Machine Learning Engineer Intern Resume

TECHNICAL SKILLS

Machine Learning: Deep/Reinforcement Learning, CNN, RNN, GAN, TensorFlow, Keras, NumPy, Scikit - learn, CUDA

Programming Language: Java, Python, Perl, C, C++, JavaScript

Frameworks/Tools: MongoDB, Node.js, Hadoop, Kafka, ElasticSearch, Cassandra, REST, Redis, MySQL, Spring MVC

PROFESSIONAL EXPERIENCE

Machine Learning Engineer Intern

Confidential

Responsibilities:

Built probabilistic graphical model in completely unsupervised setting (utilized Gibbs sampling to learn parameters wif video training dataset), to remind users of forgotten actions, achieving greater than 54 % accuracy Python
As Big Data team member, developed and deployed web crawlers to crawl resources regularly. Cleaned, visualized, and analyzed all operational data to support high-level decision-making Java, MongoDB, ElasticSearch, Hadoop
Designed and built recommender system to recommend videos to Confidential TV clients based on choices and preferences

Software Engineer

Confidential

Responsibilities:

Implemented features, including Facet, Tagging, Result Grouping, and OID/AD autantication; identified and fixed more than 100 bugs. Earned “Outstanding” rating (highest performance evaluation for Oracle employees) Java, Oracle, JSP, Perl, REST
Created SearchEngineService and implemented Admin API for Oracle Cloud Search to provide RESTful search service
Eliminated redundancy by preparing tools to automate routine activities, reducing unnecessary efforts by at least 80%

Software Engineer Intern

Confidential

Responsibilities:

Developed OSGi bundles to statistically analyze running data, tested High Availability for IBM Systems Director Java, OSGi
Trained neural networks model to detect performance anomalies of IBM servers, achieving 91 % accuracy C, Perl
Devised a novel approach “Gradient Ascent wif Noise”, computing teh perturbation by iteratively adding gradients wif uniform noise, adding teh perturbation to input natural image to generate adversarial example
Achieved 92% accuracy to attack inception-v 3 wif 0 . 6 % perturbation rate Python, TensorFlow
Extended ES technique (Tim Salimans et al., 2017 ) wif meta-learning, building much more dynamic, responsive, and tractable alternative to naive grid search. Trained agent wif teh approach to finish OpenAI Atari and Box2D tasks (could solve LunarLander-v2 after 9894 episodes) Python, Keras, TensorFlow
Ensured universality of ES wif meta-learning approach to any learning algorithms
Combined ES wif neural networks using species-based approach: maintain a set of parallel “species” or models wif varying network structure, new species would be introduced via random “mutation” after eliminating teh lowest performers