Julia First Order Optimization
Open-sourced two Julia packages for separable optimization problems

Nicholas Moehle, Ellis Brown, Mykel Kochenderfer

Created a new Julia organization, JuliaFirstOrder, dedicated to first-order optimization methods. PiecewiseQuadratics.jl allows for the representation and manipulation of piecewise-quadratic functions, including the computation of the proximal operator and the convex envelope. SeparableOptimization.jl solves the problem of minimizing a sum of piecewise-quadratic functions subject to affine equality constraints by applying the Alternating Direction Method of Multipliers (ADMM). Presented at JuliaCon 2021.

SeparableOptimization.jl PiecewiseQuadratics.jl Talk Blog Jul 2021

Single Shot MultiBox Object Detector (SSD), in PyTorch

Implented SSD, a unified framework for real-time object detection using a single network. Has become the de facto implementation in PyTorch (4.7k+ stars).

Original Paper Code Mar 2017

Keras implementation of a Bayesian Neural Network with dropout

Experiments investigating the effect of weight prior selection and network architecture on uncertainty estimates.

Poster Code May 2019

Gender Inference from Character Sequences in Multinational First Names

Used Naïve-Bayes and a Char-RNN implemented in PyTorch to extrapolate gender from character sequences in first names from around the world.

Blog Code Dec 2017

Music Genre Classification from 3 second clips, in Torch

A deep learning method for automatically labeling songs by genre using Torch. We convert audio files to spectrograms and process them as images.

Code Mar 2017