ML

Revisited the Adaptive Instance Normalization (AdaIN) style transfer, a significant advancement in the domain of style transfer. Introduced multi-level style transfer, applying the AdaIn module to various encoder levels instead of exclusively at the last level. Further enhancement was made by replacing the AdaIn module with an attention module, improving the model’s capacity to stylize images beyond color, also considering other features like texture and shape.

Every student applying for a college education surely wants to know what are the chances of getting into their dream school. Our project is designed to visualize and predict the probability of admission of certain students and the universities to which they apply, which is beneficial for the students in need.

Proposed a visualization method of discrete fracture network with randomly generated fracture trace maps based on machine vision methods and statistical principles using Python language, and thus established a complete discrete fracture network similarity evaluation system. Also proposed a pointing plus gridding algorithm to accurately calculate the position and density similarity of trace maps, innovatively took spacing similarity index into the calculation of comprehensive similarity. Finally, applied the established system to the actual project to test its reliability and applicability

Built an Auto-Regressive Integrated Moving Average model to predict the expansion of hornets; Trained a CNN-based model to distinguish Asian Giant Hornets from other bees; Implemented a multi-parameter optimization technique and finally F1-Score reached 0.992 while accuracy reached 0.987.