Haoxuan Li

Master of Science, Computer Science (expected in May 2025)

Electrical Engineering and Computer Science (Joint Program with HIT)

Bachelor of Engineering, Computer Science and Technology

Student Researcher

Mentor: Prof. Glenn Healey

• Hyperspectral Image Data Processing and Analysis

  Analyzed hyperspectral data cubes through MATLAB, extracting spectra in SWIR and NVIR for building and vegetation regions to identify dominant bands

  Digitized imaging filters into CSVs and compared them in MATLAB against specific chemical emissions

  Digitized Mollweide projection plots of different pitch types for baseball players

Student Researcher

Mentor: Prof. Hongzhi Wang,     Dr. Xiaoou Ding

• Research on Hybrid Error Data Cleaning Technology for Human In The Loop    

  Independently proposes and implements a hybrid human-in-the-loop framework for erroneous data cleaning

  Automatic Detection and Repair: The framework employs algorithms to automatically identify missing points and outliers in data, attempting auto-correction. When the algorithm's confidence in auto-correction does not reach a preset threshold, the system escalates the issue to human experts for intervention

  Human-Machine Interactive Repair: For data issues uncertain or beyond automatic correction, human experts are brought into the repair process. By recording their opinions and judgments, the system continuously adjusts the dynamic rational value range of the data, thereby enhancing the accuracy of anomaly detection and repair in subsequent processing

  Dynamic Data Quality Improvement: The efficacy of this cleaning framework has been validated through experiments assessing classification accuracy under different noise data ratios and regression accuracy following changes in the rational value environment

• Research on Data Pricing Techniques Based on Data Quality Evaluation

  Contributed to constructing the relationship between data quality and pricing. Considered the weight of different indicators in pricing decisions, and the integration with current economic theories, to determine the appropriate pricing of data based on quantified evaluation metrics

• Research on Time-Series Data Error Detection and Cleaning with Multi-Constraint

  Participated in the experiment of a new time-series data error detection and cleaning algorithm based on multiple constraints. Through comparative experiments with existing data cleaning methods, the superior performance of the newly proposed algorithm in addressing inaccuracies in time-series data was validated

  Replicated a mainstream algorithm "A symbolic representation of time series, with implications for streaming algorithms". The algorithm was broken down into viable steps, and by using or developing appropriate functions, theoretical models were converted into executable programs