Hao Liu

(looking for **** summer Internship, May – Aug)

513-***-**** adbvg1@r.postjobfree.com Cincinnati, OH 45220 LinkedIn SUMMARY

With strong background in Privacy preserving techniques, Adversarial Perturbation, NLP, Machine/Deep Learning, AI SKILLS Python (Pytorch, Tensorflow), Matlab, C++, SQL AWS, GCloud Git, HPC EDUCATION

Ph.D. Computer Science University of Cincinnati Jan 2018 – current MS Intelligent Control and Optimization Tongji University Sep 2014 – Jul 2017 BS Electrical Engineering Shanghai Dianji University Sep 2009 – Jul 2013 PROFESSIONIAL EXPERIENCES

University of Cincinnati Cincinnati, OH, USA Research Assistant Jan 2018 – current (Jan 2022)

Preserve Privacy Leakage of Searchable Encryption in an Email System

• Designed a self-injection defense to mitigate the privacy leakage of searchable encryption, resulted in attacker’s attacking ability reducing from 1 to 0.0037.

• Extracted 30,109 emails in textual format from a real-word email dataset, leveraged NLTK to preprocess data into easy- to-interpret formats for downstream NLP assignments. (Tokenization, Tagging, Chunking, Stemming, Lemmatization)

• Trained Markov Model to check semantics of self-injected files, combined with machine learning classifiers (Naïve Bayes & KNN) to distinguish and filter those files, achieved 0 FP and small (1.8%) FN.

File-Injection Attacks Implementation and Mitigation with Natural Language Processing

• Built two customized automatic text generators (N-gram & LSTM (RNN) based) to perform File-Injection attacks.

• Designed adversarial examples in word & character level as customized rules inside and outside text generators to make generated texts qualified for injection attacks.

• Designed and formulated the defense as a semantic analysis problem where sought to differentiate injected texts from benign texts, solved it by ML techniques (AdaBoost, SVM, Random Forest ). (Used NLTK to extract linguistic features, Recursive Feature Elimination to select features, ML algorithms to detect injected files.) The results shown that File-Injection attack is not a big threat in practice.

Voice Command Fingerprinting on Smart Home Speakers

• To infer a voice command of smart home speakers by analyzing its encrypted network traffic.

• Formulated the attack as machine learning problem, collected 1,000 encrypted traffic traces from Amazon smart speakers, built and trained attacking models (CNN) on collected dataset.

• Leveraged Doc2Vec method to analysis the Semantic Distance of voice commands, combine it with Accuracy as metric to access privacy leakage. The attacks can correctly infer 33.8% voice commands. Tongji University Shanghai, China Research Assistant Sep 2014 – Nov 2017

Offline Optimization System of Target Speed Curve of an Automatic Urban Railway

• Designed and built an offline optimization app to produce optimized target speed curve of an automatic urban railway. Obtained the corresponding Patent.

• Built a multi-objective target speed curve model based on the train & line condition of Shanghai Metro Line 11.

• Formulated the problem as an optimization problem and implemented it with Genetic algorithm in Matlab, achieved great performance in terms of speeding, punctuality, parking accuracy, energy consumption.

Real-time Speed Dashboard of Urban Railway

• Designed and built a real-time speed dashboard app for an Urban Railway, the app has been used as a part of the Urban Railway real-time monitoring system in Shanghai.

• Designed the mathematic model of speed dashboard system, implemented it in Microsoft Foundation Classes (in C++). Envision Energy Shanghai, China Data Analyst Intern Oct 2015 – Mar 2016

• Cleaned and analyzed data of wind turbines, visualized results by using EDA approach with matplotlib tool. Provided an efficient way to locate the root cause of wind turbine’s faults.

• Conduct multiple machine learning algorithms (Linear Regression & Random Forest) to predict the faults of wind turbines. Provided a maintaining guidance for wind turbines. SELECTED PUBLICATIONS:

• H Liu, B.Y Wang, “Mitigating File-Injection Attacks with Natural Language Processing” by the 6th ACM IWSPA 2020, New Orleans, LA, USA, 18 March, 2020

• H Liu, B.Y Wang, N Niu, S Wilson, X.T Wei, “Vaccine: Obfuscating Access Pattern Against File-Injection Attacks” by IEEE CNS 2019, Washington, D.C., USA, 10-12 June, 2019

• H Liu, C.Y Qian, Z.M Ren, G.L Wang, “Research on Running Curve Optimization of Automatic Train Operation System Based on Genetic Algorithm,” by EITRT 2017, Changsha, China, 20-22 Oct, 2017, Best Paper Award. HONORS & AWARDS:

• IEEE CNS Conference Student Travel Grant (NSF Sponsored) USA, 2019

• GSGA Research Fellowship University of Cincinnati, USA, 2019

• Best Paper Award in the 3rd EITRT 2017 Conference (3.8% recipients) Changsha, China, 2017

Contact this candidate