Sherman Ip

Oxford University

Department of Statistics

Oxford Warwick Statistics Programme (OxWaSP) 2015-16 Warwick Cohort

email: sherman.ip@@spc.ox.ac.uk

UCL Personal Webpage

Last updated: 15th November 2016

I have moved to Warwick University.

Sherman is from Maidenhead and studied at University College London. He completed his MSci Physics degree in 2014. During his physics degree, he became interested in statistics and as a result his physics project was on the computational and statistical analysis of 3D imaging of a human chromosome at the London Center for Nanotechnology. Following on, he then completed a MSc Computational Statistics and Machine Learning degree in 2015. His project was on the multivariate regression view of multi-label classification. In his spare time, Sherman enjoys shooting arrows at the local archery range and competes in tournaments.

This website contains reports and presentations produced during the first year of OxWaSP at Oxford University.

Sherman Ip

Computational Statistics and Statistical Computing

Bayesian Logistic Regression with Polya-Gamma Latent Variables

Team:

Documents:

Me and Kaspar Martens have done a project on Bayesian logistic regression using Polya-Gamma latent variables, based on the paper written by Polson, Scott and Windle (2013). This was my first time building a Gibbs sampler and learning about rejection sampling which has been useful. My main contribution to the project was conducting experiments on classifying real data using the Bayesian logistic regression.

[top]

Probability and Approximation

A Review and Demonstration of a Randomised Algorithm for Low Rank Approximation and Their Bounds

Team:

Documents:

Me and Kasper Martens paired up again to do a project on randomised algorithms for low rank approximation and their bounds, based on the paper written by Witten and Candes (2015). My task in this project was to design and conduct experiments to do low rank approximation on simulated matrices.

[top]

Stochastic Simulation

Hamiltonian Monte Carlo

Team:

Documents:

I was paired up with Jack Jewson to do a project on Hamiltonian Monte Carlo (HMC). We have mainly looked at the chapter written by Neal (2011) from the Handbook of Markov Chain Monte Carlo (MCMC) to obtain knowledge of HMC. We have implemented and experimented HMC by sampling from the multivariate and mixture of Normal distributions. I have gained a lot of practical experience on MCMC and understood the Metropolis-Hastings sampler. My main contribution to the project was using the Gelman-Rubin diagnostic to compare the burn in with HMC and random walk Metropolis-Hastings.

[top]

Scalable Methods and Analysis of Large Complex Data

messageR

Team:

Documents:

I've worked with Nathan Cunningham and Paul Vanetti to implement a R package. The package implements the median selection subset aggregation for parallel inference, proposed by Wang, Peng and Dubson (2014). My task was to implement, in C, regularization paths for generalized linear models via coordinate descent, proposed by Friedman, Hastie and Tibshirani (2010). This fits a GLM regularized by a combination of ridge regression and lasso. I have attempted to vectorize the algorithm using AVX but this was not effective because I did not considered the cache line efficiency in matrix multiplication.

[top]

Machine Learning

Kernel Learning via Random Fourier Representations

Team:

Documents:

I've worked with Leon Law, Marcin Minder, Xenia Miscouridou and Andi Wang on Kernel Learning via Random Fourier Representations. My main focus was on fitting a one layer neural network with Fourier features, described in Rahimi and Recht (2007), using different optimization methods.

[top]

Bayesian Inference

Learning our ABCs

Team:

Documents:

I've worked with Nathan Cunningham and Ella Kaye on Approximate Bayesian Computation (ABC), well explained by Marin, Pudlo, Robert and Ryder (2012). My main aim was to provide a conjugate example of using ABC to sample the posterior of the beta distribution and the Normal-Gamma distribution.

[top]

Applied Statistics

Colon Cancer Subtypes from Gene Expression Data

Team:

Documents:

I've worked with Nathan Cunningham, Giuseppe Di Benedetto and Leon Law. Our main aim was test the robustness of the results by De Sousa E Melo and et. al. (2013). I was responsible for using bootstrapping to obtain error bars, in the 10-fold cross validation and number of genes selected, when using the R package pamr (prediction analysis for microarrays) for classification. pamr was developed by Tibshirani, Hastie, Narasimhan and Chu (2002).

[top]

Time Series and Stochastic Processes

Time Series Analysis of Temperature and Activity of Four Healthy Patients

Team:

Documents:

I've worked with Nathan Cunningham and Beniamino Hadj-Amar to conduct time series analysis on data of skin temperature and activity from 4 healthy subjects. We fitted sinusoidal functions and the SARMA models onto the time series. My task was to use AIC and BIC to select which SARMA model to fit onto the time series and assess the performance of the forecasts.

[top]

Mini-project 1

Characterisation of CT Noise in Projection and Image Space with Applications to 3D Printing

Supervisors:

Documents:

X-ray computed tomography can be used to do quality control on 3D printed samples. However there are sources of error in the 3D printing, how the photons behave and in the X-ray detector. This project aims to find a relationship between the sample mean and sample variance grey values in images obtained from the X-ray detector by fitting linear regressions. In addition, latent variable models such as principle component analysis, factor analysis and the compound Poisson were attempted to be fitted to find sources of variance.

[top]

Mini-project 2

Multicore adaptive MCMC

Supervisors:

Documents:

[top]

Archery

I am shooting for Oxford University Company of Archers. My best performance was coming 3rd in the British University College Sports (BUCS) Indoor Archery Championships 2016.

I have ran the Univeristy of London Archery Club as club president between the years 2011-14, for which I have obtained the Full Purple award, for outstanding sporting achievement at a national level, and Honorary Life Membership, for outstanding and exceptional service to the club.

I started archery with Maidenhead Archers.

National Rankings

Gents. Recurve UK

YearRankWA 1440WA 720Total
201610111471123-5835705553978
201512211111109-5685555483891
201480114911451133599560-4586
201395114611151083577573-4494