Interpreting complex 3GPP telecommunications standards for question and answering (QA) poses a challenge for general-purpose LLMs due to their specialized terminology and high computational demands, limiting their use in resourceconstrained environments. This work explores an efficient, opensource approach using the TeleQnA dataset of 10,000 telecom questions and the TSpec-LLM repository of processed 3GPP documents. We enhance a lightweight Llama 3.2 (3B parameters) model, quantized from 16-bit precision to 4 bits, through finetuning and RAG to improve accuracy without heavy resource reliance. Unlike prior resource-intensive or proprietary solutions, our method reduces memory demands, enabling deployment on modest hardware like edge devices or softwarized networks. Shared via GitHub repositories [1], this approach advances costeffective, reproducible AI for telecommunications QA, supporting contexts where budgets, computation, or public internet access are limited.

Leandro De Souza, Thales Almeida, Roberto Lotufo, Rodrigo Frassetto Nogueira. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Recent work has shown that incorporating explanations into the output generated by large language models (LLMs) can significantly enhance performance on a broad spectrum of reasoning tasks. Our study extends these findings by demonstrating the benefits of explanations for neural rankers. By utilizing LLMs such as GPT-3.5 to enrich retrieval datasets with explanations, we trained a sequence-to-sequence ranking model, dubbed ExaRanker, to generate relevance labels and explanations for query-document pairs. The ExaRanker model, finetuned on a limited number of examples and synthetic explanations, exhibits performance comparable to models finetuned on three times more examples, but without explanations. Moreover, incorporating explanations imposes no additional computational overhead into the reranking step and allows for on-demand explanation generation. The codebase and datasets used in this study will be available at https://github.com/unicamp-dl/ExaRanker

The zero-shot cross-lingual ability of models pretrained on multilingual and even monolingual corpora has spurred many hypotheses to explain this intriguing empirical result. However, due to the costs of pretraining, most research uses public models whose pretraining methodology, such as the choice of tokenization, corpus size, and computational budget, might differ drastically. When researchers pretrain their own models, they often do so under a constrained budget, and the resulting models might underperform significantly compared to SOTA models. These experimental differences led to various inconsistent conclusions about the nature of the cross-lingual ability of these models. To help further research on the topic, we released 10 monolingual byte-level models rigorously pretrained under the same configuration with a large compute budget (equivalent to 420 days on a V100) and corpora that are 4 times larger than the original BERT’s. Because they are tokenizer-free, the problem of unseen token embeddings is eliminated, thus allowing researchers to try a wider range of cross-lingual experiments in languages with different scripts. Additionally, we release two models pretrained on non-natural language texts that can be used in sanity-check experiments. Experiments on QA and NLI tasks show that our monolingual models achieve competitive performance to the multilingual one, and hence can be served to strengthen our understanding of cross-lingual transferability in language models.

Text recognition tasks are commonly solved by using a deep learning pipeline called CRNN. The classical CRNN is a sequence of a convolutional network, followed by a bidirectional LSTM and a CTC layer. In this paper, we perform an extensive analysis of the components of a CRNN to find what is crucial to the entire pipeline and what characteristics can be exchanged for a more effective choice. Given the results of our experiments, we propose two different architectures for the task of text recognition. The first model, CNN + CTC, is a combination of a convolutional model followed by a CTC layer. The second model, CNN + Tr + CTC, adds an encoder-only Transformers between the convolutional network and the CTC layer. To the best of our knowledge, this is the first time that a Transformers have been successfully trained using just CTC loss. To assess the capabilities of our proposed architectures, we train and evaluate them on the SROIE 2019 data set. Our CNN + CTC achieves an F1 score of 89.66% possessing only 4.7 million parameters. CNN + Tr + CTC attained an F1 score of 93.76% with 11 million parameters, which is almost 97% of the performance achieved by the TrOCR using 334 million parameters and more than 600 million synthetic images for pretraining.

There has been mounting evidence that pretrained language models fine-tuned on large and diverse supervised datasets can transfer well to a variety of out-of-domain tasks. In this work, we investigate this transfer ability to the legal domain. For that, we participated in the legal case entailment task of COLIEE 2021, in which we use such models with no adaptations to the target domain. Our submissions achieved the highest scores, surpassing the second-best submission by more than six percentage points. Our experiments confirm a counter-intuitive result in the new paradigm of pretrained language models: given limited labeled data, models with little or no adaption to the target task can be more robust to changes in the data distribution than models fine-tuned on it. Code is available at https://github.com/neuralmind-ai/coliee.

This paper presents a novel approach named extinction profiles to model the spatial information of remote sensing images. Then, the output of the extinction profile is fed to a grid-search random forest classification method. Results indicate that the proposed approach can effectively extract spatial information from remote sensing gray scale images and provide high classification accuracies in an automatic way.

This letter proposes a new approach for the spectral-spatial classification of hyperspectral images, which is based on a novel extrema-oriented connected filtering technique, entitled as extended extinction profiles. The proposed approach progressively simplifies the first informative features extracted from hyperspectral data considering different attributes. Then, the classification approach is applied on two well-known hyperspectral data sets, i.e., Pavia University and Indian Pines, and compared with one of the most powerful filtering approaches in the literature, i.e., extended attribute profiles. Results indicate that the proposed approach is able to efficiently extract spatial information for the classification of hyperspectral images automatically and swiftly. In addition, an array-based node-oriented max-tree representation was carried out to efficiently implement the proposed approach.

With respect to recent advances in remote sensing technologies, the spatial resolution of airborne and spaceborne sensors is getting finer, which enables us to precisely analyze even small objects on the Earth. This fact has made the research area of developing efficient approaches to extract spatial and contextual information highly active. Among the existing approaches, morphological profile and attribute profile (AP) have gained great attention due to their ability to classify remote sensing data. This paper proposes a novel approach that makes it possible to precisely extract spatial and contextual information from remote sensing images. The proposed approach is based on extinction filters, which are used here for the first time in the remote sensing community. Then, the approach is carried out on two well-known high-resolution panchromatic data sets captured over Rome, Italy, and Reykjavik, Iceland. In order to prove the capabilities of the proposed approach, the obtained results are compared with the results from one of the strongest approaches in the literature, i.e., APs, using different points of view such as classification accuracies, simplification rate, and complexity analysis. Results indicate that the proposed approach can significantly outperform its alternative in terms of classification accuracies. In addition, based on our implementation, profiles can be generated in a very short processing time. It should be noted that the proposed approach is fully automatic.

With the growing use of biometric authentication systems in the recent years, spoof fingerprint detection has become increasingly important. In this paper, we use convolutional neural networks (CNNs) for fingerprint liveness detection. Our system is evaluated on the data sets used in the liveness detection competition of the years 2009, 2011, and 2013, which comprises almost 50 000 real and fake fingerprints images. We compare four different models: two CNNs pretrained on natural images and fine-tuned with the fingerprint images, CNN with random weights, and a classical local binary pattern approach. We show that pretrained CNNs can yield the state-of-the-art results with no need for architecture or hyperparameter selection. Data set augmentation is used to increase the classifiers performance, not only for deep architectures but also for shallow ones. We also report good accuracy on very small training sets (400 samples) using these large pretrained networks. Our best model achieves an overall rate of 97.1% of correctly classified samples-a relative improvement of 16% in test error when compared with the best previously published results. This model won the first prize in the fingerprint liveness detection competition 2015 with an overall accuracy of 95.5%.

This paper presents an array-based node-oriented structure for the max-tree representation, which allows direct access and flexible manipulation of its nodes, and is more suitable for OpenMP parallel processing. The proposed structure is based on two arrays called node array (NA), which stores attributes of the nodes, and node index (NI), which indicates the node that each pixel belongs to. We compare it with the pixel-oriented max-tree representation based on a parent array (parent) and an ordering array (S) that allows tree traversals. We show that our max-tree representation requires less memory when the ratio between the number of image pixels and max-tree nodes is greater than 1.6, which is often the case. It is more flexible, and can compute some attributes, such as height and dynamics, with a complexity linear on the number of max-tree nodes instead of the number of image pixels. In our experiments our structure computed the height attribute on average 11.4 faster than the parent/S representation. Also, for a single area-open filter, the sequential implementation of our structure is on average 1.14 times slower and the parallel implementation in a 4-core CPU is 1.2 times faster than the parent/S structure. For an area-open filter followed by the hmax filter, our sequential implementation is 1.34 times faster and our parallel implementation is 2.32 times faster than the parent/S structure.

The Max-Tree is an efficient data structure that represents all connected components resulting from all possible image upper threshold values. Usually, most of its nodes represent irrelevant extrema, i.e. noise, or small variations of a connected component. This paper proposes the Maximal Max-Tree Simplification (MMS) filter with a normalized threshold criterion (MMS-T) and a Maximally Stable Extremal Regions (MSER) criterion (MMS-MSER) and a methodology to apply them using the Extinction filter We show that after applying our simplification methodology which sets the number of maxima in the image, the number of Max-Tree nodes is at most twice this number. Two applications of the proposed methodology are illustrated.

In this work, a new type of watershed transform is introduced: the tie-zone watershed (TZWS). This region-based watershed transform does not depend on arbitrary implementation and provides a unique and optimal solution. Indeed, many solutions are sometimes possible when segmenting an image with a watershed algorithm. In this case, the TZWS assigns each pixel to a catchment basin (CB) if in all solutions it belongs to this CB. Otherwise, the pixel is said to belong to a tie-zone (TZ). We propose an efficient algorithm based on image foresting transform (IFT) which computes the TZWS transform as a shortest-path forest. Finally, two applications of this TZWS are presented: bounding intervals for segmented objects' extensions and a progressive segmentation procedure.

In this work we discuss an approach to characterize the human cortex in MRI data through the segmentation of the cortical layer with deep penetration into the sulci, using the Image Foresting Transform. This technique, based on graph theory, provides a sound framework for the implementation of many image processing operators, and has been successfully applied to the solution of many problems, including some in medical imaging. We show the results of applying the technique to slices of MRI data from four subjects and compare these to the computation of the morphological skeleton.

This work consists in the study, development and implementation of a toolbox for image processing using the Python language and the numerical Python package. This set has "open source" distribution and is adequate for multidimensional mathematical processing. Python is a modern and well projected language, interpreted, "very-high-level", object oriented and extremely portable, apart from being suitable for rapid application development (RAD). The system is generated using the methodology of the Adesso project for construction of scientific software. This environment is useful in education, research and development of final applications.

The notion of fuzzy connectedness captures the idea of "hanging-togetherness" of image elements in an object by assigning a strength of connectedness to every possible path between every possible pair of image elements. This concept leads to powerful image segmentation algorithms based on dynamic programming whose effectiveness has been demonstrated on 1,000s of images in a variety of applications. In the previous framework, a fuzzy connected object is defined with a threshold on the strength of connectedness. We introduce the notion of relative connectedness that overcomes the need for a threshold and that leads to more effective segmentations. The central idea is that an object gets defined in an image because of the presence of other co-objects. Each object is initialized by a seed element. An image element c is considered to belong to that object with respect to whose reference image element c has the highest strength of connectedness. In this fashion, objects compete among each other utilizing fuzzy connectedness to grab membership of image elements. We present a theoretical and algorithmic framework for defining objects via relative connectedness and demonstrate utilizing the theory that the objects defined are independent of reference elements chosen as long as they are not in the fuzzy boundary between objects. An iterative strategy is also introduced wherein the strongest relative connected core parts are first defined and iteratively relaxed to conservatively capture the more fuzzy parts subsequently. Examples from medical imaging are presented to illustrate visually the effectiveness of relative fuzzy connectedness. A quantitative mathematical phantom study involving 160 images is conducted to demonstrate objectively the effectiveness of relative fuzzy connectedness.DisclaimerA claim of priority in research and publication appeared on page 1486 in the paper "Relative Fuzzy Connectedness and Object Definition: Theory, Algorithms, and Applications in Image Segmentation" by J.K. Udupa, P.K. Saha, and' R.A. Lotufo (IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 11, pp. 1485-1500,Nov. 2002) with respect to the paper "Multiseeded Segmentation Using Fuzzy Connectedness by G.T. Herman and B.M. Carvalho" (IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 23, no. 5, pp. 460-474, May 2001). Furthermore, the wording of this claim suggests professional misconduct on the part of G.T. Herman and B.M. Carvalho. Responsibility for the content of published papers rests with the authors. The peer review process is intended to determine the overall significance of the technical contribution of a manuscript. The peer review process does not provide a way to validate every statement in a manuscript. In particular, the IEEE has not validated the claim referred to in the first paragraph above. The IEEE regrets publishing this unauthenticated statement and the pain that such publication may have caused to G.T. Herman and B.M. Carvalho.

The estimation of the number of people in an area under surveillance is very important for the problem of crowd monitoring. When an area reaches an occupation level greater than the projected one, people's safety can be in danger. This paper describes a new technique for crowd density estimation based on Minkowski fractal dimension. The fractal dimension has been widely used to characterize data texture in a large number of physical and biological sciences. The results of our experiments show that fractal dimension can also be used to characterize levels of people congestion in images of crowds. The proposed technique is compared with a statistical and a spectral technique, in a test study of nearly 300 images of a specific area of the Liverpool Street Railway Station, London, UK. Results obtained in this test study are presented.

In multidimensional image analysis, there are, and will continue to be, situations wherein automatic image segmentation methods fail, calling for considerable user assistance in the process. The main goals of segmentation research for such situations ought to be (i) to provideeffective controlto the user on the segmentation processwhileit is being executed, and (ii) to minimize the total user's time required in the process. With these goals in mind, we present in this paper two paradigms, referred to aslive wireandlive lane, for practical image segmentation in large applications. For both approaches, we think of the pixel vertices and oriented edges as forming a graph, assign a set of features to each oriented edge to characterize its ``boundariness,'' and transform feature values to costs. We provide training facilities and automatic optimal feature and transform selection methods so that these assignments can be made with consistent effectiveness in any application. In live wire, the user first selects an initial point on the boundary. For any subsequent point indicated by the cursor, an optimal path from the initial point to the current point is found and displayed in real time. The user thus has a live wire on hand which is moved by moving the cursor. If the cursor goes close to the boundary, the live wire snaps onto the boundary. At this point, if the live wire describes the boundary appropriately, the user deposits the cursor which now becomes the new starting point and the process continues. A few points (live-wire segments) are usually adequate to segment the whole 2D boundary. In live lane, the user selects only the initial point. Subsequent points are selected automatically as the cursor is moved within a lane surrounding the boundary whose width changes as a function of the speed and acceleration of cursor motion. Live-wire segments are generated and displayed in real time between successive points. The users get the feeling that the curve snaps onto the boundary as and while they roughly mark in the vicinity of the boundary. We describe formal evaluation studies to compare the utility of the new methods with that of manual tracing based on speed and repeatability of tracing and on data taken from a large ongoing application. The studies indicate that the new methods are statistically significantly more repeatable and 1.5–2.5 times faster than manual tracing.

The authors have developed an interactive environment for the creation and maintenance of dynamic, active multimedia-based teaching mechanisms. The environment is designed to be user-friendly and to facilitate the creation of educational material. This tool has already been used to create courses in multidimensional signal processing (MDSP) by researchers working together while geographically separated.

This paper presents a technique for oriented texture classification which is based on the Hough transform and Kohonen's (1990) neural network model. In this technique, oriented texture features are extracted from the Hough space by means of two distinct strategies. While the first operates on a non-uniformly sampled Hough space, the second concentrates on the peaks produced in the Hough space. The described technique gives good results for the classification of oriented textures, a common phenomenon in nature underlying an important class of images. Experimental results are presented to demonstrate the performance of the new technique in comparison with an implemented technique based on Gabor filters.

Khorosware, which is the concept of using Khoros as an "authoring system" was first reported in ICIP-94. We present here the effort of integrating the Khoros 2 system, hypertext (HTML), and network browsers to produce an interactive digital image processing course. The success of the Khorosware lies on the the broad scope of the Khoros system which encourages and facilitates collaboration. This paper describes how the course is organized, how it can impact teaching, learning, and research processes, and gives examples of how it is currently being used in several settings. This new interactive toolbox/course is available free of charge via the Internet at http://www.Khoros. Unm.Edul dip course.

The Khoros scientific software environment has become a popular tool for teaching image processing. The broad scope of Khoros and the communities emphasis on collaboration have contributed to this success. The paper describes how Khoros can impact the learning process and gives an example of how it can be utilized by an instructor to enhance an image processing course. The concept of thinking of Khoros as an "authoring system" is introduced and supported by briefly explaining the components of Khoros that make up the authoring system.>