Sergio Camelo
Computational Math PhD Student @Stanford
I am a 4th Year PhD Student at Stanford ICME, where I focus on making food supply chains more transparent and sustainable. My research, under the advice of Prof. Dan Iancu, leverages heavily on optimization and machine learning.
At Stanford I teach ICME's Data Visualization class (CME151A) and I have spent my summers working with the Data Science team @Airbnb and the Machine Learning teams @Facebook NY.
Before coming to Stanford I did my MSc thesis on approximating NPHard problems through Semidefinite Optimization, under the advice of Prof. Mauricio Velasco. I was a data scientist for two years at Quantil Applied Mathematics.

2015 
Stanford PhD Student in Computational and Mathematical Engineering
Working in the broad intersection of optimization and machine learning

2016 
Stanford Teaching Fellow in Data Viz
Teaching CME151A, a course to design interactive Data Visualizations on the Web using D3

Summer 2018
Airbnb
Designing largescale Data Visualization tools to detect and understand anomalous behavior in data

Summer 2017
Facebook NY
Building Recommendation Systems for the Local Places Team

Summer 2016
Facebook MPK
Implementing prediction algorithms for the Events and Location Teams

20132015
University of the Andes MSc in Mathematics
Thesis in approximating NPHard Graph problems through semidefinite optimization relaxations

20132015
Quantil Data Scientist
Designing machine learning models to predict crime in Colombian cities, detect fraud in the healthcare system, and find ways to make the energy market more competivive

20082013
University of the Andes BSc in Mathematics and Economics
Thesis in finding optimal bandwidths for kernel estimation and kernel classification algorithms
Research

Nearest Neighbors Methods for Support Vector Machines [PDF]
Annals of Operations Research
We propose a statistical nearestneighbors procedure to approximate the solution of the Support Vector Machines problem on big datasets. This approximation comes at a much cheaper computational cost than obtaining the exact solution.

A Structural Model to Evaluate the Transition from SelfCommitment to Centralized Unit Commitment [PDF]
Energy Economics (Minor Revisions)
We analyze the transition to a Centralized Unit Commitment energy auction. The study constructs a competitive benchmark of energy production by estimating marginal costs of energy producers and then compares it with real data.

An Estimation of Costs and Welfare for the New Colombian Healthcare Plan [PDF]
Revista de Economia del Rosario
We use an econometric decision making model to predict the effect of Colombia's major healthcare reform on the government’s budget and the country’s welfare .

Semidefinite Relaxations for Copositive Optimization [PDF]
Master's Thesis
We give new results on the Barvinok, Veomet and Laserre semidefinite approximation of the copositive cone when applied to calculating the independence number of a graph. With this, we propose an algorithm for obtaining large independent sets of graphs and evaluate its empirical performance on Hamming and DeBruijn graphs.

CrossEntropy for Detecting Anomalous Behavior in HealthCare Service [PDF]
Working Paper
We discuss the performance of a computational tool that uses crossentropy calculations to automatically identify anomalous behavior in patients’ and insurers’ spending. The tool controls for gender, age group, medical diagnosis and other socioeconomic factors. The methodology is tested with real data, the 2010 Colombian healthcare services database, where we detect strange behaviors of health providers.
Talks and Posters

Semidefinite Approximations to Copositive Programming [Slides]
ISMP 2015

Nearest Neighbor Methods for Support Vector Machines [Poster]
Foundations of Computational Mathematics 2014

Cross Entropy for Detecting Anomalous Behaviour in HealthCare Service Provision [Poster]
Latin American Congress in Probability and Statistics 2015

Clustering and Centrality for Graph Visualization [Poster]
CS448B @ Stanford