Climate change can have a serious impact on human life and occurs due to the emission of greenhouse gases, such as carbon dioxide, into the atmosphere. Deforestation is one contributing factor to climate change. It is important to understand the drivers of deforestation for mitigation efforts, but there is a lack of data-driven studies that can predict these drivers. In this work, we propose a contrastive learning architecture, called Multimodal SuperCon, to classify the drivers of deforestation in Indonesia using composite images obtained from Landsat 8. Multimodal SuperCon is an architecture that combines contrastive learning and multimodal fusion to handle the available Indonesian deforestation dataset. As a means to yield better performance, Multimodal SuperCon upgrades ordinary SuperCon so that it is able to handle auxiliary spatial variables as additional inputs. Our proposed model outperforms previous work on driver classification, giving an 8% improvement in accuracy compared to a state-of-the-art rotation equivariant model for the same task.

White Matter Hyperintensities (WMH) are areas of the brain that have a higher intensity than other normal brain regions on Magnetic Resonance Imaging (MRI) scans. WMH is often associated with small vessel disease in the brain, making early detection of WMH important. However, there are two common issues in detecting WMH: high ambiguity and difficulty detecting small WMH. In this study, we propose a method called Probabilistic TransUNet to address the precision of small object segmentation and the high ambiguity of medical images. We conducted a k-fold cross-validation and cross-dataset robustness experiment to measure model performance. Based on the experiments, Transformer-based models (TransUNet and Probabilistic TransUNet) were found to provide more precise and better segmentation results, as demonstrated by the higher DSC scores obtained compared to CNN-based models (UNet and Probabilistic UNet) and their ability to segment small WMH objects. The addition of a probabilistic model and the use of a transformer-based approach was able to achieve better performance. Our proposed method achieved a DSC score of 0.742 in K-fold cross-validation outperforming the previous method. The use of CNN and Transformer-based probabilistic models increased the average DSC score by +0.016 and +0.001, respectively. In cross-dataset robustness, our proposed method achieved the second-best result with a DSC score of 0.605

One of the major challenges in the change detection (CD) of high-resolution remote sensing images is the high requirement for computational resources. Besides, to get the best change detection result, it must spot only the important changes while omitting unimportant ones, which requires learning complex interactions between multi-scale objects on the images. Despite Convolution Neural Network (CNN) efficiently extracting features from such images, it has a limited receptive field resulting in sub-optimal representation. On the other hand, Vision Transformer (ViT) can capture long-range dependencies. Still, it suffers from quadratic complexity concerning the number of image patches, especially for high-resolution images. Furthermore, both approach can not model the interactions among multi-scale image patches, which is essential for a model to fully understand the natural images. We propose FocalCD, a CD method based on a recently proposed focal modulation architecture capable of learning short and long dependencies to solve this problem. It is attention-free and does not suffer from quadratic complexity. Also, it supports learning multi-scale interaction by adaptively selecting the discriminator regions from multi-scale levels. Besides the efficient yet powerful encoder, FocalCD has an effective multi-scale feature fusion and pyramidal decoder network. FocalCD achieves strong empirical results on various CD datasets, including CDD, LEVIR-CD, and WHU-CD. It reaches F1 scores of 0.9851, 0.952, and 0.9616 on datasets CDD, LEVIR-CD, and WHU-CD outperforming state-of-the-art CD methods while having comparable or even lower computation complexity.

Automatic car make, and model classification is essential to support activities of intelligent traffic systems in urban areas, such as surveillance, traffic information collection, statistics, etc. In order to classify this data, we need an embedded system approach for real-time car recognition. Many approaches could be made, from image processing to machine learning. Recently, the development of the Convolutional Neural Network has spurred various research in the Area. ResNet, Inception, DenseNet, and NasNet are some of the most commonly used Neural Network based method that is used to classify images. In this research, we utilize pre-processing and cropping technique to maximize the quality of dataset. Several deep learning networks are going to be compared in classifying vehicle make and model in the Stanford dataset. The dataset contains 196 different labels. Several evaluation metrics are used to compare the performance of the methods. From the experiment, the InceptionV3 method achieved the best performance of the AUROC ratio for training the dataset under 50 epochs. Other methods that achieve a high AUROC value tends to have a higher computational time. Real-time simulations have shown that the embedded system is capable of classifying a 100 % success rate for six concurrent users

Virtual assistants have gained popularity across various domains, including the legal field, where they serve to offer guidance and aid in the form of law retrieval. In this research, our aim is to develop a legal virtual assistant that combines knowledge graphs (KGs) and information retrieval (IR) techniques. This hybrid approach allows us to provide accurate answers extracted from structured interconnected data while simultaneously cater to a diverse range of legal inquiries. We categorize these inquiries into a few distinct use cases: definition lookup, law component lookup, sanctions, and domain knowledge. Our system encompasses a chatbot platform, knowledge graph querying, and information retrieval. Specifically, we construct a VA system over a legal knowledge graph pertaining to the Indonesian Act concerning Manpower or Labor (UU Ketenagakerjaan) and the Indonesian Act concerning the Creation of Jobs (UU Cipta Kerja). This marks the creation of the first legal virtual assistant in the Indonesian context that combines KG and IR methodologies. To evaluate the effectiveness of our prototype system, we conduct tests using a variety of labor law-related questions, ranging in difficulty. The integration of knowledge graphs and information retrieval proves to significantly improve the support provided for a wide range of potential applications in the legal field.

Breast cancer ranks as the second leading cause of death among women, but early screening and self-awareness can help prevent it. Hormone therapy drugs that target estrogen levels offer potential treatments. However, conventional drug discovery entails extensive, costly processes. This study presents a framework for analyzing the quantitative structure–activity relationship (QSAR) of estrogen receptor alpha inhibitors. Our approach utilizes supervised learning, integrating self-attention Transformer and molecular graph information, to predict estrogen receptor alpha inhibitors. We established five classification models for predicting these inhibitors in breast cancer. Among these models, our proposed MATH model achieved remarkable precision, recall, F1 score, and specificity, with values of 0.952, 0.972, 0.960, and 0.922, respectively, alongside an ROC AUC of 0.977. MATH exhibited robust performance, suggesting its potential to assist pharmaceutical and health researchers in identifying candidate compounds for estrogen alpha inhibitors and guiding drug discovery pathways.

-