Nghiem Nguyen Viet Dung

Constrained Clustering Via Post-Processing

This work proposed the novel approach by converting the output of many clustering algorithm to a cluster fractional allocation matrix (CFAM). Then, takes it as an optimization problem with requirement to satisfy (combined) different types of constraints. In addition, we also shows practical application namely the attribute-level constraint and the (combined) fairness constraint. Summary:

• Applicable to existing (constrained) clustering solutions
• Using ILP/Gurobi - powerful optimization model/tool
• Guarantee constrained satisfaction

Data Visualization for Human Perception - Concept mining approach

Data often views as rows and columns/records and attributes/a set of data points. However, at highest level of information, data is complex maps of concepts and relations (similar to how our mind store information).

Properties:

• Introductory including open improvements
• Make complex data easier to understand
• An important step before creating final reports, inferencing knowledge

Question-Answering using auto labeler and active learning

QA is an important task in NLP. However, the cost of labeling is high as the annotator need to read the whole context and find the appropriate answer. In this repo, we proposes to the following scheme:

1. Init a pretrain QA model
2. Inference the dataset using the current model
3. Export into excel files
4. Active model select files/questions need to check/answer by human
5. Human check machine-generated answers and rewrite if necessary
6. Update model with new labeled data
7. Test model's performance. Return to step 2 if not satisfy.

Optimization Process

An educational place for beginner to practice problem-solving.

• Analyze the problem
• State the problem by mathematical tools
• Define the optimization model
• Solving with optimization solver