Brazil

Inertial Measurement Unit (IMU) sensors are present in everyday devices such as smartphones and fitness watches. As a result, the array of health-related research and applications that tap onto this data has been growing, but little attention has been devoted to the prediction of an individual&#39;s heart rate (HR) from IMU data, when undergoing a physical activity. Would that be even possible? If so, this could be used to design personalized sets of aerobic exercises, for instance. In this work, we show that it is viable to obtain accurate HR predictions from IMU data using Recurrent Neural Networks, provided only access to HR and IMU data from a short-lived, previously executed activity. We propose a novel method for initializing an RNN&#39;s hidden state vectors, using a specialized network that attempts to extract an embedding of the physical conditioning (PCE) of a subject. We show that using a discriminator in the training phase to help the model learn whether two PCEs belong to the same individual further reduces the prediction error. We evaluate the proposed model when predicting the HR of 23 subjects performing a variety of physical activities from IMU data available in public datasets (PAMAP2, PPG-DaLiA). For comparison, we use as baselines the only model specifically proposed for this task and an adapted state-of-the-art model for Human Activity Recognition (HAR), a closely related task. Our method, PCE-LSTM, yields over 10% lower mean absolute error. We demonstrate empirically that this error reduction is in part due to the use of the PCE. Last, we use the two datasets (PPG-DaLiA, WESAD) to show that PCE-LSTM can also be successfully applied when photoplethysmography (PPG) sensors are available, outperforming the state-of-the-art deep learning baselines by more than 30%.

BRACIS 2021

Encoding Physical Conditioning from Inertial Sensors for Multi-step Heart Rate Estimation

heart-rate

time-series

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-23 is the Thirty-Seventh AAAI Conference on Artificial Intelligence. The theme of this conference is to create collaborative bridges within and beyond AI. Like previous AAAI conferences, AAAI-23 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and two new activities: a Bridge Program and a Lab Program. Many of these activities are tailored to the theme of bridges and all are selected according to the highest standards, with additional programs for students and young researchers. 
AAAI is providing you with a conference planner, which you can use to help organize your itinerary of activities. This includes talks to attend in person, talks to attend remotely, breaks with colleagues and your site seeing activities. To access this conference planner, please go to [https://aaai-2023.takemobi.io](https://aaai-2023.takemobi.io).

In order to access this site, you need to register. If you haven't already, please register [here](https://aaai.org/Conferences/AAAI-23/registration/).


AAAI 2023

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines.

Inertial Measurement Unit (IMU) sensors are present in everyday devices such as smartphones and fitness watches. As a result, the array of health-related research and applications that tap onto this data has been growing, but little attention has been devoted to the prediction of an individual's heart rate (HR) from IMU data, when undergoing a physical activity. Would that be even possible? If so, this could be used to design personalized sets of aerobic exercises, for instance. In this work, we show that it is viable to obtain accurate HR predictions from IMU data using Recurrent Neural Networks, provided only access to HR and IMU data from a short-lived, previously executed activity. We propose a novel method for initializing an RNN's hidden state vectors, using a specialized network that attempts to extract an embedding of the physical conditioning (PCE) of a subject. We show that using a discriminator in the training phase to help the model learn whether two PCEs belong to the same individual further reduces the prediction error. We evaluate the proposed model when predicting the HR of 23 subjects performing a variety of physical activities from IMU data available in public datasets (PAMAP2, PPG-DaLiA). For comparison, we use as baselines the only model specifically proposed for this task and an adapted state-of-the-art model for Human Activity Recognition (HAR), a closely related task. Our method, PCE-LSTM, yields over 10% lower mean absolute error. We demonstrate empirically that this error reduction is in part due to the use of the PCE. Last, we use the two datasets (PPG-DaLiA, WESAD) to show that PCE-LSTM can also be successfully applied when photoplethysmography (PPG) sensors are available, outperforming the state-of-the-art deep learning baselines by more than 30%.

technical paper

**BRACIS** originated from the combination of the two most important scientific events in Brazil in Artificial Intelligence (AI) and Computational Intelligence (CI): the Brazilian Symposium on Artificial Intelligence – **SBIA** (22 editions), and the Brazilian Symposium on Neural Networks – **SBRN** (13 editions). **ENIAC** is a national forum for researchers, practitioners, educators and students to present and discuss innovations, trends, experiences, developments and work in progress in the fields of Artificial and Computational Intelligence. **STIL** is the bi-annual Language Technology event.
All events are supported by the Brazilian Computer Society – Special Interest Groups: **CE-IA**, **CE-IC** and **CE-PLN**.

To access the 10th Brazilian Conference on Intelligent Systems (**BRACIS**), you need to register for the conference.  Please visit the Conference official site to register

[Register now](http://c4ai.inova.usp.br/bracis2021/)

Register now

BRACIS originated from the combination of the two most important scientific events in Brazil in Artificial Intelligence (AI) and Computational Intelligence (CI): the Brazilian Symposium on Artificial Intelligence – SBIA (22 editions), and the Brazilian Symposium on Neural Networks – SBRN (13 editions). 
ENIAC is a national forum for researchers, practitioners, educators and students to present and discuss innovations, trends, experiences, developments and work in progress in the fields of Artificial and Computational Intelligence. 
STIL is the bi-annual Language Technology event.
All events are supported by the Brazilian Computer Society – Special Interest Groups: CE-IA, CE-IC and CE-PLN

In this work we evaluate the You Only Look Once (YOLOv3) architecture for real-time detection of insect-pests in soybean. Soybean crop images were collected on different days, locations and weather conditions, between the phenological stages R1 to R6, considered of high occurrence of soybean pests. For training and testing the neural network, we used a 5 fold cross-validation considering four metrics to evaluate the classification results: precision, recall, F-score and accuracy; and three metrics to evaluate the detection results: mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination R². The experimental results showed that the YOLOv3 architecture trained with batch size 32 leads to higher classification and detection rates compared to batch sizes 4 and 16. The results indicate that the evaluated architecture can support experts and farmers in monitoring pest control action levels in soybean fields.

Performance analysis of YOLOv3 for real-time detection of pests in soybeans

The field of neural networks has seen significant advances in recent years with the development of deep and convolutional neural networks. Although many of the current works address real-valued models, recent studies reveal that neural networks with hypercomplex-valued parameters can better capture, generalize, and represent the complexity of multidimensional data. This paper explores the quaternion-valued convolutional neural network application for a pattern recognition task from medicine, namely, the diagnosis of acute lymphoblastic leukemia. Precisely, we compare the performance of real-valued and quaternion-valued convolutional neural networks to classify lymphoblasts from the peripheral blood smear microscopic images. The quaternion-valued convolutional neural network achieved better or similar performance than its corresponding real-valued network but using only 34% of its parameters. This result confirms that quaternion algebra allows capturing and extracting information from a color image with fewer parameters.

Quaternion-Valued Convolutional Neural Network Applied for Acute Lymphoblastic Leukemia Diagnosis

Aspect Sentiment Triplet Extraction (ASTE) is an Aspect-Based Sentiment Analysis subtask (ABSA). It aims to extract aspect-opinion pairs from a sentence and identify the sentiment polarity associated with them. For instance, given the sentence ``Large rooms and great breakfast", ASTE outputs the triplet T = {(rooms, large, positive), (breakfast, great, positive)}. Although several approaches to ASBA have recently been proposed, those for Portuguese have been mostly limited to extracting only aspects without addressing ASTE tasks. This work aims to develop a framework based on Deep Learning to perform the Aspect Sentiment Triplet Extraction task in Portuguese. The framework uses BERT as a context-awareness sentence encoder, multiple parallel non-linear layers to get aspect and opinion representations, and a Graph Attention layer along with a Biaffine scorer to determine the sentiment dependency between each aspect-opinion pair. The comparison results show that our proposed framework significantly outperforms the baselines in Portuguese and is competitive with its counterparts in English.

A Deep Learning approach for Aspect Sentiment Triplet Extraction in Portuguese

Genetic-based clustering meta-heuristics are important bioinspired algorithms. One such technique, termed Genetic Algorithm for Decision Boundary Analysis (GADBA), was proposed to support Structural Health Monitoring (SHM) processes in bridges. GADBA is an unsupervised, non-parametric approach that groups data into natural clusters by means of a specialized objective function. Albeit it allows a competent identification of damage indicators of SHM-related data, it achieves lackluster results on more general clustering scenarios. This study improves the objective function of GADBA based on a Cluster Validity Index (CVI) named Mutual Equidistant-scattering Criterion (MEC) to expand its applicability to any real-world problem.

Improving a genetic clustering approach with a CVI-based objective function.

Data availability in a wide variety of domains has boosted the use of Machine Learning techniques for knowledge discovery and classification. 
The performance of a technique in a given classification task is significantly impacted by specific characteristics of the dataset, which makes the problem of choosing the most adequate approach a challenging one. Meta-Learning approaches, which learn from meta-features calculated from the dataset, have been successfully used to suggest the most suitable classification algorithms for specific datasets. This work proposes the adaptation of clustering measures based on internal indices for supervised problems as additional meta-features in the process of learning a recommendation system for classification tasks. The gains in performance due to Meta-Learning and the additional meta-features are investigated with experiments based on 400 datasets, representing diverse application contexts and domains. Results suggest that (i) meta-learning is a viable solution for recommending a classifier, (ii) the use of clustering features can contribute to the performance of the recommendation system, and (iii) the computational cost of Meta-Learning is substantially smaller than that of running all candidate classifiers in order to select the best.

Evaluating Clustering Meta-Features for Classifier Recommendation

Co-clustering is a specific type of clustering that addresses the problem of simultaneously clustering objects and attributes of a data matrix. Although general clustering techniques find non-overlapping co-clusters, finding possible overlaps between co-clusters can reveal embedded patterns in the data that the disjoint clusters cannot discover. The overlapping co-clustering approaches proposed in the literature focus on finding global overlapped co-clusters and they might overlook interesting local patterns that are not necessarily identified as global co-clusters. Discovering such local co-clusters increases the granularity of the analysis, and therefore more specific patterns can be captured. This is the objective of the present paper, which proposes the new Overlapped Co-Clustering (OCoClus) method for finding overlapped co-clusters on binary data, including both global and local patterns. This is a non-exhaustive method based on the co-occurrence of attributes and objects in the data. Another novelty of this method is that it is driven by an objective cost function that can automatically determine the number of co-clusters. We evaluate the proposed approach on publicly available datasets, both real and synthetic data, and compare the results with a number of baselines. Our approach shows better results than the baseline methods on synthetic data and demonstrates its efficacy in real data.

A Co-occurrence Based Approach for Mining Overlapped Co-clusters in Binary Data

Numerous problems in machine learning require some type of dimensionality reduction. Unsupervised metric learning deals with the definition of intrinsic and adaptive distance functions of a dataset. Locally linear embedding (LLE) consists of a widely used manifold learning algorithm that applies dimensionality reduction to find a more compact and meaningful representation of the observed data through the capturing of the local geometry of the patches. In order to overcome relevant limitations of the LLE approach, we introduce the LLE Kullback–Leibler (LLE-KL) method. The proposed method employs the KL divergence between patches of the $k$-nearest neighbors (KNN) graph instead of the pointwise Euclidean metric. Our objective with such a methodological modification is to increase the robustness of the LLE to the presence of noise or outliers in the data. Our empirical results using several real-world datasets indicate that the proposed method delivers a superior clustering allocation compared to state-of-the-art methods of dimensionality reduction-based metric learning.

A Kullback-Leibler divergence-based locally linear embedding method: a novel parametric approach for cluster analysis

In this presentation, we show our approach that seeks to apply computer vision to assist robots in classifying the target team's organizational structure faced with a group of malicious target agents. Thus, robots can select the most appropriate strategy among the containment strategies implemented for each organizational structure or continue with the method proposed by literature for cases where the targets are not organized.

Intelligent Agents for Observation and Containment of Malicious Targets Organizations

Question answering (QA) systems aim to answer human questions made in natural language. This type of functionality can be very useful in the most diverse application domains, such as the biomedical and clinical. Considering the clinical context, where we have a growing volume of information stored in electronic health records, answering questions about the patient status can improve the decision-making and optimize the patient care. In this work, we carried out the first experiments to develop a QA model for clinical texts in portuguese. To overcome the lack of corpora for the required language and context, we used a transfer learning approach supported by pre-trained attention-based models from the Transformers library. We fine-tuned the BioBERTpt model with a translated version of the SQuAD dataset. The evaluation showed promising results when evaluated in different clinical scenarios, even without the application of a clinical QA corpus to support a training process. The developed model is publicly available to the scientific community.

Experiments on Portuguese Clinical Question Answering

The de-identification of clinical notes is crucial for the reuse of electronic clinical data and is a common Named Entity Recognition (NER) task. Neural language models provide a great improvement in Natural Language Processing (NLP) tasks, such as NER, when they are integrated with neural network methods. This paper evaluates the use of current state-of-the-art deep learning methods (Bi-LSTM-CRF) in the task of identifying patient names in clinical notes, for de-identification purposes. We used two corpora and three language models to evaluate which combination delivers the best performance. In our experiments, the specific corpus for the de-identification of clinical notes and a contextualized embedding with word embeddings achieved the best result: an F-measure of 0.94.

Premium content

Downloads

Next from BRACIS 2021

Performance analysis of YOLOv3 for real-time detection of pests in soybeans

Similar lecture

MHCCL: Masked Hierarchical Cluster-wise Contrastive Learning for Multivariate Time Series

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES