I majored in Probability theory, and focused on Markov Chain related problems. After graduation, I worked at ASUS on Zenfone camera algorithm development for 2 years. Then focus on self-learning platform "Coursera". I completed the series of class "Advanced Machine Learning". Now, I am a data scientist in KKday. The main projects are listed below :
Machine Learning Engineer / Data Analysis Engineer / Data Scientist
Taipei,TW
[email protected]
Camera developer for Zenfone
Master&Bachelor : National Chiao Tung University , Applied Mathematics
The target is comparing the value of customers who buy products on different platform.
I analyze from three aspects :
This is a time model. For each time, it has only 5 features (shop_id/ item_id/ Category_name/ shop_name/ Category_id), So I need generate new features. I add time delay data and embedding the shop name and item category as new features. I also train model to get new features. In the end, I use Linear model to reduce features and build 3 models to get the final answer.
Given a picture, It will generate a short description for this picture.
This model use a pre-trained InceptionV3 model for CNN encoder and extract its last hidden layer as an embedding.
Use about 82K training data and 40K validation date and each picture has 5 captions.
If you ask a programing problem, chat robot will return a closest StackOverflow link for you.
In this model, we will classify the training data to language type and use facebookresearch/StarSpace for embedding questions. For each input, we will classify it and embedding it to vector then found related link.
Use Monte Coral tree search. That is, we choose road by root scores, if meet the tree leaf, do propagation (add score to root).
We can build a tree by playing games, then tree will tell us how to playing game.
Use actor-critic training.
Input 4-frame image, train a DNN to catch the image information and predict the probability for 12 reactions. Then we according the DNN result to make a reaction for this game.
In my master thesis, it provide a simulated method, which can avoid lots of computations, to make the Markov chain approximate its stationary distribution and also give a theorem to prove. At first part, we gave a theorem to prove the convergence of new random variable. For second part, we gave two special cases of simulation and found the random variable will not converge if the chain does not satisfy the condition of theorem. In the end, we provided a way to improve the chain.