Data Mining Lab
How can we learn models for prediction and better understanding of data? How can we find useful patterns and anomalies in big data? How to handle data that are either too huge or too fast? In Data Mining Lab, we perform interesting research projects on key models, algorithms, and systems for artificial intelligence (AI), knowledge discovery, and machine learning.
Deep Learning and Machine Learning
How can we learn from massive amount of data? We work on designing
models, algorithms, and systems for deep learning and machine learning.
We focus on the following researches.
- Autonomous machine learning (AutoML): we develop methods to design an AI that learns how to learn automatically.
- Lightweight machine learning: we aim to build a fast and energy-efficient model.
- Transferable machine learning: we aim to build a high quality model even when labeled instances are scarce.
- Anomaly detection and prediction: we develop models and algorithms for finding anomalies and frauds.
- Object detection and classification using sensors: we develop methods to detect objects and humans using data from various sensors.
Graphs and Tensors
How to analyze graphs and/or multi-dimensional data? We develop models, algorithms, and systems for graphs and tensors.
- Random walk with restart for ranking and relation inference.
- Graph machine learning: we develop state-of-the-art node classification and representation learning methods, which outperform recent works including Graph Convolutional Network (GCN) and Graph Attention Network (GAT).
- Scalable graph mining: we develop large graph processing engines and related algorithms to handle very large graphs whose sizes are more than hundreds of billions of edges.
- Scalable tensor mining: we develop methods for analyzing large scale multi-dimensinoal data (tensors) including network intrusion logs, knowledge bases, and time-varying social networks.
Given who-watched-which TV transaction data, how can we recommend relevant TV programs for a given user? Given a friendship social
network, how can we recommend friends that are likely to make connections to a given user? Recommendation is an important application
of data mining, and is widely used in movie recommendation, restaurant recommendation, job recommendation, article recommendation, and friend recommendation. In this project, we work on designing and developing models, algorithms, and systems for recommendation. We focus on the following researches.
- Recommendation in multi-modality, where multi-modal data, including ratings, social networks, texts, images, and videos are available.
- Sequence recommendation where we want to predict the next item in a sequence (e.g., video and news recommendation).
- Active recommendation: we devise method to "control" the dynamics of recommender systems, instead of merely observing them.
- Network based recommendation: we work on recommendation in networks or graphs (e.g., "People You May Know" in LinkedIn, or friend recommendation in Facebook) which is a very important problem. We work on fast and scalable models and algorithms for network based recommendation.
How can we design an AI that automatically trades stocks? How can we detect financial frauds? Financial AI aims to develop models,
algorithms, and systems for financial applications (e.g., actuarial and insurance, consumer banking, and investment banking), transforming
subjective decision-making to data-driven decision-making. Despite the “efficient market hypothesis” of classic economics, we have seen many success stories of investments, and we develop state-of-the-art AI methods to exploit the tiny gap between conventional theory and what actually happens. We focus on the following researches.
- Time series prediction: we develop methods for predicting future values of time series.
- Asset value prediction and algorithm trading: we design AI that learns high-quality trading strategy.
- Consumer analytics: we design methods for effective personalization, and customer understanding.
- Fraud detection and prediction: we develop methods for detecting and predicting suspicious financial transactions and activities.