Riley F. Edmunds

I am currently a Research Assistant in Dawn Song's Lab (part of BAIR) and a Machine Learning Consultant through my company Alinea.AI. I will complete my undergraduate in Computer Science at UC Berkeley in Spring of 2019. During that time, I built the research team at Machine Learning at Berkeley - a student organization of 80 members that conducts projects with industry partners, research with labs, and hosts numerous events on campus.

I recently wrote a chapter on Adversarial Machine Learning in the recently-released "Artificial Intelligence Safety and Security".

Email  /  GitHub  /  LinkedIn

Artificial Intelligence Safety and Security
Chapter on Adversarial Machine Learning
Phillip Kuznetsov, Riley F. Edmunds, Ted Xiao, Humza Iqbal, Raul Puri, Noah Golmant, Shannon Shih
CRC Press 2018
Book / Chapter

Chapter surveying the field of Adversarial Machine Learning with connections to AI Safety.

Tensegrity Robot Locomotion under Limited Sensory Inputs via Deep Reinforcement Learning
Jianlan Luo, Riley F. Edmunds, Franklin Rice, Alice M. Agonino
International Conference on Robotics and Automation (ICRA), 2018
PDF / Video

Work demonstrating that Tensegrity robots can learn locomotion policies even with severely limited sensory inputs using Mirror Descent Guided Policy Search (MDGPS).

Transferability of Adversarial Attacks in Model-Agnostic Meta-Learning
Riley F. Edmunds, Noah Golmant, Vinay Ramasesh, Phillip Kuznetsov, Piyush Patil, Raul Puri
Deep Learning and Security Workshop (DLSW) in Singapore, 2017
PDF / Slides

Work desmonstrating that in a Meta-Learning context, adversarial attacks transfer between networks trained on different classification tasks drawn from Omniglot and mini-ImageNet.

Hierarchical Semi-Supervised Embeddings for Anomaly Detection
Riley F. Edmunds, Efraim Feinstein
Intuit I.A.T. 2017

A novel semi-supervised autoencoder model for fraud detection in event-streams having unbalanced class sizes and concept drift. Filed patent application based on invention.

Inclined Surface Locomotion Strategies for Spherical Tensegrity Robots
Lee-Huang Chen, Brian Cera, Edward L. Zhu, Riley F. Edmunds, Franklin Rice, Antonia Bronars, Ellande Tang, Saunon R. Malekshahi, Osvaldo Romero, Adrian K. Agogino, Alice M. Agonino
International Conference on Intelligent Robots and Systems (IROS), 2017
PDF / Video

The first tensegrity robot to achieve reliable locomotion on inclined surfaces of up to 24 degrees, with two novel multi-cable actuation policies, suited for steep incline climbing and speed, respectively.

Machine Learning at Berkeley

VP of Research of UC Berkeley's first Machine Learning student organization from April 2017 to May 2018. Project manager August 2016 to Present.

Complex-Valued Neural Networks
Pat Virtue, Riley F. Edmunds, Vinay Ramasesh, Renee Sweeney, Stella Yu

We derive and implement novel complex-valued layer functions for CNNs, evaluating performance on frequency-domain problems including representation learning on MRI and classification of SAR data.

Learning Transferability Metrics Across Tasks
Riley F. Edmunds, Jianlan Luo*

Work towards learning empirical transferability metrics between reinforcement learning tasks using autoencoder reconstruction errors and Model-Agnostic Meta-Learning.

Survey of Neural Architecture Search
Riley F. Edmunds

A blog post on Neural Architecture Search's applications in Deep Learning, with an emphasis on crafting Domain Specific Languages to encode expressive yet tractable search spaces.

Tutorial on Bayesian Optimization
Riley F. Edmunds, Rahil Mathur

A tutorial on Bayesian Optimization and its applications in Deep Learning, with a focus on providing as much intuition behind the underlying theory as possible.

VR Virtual Campanile
Yulin Zheng, James Lin, Riley F. Edmunds, Kyle Provencher

Using the Unity game engine and the HTC Vive, we created a 3D modeled virtual version of the UC Berkeley Campanile along with an interactive carillon.

Capsule - Travel the World through Photos
Riley F. Edmunds, Connor Killion, Aparna Krishnan
iOS App

Capsule connects the world through pictures. Travel to real world locations, pick up a capsule, and view images of everywhere else it's been. When you're done, travel somewhere else, take a picture, and leave it there for future explorers. Run up your score with frequent contributions, or use those points to add new capsules to the world. Happy travels!