Breast cancer remains one of the most prevalent malignancies affecting women worldwide, with over 2.3 million new cases diagnosed annually. While conventional diagnostic methods like mammography have reduced mortality rates, artificial intelligence offers opportunities to further enhance detection accuracy. In this paper, we present a comprehensive framework for breast cancer identification using deep learning techniques applied to mammographic images. Our approach leverages transfer learning from ImageNet pretrained models and incorporates an auxiliary classification mechanism to improve diagnostic performance. We evaluated multiple state-of-the-art convolutional neural network architectures, including MobileNetV3, ResNet50, ConvNeXtV2-nano, ResNeXt50, MobileViT-small, and EfficientNet-B3, trained on four publicly available datasets (BMCD, CDD-CESM, CMMD, and MiniDDSM). Performance was assessed on separate external test datasets (VinDr and RSNA) to simulate real-world clinical deployment scenarios. Our experimental results demonstrate that EfficientNet-B3 with auxiliary classification achieved superior performance across most metrics, with an accuracy of 86.14%, F1-score of 85.80%, and AUC of 90.39% on validation data. We further introduce the Probabilistic F1-score as a clinically relevant evaluation metric that accounts for prediction confidence rather than binary decisions alone. The proposed framework, available as an open-source implementation, offers a promising approach for enhancing breast cancer detection while providing insights into the challenges of deploying AI systems in diverse clinical environments. We released our codebase at: https://github.com/thanhhnvnqb/mmbreast

Multiple intent detection and slot filling are two fundamental and crucial tasks in spoken language understanding. Motivated by the fact that the two tasks are closely related, joint models that can detect intents and extract slots simultaneously are preferred to individual models that perform each task independently. The accuracy of a joint model depends heavily on the ability of the model to transfer information between the two tasks so that the result of one task can correct the result of the other. In addition, since a joint model has multiple outputs, how to train the model effectively is also challenging. In this paper, we present a method for multiple intent detection and slot filling by addressing these challenges. First, we propose a bidirectional joint model that explicitly employs intent information to recognize slots and slot features to detect intents. Second, we introduce a novel method for training the proposed joint model using supervised contrastive learning and self-distillation. Experimental results on two benchmark datasets MixATIS and MixSNIPS show that our method outperforms state-of-the-art models in both tasks. The results also demonstrate the contributions of both bidirectional design and the training method to the accuracy improvement. Our source code is available at https://github.com/anhtunguyen98/BiSLU.

Intent detection and slot filling are two fundamental and important tasks in spoken language understanding (SLU). Motivated by the fact that the intent and slots in a user utterance have a strong relationship, joint models that deal with both tasks in a single framework have become a predominant choice in SLU research. Most existing joint models build two different decoders on top of a shared weight encoder or exploit intent information to detect slots. Some joint models transfer information between two tasks implicitly. In this paper, we propose a bidirectional joint model for SLU that explicitly incorporates intent information into slot filling and slot information into intent detection. Specifically, we first predict a soft intent signal, which is fed into a biaffine classifier to recognize slots. Slot features are then employed along with the utterance representation to predict the final intent. We also introduce a loss function that takes into account three types of losses: soft intent detection, final intent detection, and slot filling. Experimental results on three benchmark datasets ATIS, Snips, and PhoATIS show that our model outperforms previous state-of-the-art models in both tasks with relative error reductions ranging from 6% to 22%.

Text representation plays a vital role in retrieval-based question answering, especially in the legal domain where documents are usually long and complicated.The better the question and the legal documents are represented, the more accurate they are matched.In this paper, we focus on the task of answering legal questions at the article level.Given a legal question, the goal is to retrieve all the correct and valid legal articles, that can be used as the basic to answer the question.We present a retrieval-based model for the task by learning neural attentive text representation.Our text representation method first leverages convolutional neural networks to extract important information in a question and legal articles.Attention mechanisms are then used to represent the question and articles and select appropriate information to align them in a matching process.Experimental results on an annotated corpus consisting of 5,922 Vietnamese legal questions show that our model outperforms state-of-the-art retrieval-based methods for question answering by large margins in terms of both recall and NDCG.

In many real-life recommendation settings, user profiles and past activities are not available. The recommender system should make predictions based on session data, e.g. session clicks and descriptions of clicked items. Conventional recommendation approaches, which rely on past user-item interaction data, cannot deliver accurate results in these situations. In this paper, we describe a method that combines session clicks and content features such as item descriptions and item categories to generate recommendations. To model these data, which are usually of different types and nature, we use 3-dimensional convolutional neural networks with character-level encoding of all input data. While 3D architectures provide a natural way to capture spatio-temporal patterns, character-level networks allow modeling different data types using their raw textual representation, thus reducing feature engineering effort. We applied the proposed method to predict add-to-cart events in e-commerce websites, which is more difficult then predicting next clicks. On two real datasets, our method outperformed several baselines and a state-of-the-art method based on recurrent neural networks.

This paper presents a general, simple, yet effective method for weakly supervised sentiment classification in resource-poor lan- guages. Given as input weak training signals in forms of textual reviews and associated ratings, which are available in many e-commerce websites, our method computes class distributions for sentences using the statistical information of n-grams in the reviews. These distributions can then be used directly to build sentiment classifiers in unsupervised settings, or they can be used as extra features to boost the classification accuracy in semi-supervised settings. We empirically verified the effectiveness of the proposed method on two datasets in Japanese and Vietnamese languages. The results are promising, showing that the method is able to make relatively accurate predictions even when no labeled data are given. In the semi-supervised settings, the method achieved from 1.8% to 4.7% relative improvement over the pure supervised baseline method, depending on the amount of labeled data.

A common strategy for predicting gene function from heterogeneous data sources is to construct a combined functional association network and use this network to infer gene function. In such approaches, the prediction accuracy largely depends on the quality of the network, and network optimization steps can lead to more accurate results. Existing methods, however, construct combined networks, which are then fixed, and no further optimization steps are performed. We propose a method that improves functional association networks before using them to predict gene function. The method uses an online learning algorithm to learn a similarity measure between pairs of genes, then uses this measure to construct new networks. The learning algorithm can handle noisy training signals and is fast enough to be practical. We evaluated the proposed method in predicting gene functions in two species (yeast and human). We found that our method produced networks with improved prediction accuracy, and outperformed two other state-of-the-art gene function prediction methods. A Matlab implementation of the method is available upon request.

Defects are unavoidable in software development and fixing them is costly and resource-intensive. To build defect prediction models, researchers have investigated a number of factors related to the defect-proneness of source code, such as code complexity, change complexity, or socio-technical factors. In this paper, we propose a new approach that emphasizes on technical concerns/functionality of a system. In our approach, a software system is viewed as a collection of software artifacts that describe different technical concerns/-aspects. Those concerns are assumed to have different levels of defect-proneness, thus, cause different levels of defectproneness to the relevant software artifacts. We use topic modeling to measure the concerns in source code, and use them as the input for machine learning-based defect prediction models. Preliminary result on Eclipse JDT shows that the topic-based metrics have high correlation to the number of bugs (defect-proneness), and our topic-based defect prediction has better predictive performance than existing state-of-the-art approaches.

MOTIVATION: The transcription of a gene is largely determined by short sequence motifs that serve as binding sites for transcription factors. Recent findings suggest direct relationships between the motifs and gene expression levels. In this work, we present a method for identifying regulatory motifs. Our method makes use of tree-based techniques for recovering the relationships between motifs and gene expression levels. RESULTS: We treat regulatory motifs and gene expression levels as predictor variables and responses, respectively, and use a regression tree model to identify the structural relationships between them. The regression tree methodology is extended to handle responses from multiple experiments by modifying the split function. The significance of regulatory elements is determined by analyzing tree structures and using a variable importance measure. When applied to two data sets of the yeast Saccharomyces cerevisiae, the method successfully identifies most of the regulatory motifs that are known to control gene transcription under the given experimental conditions, and suggests several new putative motifs. Analysis of the tree structures also reconfirms several pairs of motifs that are known to regulate gene transcription in combination. AVAILABILITY: http://if.kaist.ac.kr/~phuong/RegTree

Computational discovery of transcriptional regulatory regions in DNA sequences provides an efficient way to broaden our understanding of how cellular processes are controlled. We formulate the regulatory element discovery problem in the regression framework with regulatory regions treated as predictor variables and gene expression levels as responses. We use regression tree models to identify structural relationships between predictors and responses. The regression tree methodology is extended to handle multiple responses from different experiments by modifying the split function. We apply this method to two data sets of the yeast Saccharomyces cerevisiae. The method successfully identifies most of regulatory motifs that are known to control gene transcription under the given experimental conditions. Our method also suggests several putative motifs that present novel regulatory motifs.