Yika Luo
Yika Luo
Be interesting, and be interested.
 

Tech Portfolio

Here is a selected list of my tech projects, which involve a wide range of topics such as autonomous driving, distributed systems, machine learning infrastructure, deep learning, robotics and data science. They are mostly group projects that I have led or participated in at university research labs and tech companies.

 
 
 Jan 2018 - Present

Jan 2018 - Present

SOS- A Real-time Dataflow System for Robotics Applications 

 

 

 
 May - Aug 2017

May - Aug 2017

Hyperplane - A Distributed Deep Learning Training Platform

  • Affiliated Company: Pinterest

  • Advisors: Dmitry Kislyuk, Michael Xu

  • Tech Frameworks: Mesosphere, Chronos, Docker, AWS S3, AWS EFS, MongoDB, Caffe, Python, JavaScript, CSS, HTML

  • Description

    Hyperplane is a distributed model training platform. It has a web interface for users to specify how they want to train the models - what model to train, what data to use, which hyperparameters to set, how many workers you want, etc. After one click, Hyperplane is able to automatically launch jobs on several machines; more specifically, it processes queries, schedules training jobs, allocates GPU resources, and executes model training scripts in containers. (Code/Docs for this project is non-disclosable)

 
 Jan - May 2017

Jan - May 2017

IDK Cascade - A Low-Latency Deep Learning Framework 

  • Affiliated Company: Real-Time Intelligent Secure Execution Lab (RISELab)

  • Advisor: Prof. Joseph Gonzalez

  • Collaborators: Xin WangDaniel Crankshaw, Alexey Tumanov, Fisher Yu

  • Tech Frameworks: Tensorflow

  • Description

    IDK Cascade is an ensemble Convolutional Neural Network (CNN) structure. The motivation is that in most datasets there's distinction between easy and hard examples, so we want to use fast CNN to deal with easy examples and accurate CNN to deal with the hard ones so as to speed up the prediction process. We stack multiple CNNs together in an order of prediction latency, and train the whole structure end-to-end to make the model learn to tell apart easy and hard examples. IDK Cascade achieves 1.6x speedups in image classification tasks while maintaining the same accuracy; it also achieves nearly perfect accuracy (> 95%) in self-driving car tasks. Click below for our published paper.

 
 
 
 Oct - Dec 2016

Oct - Dec 2016

Clipper - A Low-latency Prediction Serving System For Machine Learning

  • Affiliated Research Group: Real-Time Intelligent Secure Execution Lab (RISELab)

  • Advisors: Prof. Ion Stoica, Prof. Joseph E. Gonzalez

  • Main Collaborators: Dan CrankshawCorey Zumar

  • Tech Frameworks: C++, Redis, REST, RPC, Docker

  • Description

    Clipper is a low-latency and intelligent model prediction serving system. For my part, I built a model selection layer that supports selecting the best performing model for different applications using multi-arm bandit algorithms, such as Epsilon Greedy, Thompson Sampling and Upper Confidence Bound (UCB). I also helped investigating a real-time profiling system to automatically find optimal model composition to achieve highest accuracy while satisfying given latency constraints. Click below for Clipper's official website.

 
 Feb 2018 - Present

Feb 2018 - Present

RARL - An Adversarial Reinforcement Learning Framework

  • Affiliated Course: Special Topics in Deep Learning (Graduate Course in Berkeley)

  • Collaborators: Zisu Dong, Xin Wang, Xingyou Song

  • Tech Frameworks: Tensorflow, OpenAI, MuJoCo

  • Description

    This is a reimplementation and extended analysis for an existing paper called ”Robust Adversarial Reinforcement Learning” (RARL). It uses two agents (protagonist and adversary) to make general Reinforcement Learning (RL) algorithms robust on different testing scenarios. For extended analysis, we tested that RARL also works for general Deep Learning (DL) algorithms. We tested on both benchmark image dataset (high dimension) and real-world datasets. Click below for our report.

 
 Feb - May 2016

Feb - May 2016

Rent Estimator - A Data Science Project on Berkeley Housing

  • Affiliated Organization: The Daily Californian

  • Collaborators: Aakash Japi, Shashank Bhargava

  • Tech Frameworks: Python, Pandas, Bokeh, Google Maps

  • Description

    This is a data science project that investigates housing situation around UC Berkeley campus. We constructed a full data pipeline including data scrapping (from Yelp and Craigslist), cleaning, transformation, analysis and visualization. We analyzed features that can potentially influence the rent price, such as the house size, the distance to campus, the restaurant quality and the crime rate. We built a price estimator that estimates house price to determine if the house is overpriced. Click below for our journal.

University Coursework Highlights

Computer Systems

  • Applications of Parallel Computing (CS 267)

  • Operating Systems (CS 162)

  • Databases (CS 186): Have to take it with Joe Hellestein!

Machine Learning

  • Deep Reinforcement Learning (CS 294-112)

  • Safety and Control of Artificial General Intelligence (CS 294-149)

  • Special Topics in Deep Learning (CS 294-131): graduate seminar, they have different speaker every week, take it if you want to be up-to-date on deep learning

  • Machine Learning (CS 189): Have to take it with Jonathan Shewchuk! I have never seen someone who can explain super abstract concepts in such a clear and intuitive way. And the best part is he is able to visualize anything.

  • Artificial Intelligence (CS 188)

Non-Tech

  • Fundamentals of Business (MBA 209F)

  • Technology Entrepreneurship (IEOR 191): Naeem Zafar is an amazing professor. He is an entrepreneur himself. It's definitely a course that teaches you a lot of soft skills. Most important thing I learned is how to SELL.

  • Thinking Through Art and Design: Californian Counterculture (L&S 25): Michael Cohen is such a fun speaker. I found the course very inspiring and eye-opening. It talks about all the non-mainstream culture.