Revenue Predict for Low-Click Keywords in Online Advertising Platform
Research Project in Marin Software, 9/2014 - 5/2015
We used some machine learning algorithm with Spark to predict the revenue-per-click value for low-click
keywords in online advertising platforms. We optimized the algorithms from Spark MLlib for better performance
on our task in distributed system.
The image on the left is from here.
Advertisement Keywords Clustering By Keyword Text
Research Project in Marin Software, 9/2014 - 5/2015
Our algorithm clusters keywords in online advertising platforms by analysing on the keyword text without external data set. We designed some natural language processing algorithm, based on machine learning, using Spark framework for distributed system.
Ray Tracer
Course project of CSC418 (Computer Graphics), 3/2014 - 4/2014
This ray tracer rendering engine was built from scratch, and it supports multiple features like anti aliasing, texture mapping, glossy reflection, and soft shallows. We have pre-defined multiple shapes in the code, including circle, triangle, quadangle, cylinder and sphere (part of the shape codes are modified from open source). What's more, we have carefully considered texture mapping on non-quadangle shapes and 3D objects to avoid distortion.
Properties of More Sums Than Differences Sets
Course project of APM461 (Additive Combinatorics), 3/2014 - 4/2014
In this project, we looked into a special kind of subset of integers, where its sumset (Minkowski sum) with itself has larger size than its difference set (Minkowski difference). The main property we have shown is there are infinitely many such sets and the density of such sets is not 0.
Facial Expression Classification
Course project of CSC411 (Machine Learning & Data Mining), 11/2013 - 12/2013
This project is an image multi-class classification problem. We designed machine learning algorithm to learn seven different facial expressions from nearly 3000 labeled images. In the algorithm, we proposed an innovative method which embedded SVM into neural networks to improve accuracy for multi-class classification problem.
Photometric Stereo with Primal-Dual Coded Camera
Summer research project, supervised by Kyros Kutulakos, 5/2013 - 8/2013
This project is aimed to create 3D model of object with Primal-Dual Coded Camera, which is able to record not only the intensity of coming light, but also the type of path where the light travels from light source to the camera. This project is not completed yet.
Minimum Clue Problem in Sudoku
Course project of MAT157 (Analysis I), 11/2012 - 3/2013
We looked into the mathematical properties of Sudoku, especially the minimum possible clues of low-rank Sudoku. During the project, we provided a theoretical proof for 2×2 Sudoku, and proposed some key steps for solving 3×3 minimum clue problem. Note that the problem for 3×3 Sudoku has been solved by exhaustive search, but there is no theoretical proof so far.