Projects

In reverse chronological order:

• Identify Misinformation/Fake news from social media
  • Research in Deep Learning and NLP approach to identify fake news and rumour
  • Collect fake news data in Singapore, crawl and annotate rumour/non-rumour data gleaned from social media (Twitter, Facebook, Forums)
  • Work in progress
• Topic modelling for scientific authors | [code]
  • Researched in topic model using Latent Dirichlet Allocation. Extended LDA for a hierarchical collection of documents. Achieved good results with Author-Topic Model
  • Built a ranking system to assign relevant reviewers to papers and automatically nominate reviewers during review process
  • Advisor: Prof Bryan Low
• Financial Reporting using Machine Learning
  • Applied ML methods on financial transaction data. Solution using Neural Network and Decision Tree.
  • Enhanced interpretability of ML system by displaying Decision Tree result. Certain level of pruning was done to make the tree more readable
  • Replace hand-coded rules with >95% accuracy
• Enhance Customer Churn prediction through Semi-Supervised Learning | [code]
  • Modeled customer and their calls as one big graph, with edge weights as a function of calls and text activities
  • Researched on Label Propagation to propagate labeled nodes (already churned customers) to unlabeled nodes in a graph. Churned customers might influence their friends in the network to churn
  • Used Label Propagation score as a feature to train churn classifier. Winner classifier is Logistic Regression with accuracy of 96.9% accuracy
  • Designed an architecture to train the data in a distributed manner using MapReduce, speeding up training time
  • Advisor: Prof Ng Wee Keong
• Prediction of genetic sequence mutation using Neural Network | [code]
  • Implemented Recurrent Neural Network to train and predict how genetic sequence mutates under influence of HIV-1 drugs. Used one-hot encoding to encode 4 proteins (T, G, A, C) and LSTM to process the sequence. Correctly predicted positions of mutation and corresponding types up to 90%
  • Researched and experimented alternative approach using Game Theory and Cellular Automata
  • Advisor: Dr Samuel Gan, Dr Su Tran To Chinh
• Privacy-preserving in Outsourced Database | [code: ElGamal, Paillier]
  • Implemented ElGamal and Paillier Homomorphic Cryptosystems to encode plaintext SQL queries and retrieve result from encrypted database
  • Built different handlings for different types of queries (AND, OR, and aggregate functions)
  • Advisor: Prof Ng Wee Keong, Dr Vasily Sidorov
• Preserving Data Integrity in Cloud Storage | Technical Report
  • Researched on a tree-based hashing function to detect changes in files. Idea: File is split into blocks, and forms leaves of a B-tree. Root of the tree is hashed in a bottom-up manner, becoming signature of the file. Signature is computed regularly on the third-party storage site and sent to a server for integrity checks
  • Implemented a working solution for the generation of signature and server check. Confirmed the runtime complexity of calculating signature is less than O(n)
  • Work is selected to present at International Conference of Undergraduate Research 2015
  • Advisor: Prof Ng Wee Keong