Malignant melanoma is a lethal form of skin cancer, where early and precise diagnosis is essential for improving survival. While convolutional neural networks (CNNs) have shown promise in dermoscopic image analysis, they often neglect patient metadata that clinicians routinely consider. Vision-language models (VLMs) provide a unified architecture for multimodal understanding; however, their utility in clinical domains remains limited due to domain mismatch and lack of adaptation. To address these limitations, we propose a retrieval-augmented framework that integrates pre-trained VLMs with a multimodal vector database of dermoscopic images and patient metadata. For each input, the system retrieves semantically similar cases and inserts them into structured prompts, which enable few-shot classification without fine-tuning. Our method outperforms image-based, text-based, and embedding-fusion baselines by yielding higher F1-scores and reducing both false positives and false negatives. Our findings support retrieval-augmented prompting as a reliable and scalable approach to clinical decision support in melanoma diagnosis.

The exponential growth of video content has made it increasingly difficult to locate specific information within large-scale video repositories. To address this challenge, we present a multimodal video indexing framework that integrates audio transcription and visual scene analysis into a unified, searchable structure. The system uses speech-to-text models to transcribe spoken content and applies vision-language models to generate semantic descriptions from video frames. To evaluate the effectiveness of the proposed framework, we conducted experiments on 174 hours of broadcast television content. The results, which include both retrieval accuracy and query latency, confirm the ability of the system to locate relevant sections efficiently and support scalable search in data-intensive environments.

We present Quantum Deepflow, a forecasting and decision support platform that integrates classical and quantum sequence modeling to address volatility and data irregularity in raw material procurement. The system combines an LSTM autoencoder with a Quantum Long Short-Term Memory (QLSTM) model, which enables robust and accurate forecasts from noisy time-series inputs. Users can interact with the platform through a visual interface that links forecast outputs to strategic key performance indicators such as purchase timing, cost estimates, and inventory risk. In a real-world deployment at a Korean steel manufacturer, the system achieved a 32.5% reduction in overstocking and saved $1.8 million in inventory costs. This work demonstrates a practical approach to exposing quantum-enhanced forecasting capabilities through an automated, cloud-based interface that bridges the gap between emerging quantum technology and enterprise-scale decision-making.

As car-sharing services evolve, there is a growing effort to analyze users’ safe driving behaviors and effectively manage shared vehicles. Unlike previous researches that focus on simple situations like sudden acceleration and lane departure using cameras with additional sensors, we introduce a new approach that detects more complex traffic rule violation, especially red-light violation, using only the monocular dashcam videos. The proposed framework employs the attention mechanism of Transformer, and effectively encodes the traffic signal objects and contextual information within the video. It utilizes a novel method, POISE (Positional Object Information by Spatial Encoding), to handle the positional information of traffic signal objects. Our quantitative and qualitative evaluations demonstrate the effectiveness of our proposed framework in detecting red-light violations compared to existing methods.

Video anomaly detection involves automatically identifying unusual or abnormal events in videos, such as crimes and accidents. This paper proposes a novel framework for video anomaly detection named Video Anomaly Transformer for Monitoring Accidents and Nefariousness (VATMAN). Existing approaches often implicitly train the model to generalize the normal data into a single distribution, then detect data that are less generalized as anomalies during the evaluation. However, such methods often struggle with complex data. To address this, our framework leverages the self-attention mechanisms of Transformers, combined with pre-trained 3D Convolutional Neural Networks (3DCNNs). The proposed anomaly score based on the self-attention mechanism can achieve anomaly detection that is less sensitive to data complexity. Experiments on the Abnormal Behavior CCTV Video Dataset demonstrate that VATMAN outperforms existing anomaly detection methods and shows many favorable properties.

In data analysis, unsupervised anomaly detection holds an important position for identifying statistical outliers that signify atypical behavior, erroneous readings, or interesting patterns within data. The Transformer model, known for its ability to capture dependencies within sequences, has revolutionized areas such as text and image data analysis. However, its potential for tabular data, where sequence dependencies are not inherently present, remains underexplored. This paper introduces Transformer for Point Anomaly Detection (TransPAD), a novel Transformer-based AutoEncoder framework specifically designed for point anomaly detection. Our method captures interdependencies across entire datasets, addressing the challenges posed with non-sequential, tabular data. It incorporates unique random and criteria sampling strategies for effective training and anomaly identification, and avoids the common pitfall of trivial generalization that affects many conventional methods. By leveraging an attention weight-based anomaly scoring system, TransPAD offers a more precise approach to detect anomalies. Extensive testing on a range of benchmark tabular datasets shows that TransPAD consistently outperforms existing methods. Our source code is available at https://github.com/nth221/TransPAD.

This paper overviews and discusses our recent work on a multivariate conditional outlier detection framework for clinical applications.

This paper overviews the background, goals, past achievements and future directions of our research that aims to build a multivariate conditional anomaly detection framework for the clinical application.

We develop a novel probabilistic ensemble framework for multi-label classification that is based on the mixtures-of-experts architecture. In this framework, we combine multi-label classification models in the classifier chains family that decompose the class posterior distribution P(Y1, …, Yd|X) using a product of posterior distributions over components of the output space. Our approach captures different input–output and output–output relations that tend to change across data. As a result, we can recover a rich set of dependency relations among inputs and outputs that a single multi-label classification model cannot capture due to its modeling simplifications. We develop and present algorithms for learning the mixtures-of-experts models from data and for performing multi-label predictions on unseen data instances. Experiments on multiple benchmark datasets demonstrate that our approach achieves highly competitive results and outperforms the existing state-of-the-art multi-label classification methods.

We propose a new probabilistic approach for multi-label classification that aims to represent the class posterior distribution P(Y|X). Our approach uses a mixture of tree-structured Bayesian networks, which can leverage the computational advantages of conditional tree-structured models and the abilities of mixtures to compensate for tree-structured restrictions. We develop algorithms for learning the model from data and for performing multi-label predictions using the learned model. Experiments on multiple datasets demonstrate that our approach outperforms several state-of-the-art multi-label classification methods.

This paper studies multi-label classification problem in which data instances are associated with multiple, possibly high-dimensional, label vectors. This problem is especially challenging when labels are dependent and one cannot decompose the problem into a set of independent classification problems. To address the problem and properly represent label dependencies we propose and study a pairwise conditional random Field (CRF) model. We develop a new approach for learning the structure and parameters of the CRF from data. The approach maximizes the pseudo likelihood of observed labels and relies on the fast proximal gradient descend for learning the structure and limited memory BFGS for learning the parameters of the model. Empirical results on several datasets show that our approach outperforms several multi-label classification baselines, including recently published state-of-the-art methods.

The objective of multi-dimensional classification is to learn a function that accurately maps each data instance to a vector of class labels. Multi-dimensional classification appears in a wide range of applications including text categorization, gene functionality classification, semantic image labeling, etc. Usually, in such problems, the class variables are not independent, but rather exhibit conditional dependence relations among them. Hence, the key to the success of multi-dimensional classification is to effectively model such dependencies and use them to facilitate the learning. In this paper, we propose a new probabilistic approach that represents class conditional dependencies in an effective yet computationally efficient way. Our approach uses a special tree-structured Bayesian network model to represent the conditional joint distribution of the class variables given the feature variables. We develop and present efficient algorithms for learning the model from data and for performing exact probabilistic inferences on the model. Extensive experiments on multiple datasets demonstrate that our approach achieves highly competitive results when it is compared to existing state-of-the-art methods.