hz is detected in the first and is absent in the second observation. Modeling the energy-dependent r.m.s and time lag of the kHz QPO reveals a corona size of 30 km. A similar model can explain the energy dependence of the broadband noise at Hz for the second observation. The results suggest that kHz QPOs are associated with a compact corona surrounding the neutron star and may occur when the disk is truncated at large distances. We emphasize the need for more wide-band observations of the source to confirm these results.12 Pages, 7 Figures, Accepted for publication in The Astrophysical Journa
Beyond adaptive gradient: Fast-Controlled Minibatch Algorithm for large-scale optimization
No full textAdaptive gradient methods have been increasingly adopted by deep learning community due to their fast convergence and reduced sensitivity to hyper-parameters. However, these methods come with limitations, such as increased memory requirements for elements like moving averages and a poorly understood convergence theory. To overcome these challenges, we introduce F-CMA, a Fast-Controlled Mini-batch Algorithm with a random reshuffling method featuring a sufficient decrease condition and a line-search procedure to ensure loss reduction per epoch, along with its deterministic proof of global convergence to a stationary point. To evaluate the F-CMA, we integrate it into conventional training protocols for classification tasks involving both convolutional neural networks and vision transformer models, allowing for a direct comparison with popular optimizers. Computational tests show significant improvements, including a decrease in the overall training time by up to 68%, an increase in per-epoch efficiency by up to 20%, and in model accuracy by up to 5%.There is an error in the literature review, in section 1. In particular, we noticed that there is a wrong citation, the [65], which has been erroneously associated with another author\u27s claim
Curriculum Demonstration Selection for In-Context Learning
No full textLarge Language Models (LLMs) have shown strong in-context learning (ICL) abilities with a few demonstrations. However, one critical challenge is how to select demonstrations to elicit the full potential of LLMs. In this paper, we propose Curriculum Demonstration Selection (CDS), a novel demonstration selection method for ICL. Instead of merely using similarity, CDS additionally partitions samples by their complexity measurements. Following curriculum learning, CDS then selects demonstrations from easy to difficult. Thus the selected demonstrations cover a wide range of difficulty levels, enabling LLMs to learn from varied complexities within the training set. Experiments demonstrate that our CDS consistently outperforms baseline methods, achieving notable improvements across nine LLMs on three benchmarks. Moreover, CDS proves especially effective in enhancing LLM performance in solving challenging problems.Accepted at the 40th ACM/SIGAPP Symposium On Applied Computing (SAC 2025), Main Conferenc
Computing the multimodal stochastic dynamics of a nanobeam in a viscous fluid
No full textThe stochastic dynamics of small elastic objects in fluid are central to many important and emerging technologies. It is now possible to measure and use the higher modes of motion of elastic structures when driven by Brownian motion alone. Although theoretical descriptions exist for idealized conditions, computing the stochastic multimodal dynamics for the complex conditions of experiment is very challenging. We show that this is possible using deterministic finite element calculations with the fluctuation dissipation theorem by exploring the multimodal stochastic dynamics of a doubly-clamped nanobeam. We use a very general, and flexible, finite-element computational approach to quantify the stochastic dynamics of multiple modes simultaneously using only a single deterministic simulation. We include the experimentally relevant features of an intrinsic tension in the beam and the influence of a nearby rigid boundary on the dynamics through viscous fluid interactions. We quantify the stochastic dynamics of the first eleven flexural modes of the beam when immersed in air or water. We compare the numerical results with theory, where possible, and find excellent agreement. We quantify the limitations of the computational approach and describe its range of applicability. These results pave the way for computational studies of the stochastic dynamics of complex 3D elastic structures in a viscous fluid where theoretical descriptions are not available.29 pages, 5 figure
EACO: Enhancing Alignment in Multimodal LLMs via Critical Observation
No full textMultimodal large language models (MLLMs) have achieved remarkable progress on various visual question answering and reasoning tasks leveraging instruction fine-tuning specific datasets. They can also learn from preference data annotated by human to enhance their reasoning ability and mitigate hallucinations. Most of preference data is generated from the model itself. However, existing methods require high-quality critical labels, which are costly and rely on human or proprietary models like GPT-4V. In this work, we propose Enhancing Alignment in MLLMs via Critical Observation (EACO), which aligns MLLMs by self-generated preference data using only 5k images economically. Our approach begins with collecting and refining a Scoring Evaluation Instruction-tuning dataset to train a critical evaluation model, termed the Critic. This Critic observes model responses across multiple dimensions, selecting preferred and non-preferred outputs for refined Direct Preference Optimization (DPO) tuning. To further enhance model performance, we employ an additional supervised fine-tuning stage after preference tuning. EACO reduces the overall hallucinations by 65.6% on HallusionBench and improves the reasoning ability by 21.8% on MME-Cognition. EACO achieves an 8.5% improvement over LLaVA-v1.6-Mistral-7B across multiple benchmarks. Remarkably, EACO also shows the potential critical ability in open-source MLLMs, demonstrating that EACO is a viable path to boost the competence of MLLMs.19 page
Characterization of subordinate symmetric Markov processes
No full textIn this paper, we consider subordinate symmetric Markov processes which correspond to non-killing Dirichlet forms enjoying heat kernel estimates on a metric measure space with the volume doubling property. We obtain estimates of the jump kernel of the subordinate process and establish equivalent conditions for the jump kernel following Liu-Murugan. In particular, we clarify the scale of the jump kernel, which is different from the diffusion type. This result is appliable to non-subordinate processes by the transferring method, which uses stability of Dirichlet forms
Block Coordinate Descent Methods for Structured Nonconvex Optimization with Nonseparable Constraints: Optimality Conditions and Global Convergence
No full textCoordinate descent algorithms are widely used in machine learning and large-scale data analysis due to their strong optimality guarantees and impressive empirical performance in solving non-convex problems. In this work, we introduce Block Coordinate Descent (BCD) method for structured nonconvex optimization with nonseparable constraints. Unlike traditional large-scale Coordinate Descent (CD) approaches, we do not assume the constraints are separable. Instead, we account for the possibility of nonlinear coupling among them. By leveraging the inherent problem structure, we propose new CD methods to tackle this specific challenge. Under the relatively mild condition of locally bounded non-convexity, we demonstrate that achieving coordinate-wise stationary points offer a stronger optimality criterion compared to standard critical points. Furthermore, under the Luo-Tseng error bound conditions, our BCD methods exhibit Q-linear convergence to coordinate-wise stationary points or critical points. To demonstrate the practical utility of our methods, we apply them to various machine learning and signal processing models. We also provide the geometry analysis for the models. Experiments on real-world data consistently demonstrate the superior objective values of our approaches compared to existing methods
EvoSampling: A Granular Ball-based Evolutionary Hybrid Sampling with Knowledge Transfer for Imbalanced Learning
No full textClass imbalance would lead to biased classifiers that favor the majority class and disadvantage the minority class. Unfortunately, from a practical perspective, the minority class is of importance in many real-life applications. Hybrid sampling methods address this by oversampling the minority class to increase the number of its instances, followed by undersampling to remove low-quality instances. However, most existing sampling methods face difficulties in generating diverse high-quality instances and often fail to remove noise or low-quality instances on a larger scale effectively. This paper therefore proposes an evolutionary multi-granularity hybrid sampling method, called EvoSampling. During the oversampling process, genetic programming (GP) is used with multi-task learning to effectively and efficiently generate diverse high-quality instances. During the undersampling process, we develop a granular ball-based undersampling method that removes noise in a multi-granular fashion, thereby enhancing data quality. Experiments on 20 imbalanced datasets demonstrate that EvoSampling effectively enhances the performance of various classification algorithms by providing better datasets than existing sampling methods. Besides, ablation studies further indicate that allowing knowledge transfer accelerates the GP\u27s evolutionary learning process