- To make machines understand human commonsense better, by generalizing beyond human-curated resources.
- Trained models on commonsense facts and evaluate by predicting held out facts, currently achieving 82% accuracy (human performance is 89%) and outperforming multiple baselines. Also evaluate our models as a method for the COPA task (choice of plausible alternatives), a commonsense causal reasoning task.
- Experienced several neural network architectures, including deep neural network, LSTM and so on
- Implemented neural network models using Theano and Lasagna
- Details Here: http://ttic.uchicago.edu/~kgimpel/commonsense.html
Linear equation solvers Implementation on GPU
- I was one of the eight undergraduates selected from China as part of the UA intern program.
- I implemented three linear equation solvers on GPU and perform efficiency tests to determine the best one should be used in energy systems. I found that Preconditioned Conjugate Gradient(PCG) algorithm performs the best on general energy system datasets. However, LU(Lower Upper Decomposition) performs better when the matrix is block diagonal.
Stock Return Prediction based on Patents and Twitter
- Designed stock prediction model using kernel SVM, logistic regression and decision trees with features extracted from twitter and patents data.
- Performed LDA topic model and dynamic topic model on patents data to extract patent features and preformed sentiment analysis on twitter data to extract twitter features.
- Based on our model’s trading decision on Google, Microsoft and Yahoo stock from January to June 2015, investors will benefit 1.05% to 1.25%.
Serendipitous Recommendation via User’s Recent Research Interests
- Constructed researchers’ basic profiles from Google Scholar etc more than 5 data resources.
- Integrated the basic profile with other features regarding the most dissimilar researcher from a group of similar researchers. Recommendation result is obtained by calculating cosine similarity on TF-IDF matrix representations.
I love trying new things and chanllange myself everyday. So I went to many places and try to make beautiful memories for all of them. I'm hoping to make more memories in the near future about my academic life!