Academia Sinica Intern, July 2018 ~ Dec 2018

After completing the courses. I applied for the internship program of the Academia Sinica. During the period of Academia Sinica, my job is to improve the performance of the algorithm.

Data Innovation Application Competition, April 2018

Our work is to make a model that uses deep learning to predict weather profiles. The main information is tides, waves, and wind direction. Mainly for people who love surfing.

Graph Sparsification with Generative Adversarial Networks, Jan 2019 ~ Dec 2019

Graph sparsification aims to reduce the number of edges of a network while maintaining its accuracy for given tasks. In this study, we propose a novel method that is able to sparsify networks for community detection tasks. Our method is able to capture those relationships that are not shown in the original graph but are relatively important, and creating artificial edges to reflect these relationships and further increase the effectiveness of the community detection task.

We use GAN as a learning model and let the generator learn to generate random walks that are able to capture the structure of a network. During the training phase, in addition to judging the authenticity of the random walk, discriminator also considers the relationship between nodes at the same time. We design a reward function to guide the generator creating random walks that contain useful hidden relation information. These random walks are then combined to form a new social network that is efficient and effective for community detection. Experiments demonstrate that our sparsification method is effective in real-world networks and able to be applied in different community discovery clustering algorithms.

C++|Hyperedge Aware Greedy, Mar 2018 ~ Dec 2018

This is the seed diffusion algorithm I implemented during my internship at the SINICA. It can be applied to graphs with hyperedge. In addition, I also make three pruning techniques to speed up the execution time, and use multithread to parallelize the algorithm.

Python|Continuous Bag-of-Words Modeling, Jan 2018

The order of the text affects the overall semantics, Using Deep Learning, let the first and last words as input, and the middle as output. If each word is made as a one-hot vector, data-sparse problems may occur. In order to avoid this, the word embedding technique is used to reach the compression of the text vector representation.

Python|Taxi Travel Time Prediction, Nov 2017

It's a 2015 ECML PKDD competition, I use Deep Learning approaches to predict the taxi driving time. Although it's expired when I joined this competition on Kaggle. I still submit my result and the score is 0.60 (mean-squared-error). The first place score was 0.53.

Python|Probabilistic Latent Semantic Analysis , Oct 2017

From the perspective of probability, use the Expectation-Maximization algorithm to calculate the probability of generating the current Topic for each document to match the degree of the query. After that, I have more knowledge of information retrieval.

Python|Vector Space Model , Sep 2017

The classic data retrieval approaches converts the documents and query into vectors, then calculates the cosine similarity between matching query and the document. This helps me know how the search engine work although it's just a basic approach and not advanced.

Python|MNIST Handwriting Recognition, Aug 2017

MNIST handwriting recognition is a very basic machine learning example. As a beginner for machine learning, I also try to start with this.

Python|Water Level Prediction, July 2017

In order to know well with python and machine learning, I got some practices on the internet. This is a water level prediction with the Scikit tool.