Custom NER (Name Entity Recognition) can be useful to identify security related commit messages, comments in Github to identify software security related activities in a repository

Active learning is particularly useful when we do not have big number of ground truth dataset. I used active learning with random forest classifier to train a model that detects heart disease form 350 instances of patients.

Meaningful Sentence Generation using LSTM

(Spring, 2018)

Using word to vector technique and LSTM, machine can generate meaningful text.

Defense Against Adversarial Examples

(Summer, 2018)

Neural network can make mistakes easily to infer information from degraded image. Sometimes it can be fatal for critical applications like automated car. We used bagging+ adding noise and adversarial retraining as a defense

Multi User Chat Application to monitor data flow in a server/client environment in TCP/IP

(Spring, 2018 )

Acceptance problem of Turing Machine

(Fall, 2017)

Course project, Automata

Trojan Detection

(Fall, 2017)

We proposed a method that can check the result of multiplication to detect Trojan for this function in a FPGA