Search

Data contamination -- the accidental consumption of evaluation examples within the pre-training data -- can undermine the validity of evaluation benchmarks. In this paper, we present a rigorous analysis of the effects of contamination on language models at 1B and 8B scales on the machine translation task. Starting from a carefully decontaminated train-test split, we systematically introduce contamination at various stages, scales, and data formats to isolate its effect and measure its impact on performance metrics. Our experiments reveal that contamination with both source and target substantially inflates BLEU scores, and this inflation is 2.5 times larger (up to 30 BLEU points) for 8B compared to 1B models. In contrast, source-only and target-only contamination generally produce smaller, less consistent over-estimations. Finally, we study how the temporal distribution and frequency of contaminated samples influence performance over-estimation across languages with varying degrees of data resources.

In [ Ge], Bert van Geemen computed the dimension of the space of the fourth power of the theta nullwerte. In [SM2], it has been observe that all linear relations between the $\theta_m^4$ are consequences of the quartic Riemann relations. In this note, we want to give a new proof of these result and extend them.

Dynamic inverse problems are challenging to solve due to the need to identify and incorporate appropriate regularization in both space and time. Moreover, the very large scale nature of such problems in practice presents an enormous computational challenge. In this work, in addition to the use of edge-enhancing regularization of spatial features, we propose a new regularization method that incorporates a temporal model that estimates the motion of objects in time. In particular, we consider the optical flow model that simultaneously estimates the motion and provides an approximation for the desired image, and we incorporate this information into the cost functional as an additional form of temporal regularization. We propose a computationally efficient algorithm to solve the jointly regularized problem that leverages a generalized Krylov subspace method. We illustrate the effectiveness of the prescribed approach on a wide range of numerical experiments, including limited angle and single-shot computerized tomography., Comment: 27 pages, 6 figures, 5 tables

We report the confirmation and characterization of four hot Jupiter-type exoplanets initially detected by TESS: TOI-1295 b, TOI-2580 b, TOI-6016 b, and TOI-6130 b. Using observations with the high-resolution echelle spectrograph MaHPS on the 2.1m telescope at Wendelstein Observatory, together with NEID at Kitt Peak National Observatory and TRES at the Fred Lawrence Whipple Observatory, we confirmed the planetary nature of these four planet candidates. We also performed precise mass measurements. All four planets are found to be hot Jupiters with orbital periods between 2.4 and 4.0 days. The sizes of these planets range from 1.29 to 1.64 Jupiter radii, while their masses range from 0.6 to 1.5 Jupiter masses. Additionally, we investigated whether there are signs of other planets in the systems but have found none. Lastly, we compared the radii of our four objects to the results of an empirical study of radius inflation and see that all four demonstrate a good fit with the current models. These four planets belong to the first array of planets confirmed with MaHPS data, supporting the ability of the spectrograph to detect planets around fainter stars as faint as V=12.

The increasing urbanization and the growing number of vehicles in cities have underscored the need for efficient parking management systems. Traditional smart parking solutions often rely on sensors or cameras for occupancy detection, each with its limitations. Recent advancements in deep learning have introduced new YOLO models (YOLOv8, YOLOv9, YOLOv10, and YOLOv11), but these models have not been extensively evaluated in the context of smart parking systems, particularly when combined with Region of Interest (ROI) selection for object detection. Existing methods still rely on fixed polygonal ROI selections or simple pixel-based modifications, which limit flexibility and precision. This work introduces a novel approach that integrates Internet of Things, Edge Computing, and Deep Learning concepts, by using the latest YOLO models for vehicle detection. By exploring both edge and cloud computing, it was found that inference times on edge devices ranged from 1 to 92 seconds, depending on the hardware and model version. Additionally, a new pixel-wise post-processing ROI selection method is proposed for accurately identifying regions of interest to count vehicles in parking lot images. The proposed system achieved 99.68% balanced accuracy on a custom dataset of 3,484 images, offering a cost-effective smart parking solution that ensures precise vehicle detection while preserving data privacy, Comment: Submitted to Elsevier Internet of Things, 22 pages, 11 figures, 6 tables

As LLMs continue to become more powerful and versatile, human evaluation has quickly become intractable at scale and reliance on automatic metrics has become the norm. Recently, it has been shown that LLMs are themselves state-of-the-art evaluators for many tasks. These Autoraters are typically designed so that they generalize to new systems and test sets. In practice, however, evaluation is performed on a small set of fixed, canonical test sets, which are carefully curated to measure certain capabilities of interest and are not changed frequently. In this work, we design a method which specializes a prompted Autorater to a given test set, by leveraging historical ratings on the test set to construct in-context learning (ICL) examples. We evaluate our Specialist method on the task of fine-grained machine translation evaluation, and show that it dramatically outperforms the state-of-the-art XCOMET metric by 54% and 119% on the WMT'23 and WMT'24 test sets, respectively. We perform extensive analyses to understand the representations learned by our Specialist metrics, and how variability in rater behavior affects their performance. We also verify the generalizability and robustness of our Specialist method for designing automatic metrics across different numbers of ICL examples, LLM backbones, systems to evaluate, and evaluation tasks.

Objective: Enhancing the reliability of myoelectric controllers that decode motor intent is a pressing challenge in the field of bionic prosthetics. State-of-the-art research has mostly focused on Supervised Learning (SL) techniques to tackle this problem. However, obtaining high-quality labeled data that accurately represents muscle activity during daily usage remains difficult. We investigate the potential of Reinforcement Learning (RL) to further improve the decoding of human motion intent by incorporating usage-based data. Methods: The starting point of our method is a SL control policy, pretrained on a static recording of electromyographic (EMG) ground truth data. We then apply RL to fine-tune the pretrained classifier with dynamic EMG data obtained during interaction with a game environment developed for this work. We conducted real-time experiments to evaluate our approach and achieved significant improvements in human-in-the-loop performance. Results: The method effectively predicts simultaneous finger movements, leading to a two-fold increase in decoding accuracy during gameplay and a 39\% improvement in a separate motion test. Conclusion: By employing RL and incorporating usage-based EMG data during fine-tuning, our method achieves significant improvements in accuracy and robustness. Significance: These results showcase the potential of RL for enhancing the reliability of myoelectric controllers, of particular importance for advanced bionic limbs. See our project page for visual demonstrations: https://sites.google.com/view/bionic-limb-rl

The collection of speech data carried out in Sociolinguistics has the potential to enhance large language models due to its quality and representativeness. In this paper, we examine the ethical considerations associated with the gathering and dissemination of such data. Additionally, we outline strategies for addressing the sensitivity of speech data, as it may facilitate the identification of informants who contributed with their speech., Comment: in Portuguese language. Paper accepted to LAAI-Ethics 2024

While Minimum Bayes Risk (MBR) decoding using metrics such as COMET or MetricX has outperformed traditional decoding methods such as greedy or beam search, it introduces a challenge we refer to as metric bias. As MBR decoding aims to produce translations that score highly according to a specific utility metric, this very process makes it impossible to use the same metric for both decoding and evaluation, as improvements might simply be due to reward hacking rather than reflecting real quality improvements. In this work we find that compared to human ratings, neural metrics not only overestimate the quality of MBR decoding when the same metric is used as the utility metric, but they also overestimate the quality of MBR/QE decoding with other neural utility metrics as well. We also show that the metric bias issue can be mitigated by using an ensemble of utility metrics during MBR decoding: human evaluations show that MBR decoding using an ensemble of utility metrics outperforms a single utility metric., Comment: To appear at WMT2024

This study examines the applicability of the Cloze test, a widely used tool for assessing text comprehension proficiency, while highlighting its challenges in large-scale implementation. To address these limitations, an automated correction approach was proposed, utilizing Natural Language Processing (NLP) techniques, particularly word embeddings (WE) models, to assess semantic similarity between expected and provided answers. Using data from Cloze tests administered to students in Brazil, WE models for Brazilian Portuguese (PT-BR) were employed to measure the semantic similarity of the responses. The results were validated through an experimental setup involving twelve judges who classified the students' answers. A comparative analysis between the WE models' scores and the judges' evaluations revealed that GloVe was the most effective model, demonstrating the highest correlation with the judges' assessments. This study underscores the utility of WE models in evaluating semantic similarity and their potential to enhance large-scale Cloze test assessments. Furthermore, it contributes to educational assessment methodologies by offering a more efficient approach to evaluating reading proficiency.

Linguistic diversity is a human attribute which, with the advance of generative AIs, is coming under threat. This paper, based on the contributions of sociolinguistics, examines the consequences of the variety selection bias imposed by technological applications and the vicious circle of preserving a variety that becomes dominant and standardized because it has linguistic documentation to feed the large language models for machine learning., Comment: in Portuguese language. paper aceepted to LAAI-Ethics 2024

Reinforcement learning (RL) shows promise in control problems, but its practical application is often hindered by the complexity arising from intricate reward functions with constraints. While the reward hypothesis suggests these competing demands can be encapsulated in a single scalar reward function, designing such functions remains challenging. Building on existing work, we start by formulating preferences over trajectories to derive a realistic reward function that balances goal achievement with constraint satisfaction in the application of mobile robotics with dynamic obstacles. To mitigate reward exploitation in such complex settings, we propose a novel two-stage reward curriculum combined with a flexible replay buffer that adaptively samples experiences. Our approach first learns on a subset of rewards before transitioning to the full reward, allowing the agent to learn trade-offs between objectives and constraints. After transitioning to a new stage, our method continues to make use of past experiences by updating their rewards for sample-efficient learning. We investigate the efficacy of our approach in robot navigation tasks and demonstrate superior performance compared to baselines in terms of true reward achievement and task completion, underlining its effectiveness.

Despite growing interest in incorporating feedback to improve language models, most efforts focus only on sequence-level annotations. In this work, we explore the potential of utilizing fine-grained span-level annotations from offline datasets to improve model quality. We develop a simple finetuning algorithm, called Training with Annotations (TWA), to directly train machine translation models on such annotated data. TWA utilizes targeted span-level error information while also flexibly learning what to penalize within a span. Moreover, TWA considers the overall trajectory of a sequence when deciding which non-error spans to utilize as positive signals. Experiments on English-German and Chinese-English machine translation show that TWA outperforms baselines such as Supervised FineTuning on sequences filtered for quality and Direct Preference Optimization on pairs constructed from the same data.

Collecting high-quality translations is crucial for the development and evaluation of machine translation systems. However, traditional human-only approaches are costly and slow. This study presents a comprehensive investigation of 11 approaches for acquiring translation data, including human-only, machineonly, and hybrid approaches. Our findings demonstrate that human-machine collaboration can match or even exceed the quality of human-only translations, while being more cost-efficient. Error analysis reveals the complementary strengths between human and machine contributions, highlighting the effectiveness of collaborative methods. Cost analysis further demonstrates the economic benefits of human-machine collaboration methods, with some approaches achieving top-tier quality at around 60% of the cost of traditional methods. We release a publicly available dataset containing nearly 18,000 segments of varying translation quality with corresponding human ratings to facilitate future research.

Different of biases are reproduced in LLM-generated responses, including dialectal biases. A study based on prompt engineering was carried out to uncover how LLMs discriminate varieties of Brazilian Portuguese, specifically if sociolinguistic rules are taken into account in four LLMs: GPT 3.5, GPT-4o, Gemini, and Sabi.-2. The results offer sociolinguistic contributions for an equity fluent NLP technology., Comment: 8 pages, XI Jornada de Descri\c{c}\~ao do Portugu\^es

Quantum entanglement is a key resource for quantum technologies, including emerging ground-to-satellite quantum communication. In such a scenario, an important challenge to be overcome is to consider entanglement between two or more quantum particles in different inertial frames, potentially experiencing relativistic effects affecting quantum correlations. In this paper, we present a consistent framework that overcomes this challenge. To this end, we establish the notion of factorizable and entangled multi-time trajectories and derive a class of Euler--Lagrange equations under the constraint of a non-entangling behavior. Comparing this restricted evolution to the solutions of the unrestricted equations of motion allows one to investigate the trajectory-based entanglement of general systems. We solve our equations for interacting particles in a Klein--Gordon-type setting, thereby quantifying the dynamic and relativistic impact of entanglement in a self-consistent manner.

In this paper, we present the MetricX-24 submissions to the WMT24 Metrics Shared Task and provide details on the improvements we made over the previous version of MetricX. Our primary submission is a hybrid reference-based/-free metric, which can score a translation irrespective of whether it is given the source segment, the reference, or both. The metric is trained on previous WMT data in a two-stage fashion, first on the DA ratings only, then on a mixture of MQM and DA ratings. The training set in both stages is augmented with synthetic examples that we created to make the metric more robust to several common failure modes, such as fluent but unrelated translation, or undertranslation. We demonstrate the benefits of the individual modifications via an ablation study, and show a significant performance increase over MetricX-23 on the WMT23 MQM ratings, as well as our new synthetic challenge set., Comment: Accepted to WMT24

This paper investigates the impact of verbose LLM translations on evaluation. We first demonstrate the prevalence of this behavior across several LLM outputs drawn from the WMT 2024 general shared task on machine translation. We then identify the primary triggers of verbosity, including safety, copyright concerns, and insufficient context in short input queries. Finally, we show that ignoring this behavior unfairly penalizes more verbose LLMs according to both automatic and human evaluations, highlighting the need to address this issue for more accurate future evaluations.

Deep learning-based methods have been widely researched in the areas of language and vision, demonstrating their capacity to understand long sequences of data and their usefulness in numerous helio-physics applications. Foundation models (FMs), which are pre-trained on a large-scale datasets, form the basis for a variety of downstream tasks. These models, especially those based on transformers in vision and language, show exceptional potential for adapting to a wide range of downstream applications. In this paper, we provide our perspective on the criteria for designing an FM for heliophysics and associated challenges and applications using the Solar Dynamics Observatory (SDO) dataset. We believe that this is the first study to design an FM in the domain of heliophysics., Comment: 31 Pages, 12 figures

In this paper we present a step-by-step approach to long-form text translation, drawing on established processes in translation studies. Instead of viewing machine translation as a single, monolithic task, we propose a framework that engages language models in a multi-turn interaction, encompassing pre-translation research, drafting, refining, and proofreading, resulting in progressively improved translations. Extensive automatic evaluations using Gemini 1.5 Pro across ten language pairs show that translating step-by-step yields large translation quality improvements over conventional zero-shot prompting approaches and earlier human-like baseline strategies, resulting in state-of-the-art results on WMT2024.

Recent research in neural machine translation (NMT) has shown that training on high-quality machine-generated data can outperform training on human-generated data. This work accompanies the first-ever release of a LLM-generated, MBR-decoded and QE-reranked dataset with both sentence-level and multi-sentence examples. We perform extensive experiments to demonstrate the quality of our dataset in terms of its downstream impact on NMT model performance. We find that training from scratch on our (machine-generated) dataset outperforms training on the (web-crawled) WMT'23 training dataset (which is 300 times larger), and also outperforms training on the top-quality subset of the WMT'23 training dataset. We also find that performing self-distillation by finetuning the LLM which generated this dataset outperforms the LLM's strong few-shot baseline. These findings corroborate the quality of our dataset, and demonstrate the value of high-quality machine-generated data in improving performance of NMT models.

This is the preliminary ranking of WMT24 General MT systems based on automatic metrics. The official ranking will be a human evaluation, which is superior to the automatic ranking and supersedes it. The purpose of this report is not to interpret any findings but only provide preliminary results to the participants of the General MT task that may be useful during the writing of the system submission.

A nonintrusive model order reduction method for bilinear stochastic differential equations with additive noise is proposed. A reduced order model (ROM) is designed in order to approximate the statistical properties of high-dimensional systems. The drift and diffusion coefficients of the ROM are inferred from state observations by solving appropriate least-squares problems. The closeness of the ROM obtained by the presented approach to the intrusive ROM obtained by the proper orthogonal decomposition (POD) method is investigated. Two generalisations of the snapshot-based dominant subspace construction to the stochastic case are presented. Numerical experiments are provided to compare the developed approach to POD.

Minimum Bayes Risk (MBR) decoding is a powerful decoding strategy widely used for text generation tasks, but its quadratic computational complexity limits its practical application. This paper presents a novel approach for approximating MBR decoding using matrix completion techniques, focusing on the task of machine translation. We formulate MBR decoding as a matrix completion problem, where the utility metric scores between candidate hypotheses and pseudo-reference translations form a low-rank matrix. First, we empirically show that the scores matrices indeed have a low-rank structure. Then, we exploit this by only computing a random subset of the scores and efficiently recover the missing entries in the matrix by applying the Alternating Least Squares (ALS) algorithm, thereby enabling a fast approximation of the MBR decoding process. Our experimental results on machine translation tasks demonstrate that the proposed method requires 1/16 utility metric computations compared to vanilla MBR decoding while achieving equal translation quality measured by COMET22 on the WMT22 dataset (en<>de and en<>ru). We also benchmark our method against other approximation methods and we show gains in quality when comparing to them.

Preservation of the Nasca geoglyphs at the UNESCO World Heritage Site in Peru is urgent as natural and human impact accelerates. More frequent weather extremes such as flashfloods threaten Nasca artifacts. We demonstrate that runoff models based on (sub-)meter scale, LiDAR-derived digital elevation data can highlight AI-detected geoglyphs that are in danger of erosion. We recommend measures of mitigation to protect the famous "lizard", "tree", and "hand" geoglyphs located close by, or even cut by the Pan-American Highway., Comment: accepted at IGARSS 2024

We introduce some properties describing dependence in indiscernible sequences: $F_{ind}$ and its dual $F_{Mb}$, the definable Morley property, and $n$-resolvability. Applying these properties, we establish the following results: We show that the degree of nonminimality introduced by Freitag and Moosa, which is closely related to $F_{ind}$ (equal in $\mathrm{DCF}_{0}$), may take on any positive integer value in an $\omega$-stable theory, answering a question of Freitag, Jaoui, and Moosa. Proving a conjecture of Koponen, we show that every simple theory with quantifier elimination in a finite relational language has finite rank and is one-based. The arguments closely rely on finding types $q$ with $F_{Mb}(q) = \infty$, and on $n$-resolvability. We prove some variants of the simple Kim-forking conjecture, a generalization of the stable forking conjecture to $\mathrm{NSOP}_{1}$ theories. We show a global analogue of the simple Kim-forking conjecture with infinitely many variables holds in every $\mathrm{NSOP}_{1}$ theory, and show that Kim-forking with a realization of a type $p$ with $\mathrm{F}_{Mb}(p) < \infty$ satisfies a finite-variable version of this result. We then show, in a low $\mathrm{NSOP}_{1}$ theory or when $p$ is isolated, if $p \in S(C)$ has the definable Morley property for Kim-independence, Kim-forking with realizations of $p$ gives a nontrivial instance of the simple Kim-forking conjecture itself. In particular, when $F_{Mb}(p) < \infty$ and $|S^{F_{Mb}(p) + 1}(C)| < \infty$, Kim-forking with realizations of $p$ gives us a nontrivial instance of the simple Kim-forking conjecture. We show that the quantity $F_{Mb}$, motivated in simple and $\mathrm{NSOP}_{1}$ theories by the above results, is in fact nontrivial even in stable theories., Comment: 63 pages

Linguistic ambiguity continues to represent a significant challenge for natural language processing (NLP) systems, notwithstanding the advancements in architectures such as Transformers and BERT. Inspired by the recent success of instructional models like ChatGPT and Gemini (In 2023, the artificial intelligence was called Bard.), this study aims to analyze and discuss linguistic ambiguity within these models, focusing on three types prevalent in Brazilian Portuguese: semantic, syntactic, and lexical ambiguity. We create a corpus comprising 120 sentences, both ambiguous and unambiguous, for classification, explanation, and disambiguation. The models capability to generate ambiguous sentences was also explored by soliciting sets of sentences for each type of ambiguity. The results underwent qualitative analysis, drawing on recognized linguistic references, and quantitative assessment based on the accuracy of the responses obtained. It was evidenced that even the most sophisticated models, such as ChatGPT and Gemini, exhibit errors and deficiencies in their responses, with explanations often providing inconsistent. Furthermore, the accuracy peaked at 49.58 percent, indicating the need for descriptive studies for supervised learning., Comment: in Portuguese language, 16 p\'aginas, 5 p\'aginas de ap\^endice e 4 imagens

Reliable human evaluation is critical to the development of successful natural language generation models, but achieving it is notoriously difficult. Stability is a crucial requirement when ranking systems by quality: consistent ranking of systems across repeated evaluations is not just desirable, but essential. Without it, there is no reliable foundation for hill-climbing or product launch decisions. In this paper, we use machine translation and its state-of-the-art human evaluation framework, MQM, as a case study to understand how to set up reliable human evaluations that yield stable conclusions. We investigate the optimal configurations for item allocation to raters, number of ratings per item, and score normalization. Our study on two language pairs provides concrete recommendations for designing replicable human evaluation studies. We also collect and release the largest publicly available dataset of multi-segment translations rated by multiple professional translators, consisting of nearly 140,000 segment annotations across two language pairs., Comment: To appear at NAACL 2024

Various factors influence the degree of cybersickness a user can suffer in an immersive virtual environment, some of which can be controlled without adapting the virtual environment itself. When using HMDs, one example is the size of the field of view. However, the degree to which factors like this can be manipulated without affecting the user negatively in other ways is limited. Another prominent characteristic of cybersickness is that it affects individuals very differently. Therefore, to account for both the possible disruptive nature of alleviating factors and the high interpersonal variance, a promising approach may be to intervene only in cases where users experience discomfort symptoms, and only as much as necessary. Thus, we conducted a first experiment, where the field of view was decreased when people feel uncomfortable, to evaluate the possible positive impact on sickness and negative influence on presence. While we found no significant evidence for any of these possible effects, interesting further results and observations were made.