Selected Publications

Ranking functions in information retrieval are often used in search engines to extract the relevant answers to the query. This paper makes use of this notion of information retrieval and apply onto the problem domain of cognate detection. The main contributions of this paper are: (1) positional tokenization, which incorporates the sequential notion; (2) graphical error modelling, which calculates the morphological shifts. The current research work only distinguishes whether a pair of words are cognates or not. However, we also study if we could predict a possible cognate from the given input. Our study shows that language modelling based retrieval functions with positional tokenization and error modelling tend to give better results than competing baselines.
NAACL SRW (Under review), 2018.

Since the blockchain that lies behind Bitcoin is a decentralized one, there is no daily limit unlike regular financial instruments such as stocks, bonds, and securities. This indicates that Bitcoin can be highly profitable if a trading algorithm is used. Three approaches are presented to accurately predict the price movement of Bitcoin: deep learning models based on historical bitcoin price data, predictive models based on Natural Language Processing (NLP) for social media, and traditional machine learning models based on Dynamic Time Wrapping (DTW) from historical data of exchange platforms.
Work in progress, 2017.

To analyze large amounts of text data, machine learning algorithms are required. A binary sentiment classi- fication of tweets needs to be given. The final optimal model is a weighted BoW combined with GloVe and trained with logistic regression. For this final model elongated words are mapped on their base forms. Neural networks seem promising, but need strong hardware to optimise.
To be submitted, 2016.


Ethereum ERC 860: Custodian-Client Contract Standard

A standard interface of custodian smart contracts that allows on-chain tracking of client smart contract instances.

Transfer Learning for Flower Classification by ResNet

Five categories of flowers’ pictures are being trained by a CNN, based on ResNet and through transfer learning, to provide a classifier on flower categories.

Artificial Intelligence on Ascending Triangle Pattern of Stock Markets

To make a forecast for a rising trend of stock market, the ascending triangle patterns are recognized by the use of RNN with logistic regression and SVM.

The Neon Network: Heuristic E-coupons Ecosystem on Blockchain

A Ethereum Research Project on E-Coupon smart contract and secure value distribution gateway at HKUST.

Crib Dragging Auto Cracker

It auto-cribdrags the 5000 most common English words on the cipher text to speed up the efficiency of conventional cribdragging technique.

Padding Oracle Cracker

This project exploit the vunlerability of padding oracle to crack the given CBC encrypted message and at last reveal its plain content.

Game Theory Algorithm in P2P WifiDirect Transmission Gateway

An Android research app aiming at providing peer-to-peer connection platform. Offering chat room, file transfer and even class reflection implementations for Game Theory implementations.


The Neon Network scaffolds a sustainable and scalable ticket selling framework upon Ethereum network.

A deep reinforcement learning agent solving simulated Hong Kong traffic model.

A machine learning model for chocolate consumption prediction of Godiva

The overview of IOTA Tangle with its possible attack scenarios and countermeasures.

Recent Posts

Writer at

More Posts

Published at Toward Data Science. You will be able to program and build a vanilla Feedforward Neural Network (FNN) starting today via PyTorch. Here is the python jupyter codebase for the FNN. Read More.


Published at Breathe Publication with over 750 upvotes. This post serves as a very beginner-level overview of the crypto with 9th largest market cap , IOTA. All the graphs and contents are fully credit to IOTA’s white paper. Read More.


Recently, as the flaws of Proof-of-Work(PoW), the consensus algorithm employed by Bitcoin, grow harder to disguise. Various alternatives such as Proof-of-Stake(PoS), Proof-of-Burn(PoB) and many of them emerged, contriving hard to take over their ancient precursor, PoW. Read More.


To save words and time, a table is presented for you to clearly skim through the differences between a relational database (e.g. MySQL) and a non-relational database (e.g. MongoDB). There are 10 features of both types of databases are compared to provide you a high-level understanding about their distinctions. Read More.


Today, you and your friends wanna organize a lunch gathering with the other friends, but you are not sure who will attend, and who will not. You have reserved a group lunch with given price and limited seats in a nice Taiwanese restaurant, and you hope to split the cost with the attendees eventually. How could the smart contract help you? Read More.



Headstart Fellowship @ Silicon Valley

HKUST Engineering Student Ambassador

FireBird CTF Core Member 2017

Semi-pro Magician (Hong Kong & Taiwan)

External Executive (Mainland Student Association)