Mời Freelancer vào Dự Án
- N/ACông việc đã Hoàn thành
- N/AĐúng ngân sách
- N/AĐúng hạn
- N/ATỉ lệ thuê lại
Các nhận xét gần đây
Recommender System PlatformSep 2016
My company business grow, we need a system that must to be scalable, flexible, easy to integrate new model, support AB testing which is an well-known business metrics, and lastly, realtime adapt to user when they watch new movies or musics. I proposal an architecture which combine Spofity’s recommender system and Netflix’s recommender system, and lead 4 teammates to build the system.
Music recommender systemSep 2016
I proposal and build a music recommender system using Generative Adversarial Networks which is a novel deep learning model. The input to system is rely on only raw audio so songs can be suggested without any meta-data. The input is converted to MFCC features and pass to my model to learn representation, then i use cosine similarity for ranking songs. My system result is impressive than other based on meta-data
Lead Recommender SystemSep 2016
I build a movie recommender system to serve 600k users using PredictioIO, MLLib Spark, Scala. In detail, I implement Alternative Least Square for Collaborative Filtering approach and Single Value Decomposition for Content-Base approach, after that i combine these models using a linear model. My system improves performance to 500% rps and Click-Through-Rate increase 40%
Side ProjectsJan 2016
- Research and build chat bot using seq2seq which is LSTM encoder that encode sequence to fixed vector and try to decode into answer. I crawl tweet and comments as question and answer to train my model. - Design architecture and build deep learning model ( for localization, detection, classification) that can run on small cheap device that have less than 512MByte memory and slow processor.
Bachelor2012 - 2016 (4 years)
The Best Thesis (2016)Ho Chi Minh City University of Science
We implement a novel model to generate image description. Our approach is association of two most successful deep learning models in Computer Vision and Natural Language Modeling that known as Convolution Network and Long Short-Term Memory. Our result can help blind man to see the world
Message Prediction Competition (2017)Unpossib.ly
8th/220, Predict a message given the first letters of each word
CIKM AnalytiCup 2017: Lazada Product Title Quality Challenge (2017)Lazada
20th/450, build a product title quality model that can automatically grade the clarity and the conciseness of a product title.