MScProjects 

 Predicting Galatic Photometric Redshift

Redshift is a central concept in cosmology, enabling the speed of distant galaxies to be determined. Redshifts are determined on the basis of galatic colours in three or more filters, giving a coarse approximation to the spectral energy distribution. Together with the UCL Astrophysics group, this project aims to estimate the redshift based on a training database of known redshifts and coarse spectra. In addition to an accurate estimation, a key aim is to produce uncertainty measures in the redshift estimate. The project will attempt to improve the Neural Network techniques currently used by the Astrophysics group, using a form of Bayesian technique based on constrained Gaussian Processes. This is a potentially fun project for which a successful outcome might result in software used by astronomers.

You will learn : advanced Machine Learning techniques, Bayesian techniques, Generative Models, Gaussian Processes.

Suitability : the project would be suitable for a student with

strong mathematical abilities and an interest in advanced machine learning.

 Netflix

Netflix is a competition that aims to predict which movies a user might like, based on which movies they have currently ranked. See netflix. If we can do this well, we'll win a million dollars.

You will learn : advanced Machine Learning techniques, Bayesian techniques, Generative Models, Collaborative Filtering techniques.

Suitability : the project would be suitable for a student with

strong mathematical abilities and an interest in advanced machine learning.

 Science Network

Scientific articles often use jargon and subfield specific language. For example, a paper may talk about `vertices' and `edges', whilst another uses the notation `nodes' and `links'. Such jargon makes finding similar papers and ideas a difficulty in science. The aim of the project is to automatically form a network of papers, linked by their similarity, *regardless of the specific jargon employed*. The project will be based on a corpus of scientific documents with citations used to help automatically understand what elements the two texts are similar.

You will learn : advanced Machine Learning techniques, Bayesian techniques, Generative Models, Latent Semantic Analysis, Latent Variable Models. Elementary Graph Theory concepts.

Suitability : the project would be suitable for a student with

strong mathematical abilities and an interest in advanced machine learning.

 Grammar Checker

Ever read the BBC website and get annoyed how often grammar mistakes occur? Wonder what happened to your license fee? This projects aims to make a simple and fast grammar checker that we can run on any website, and tell the BBC quickly where it has (probably) made a mistake. The method will be based on machine learning, understanding the role of words in a sentence and checking that they are grammatically correct.

You will learn : advanced Machine Learning techniques, Bayesian techniques, Generative Models, Latent Variable Models.

Suitability : the project would be suitable for a student with strong

mathematical abilities and an interest in advanced machine learning.

 Symmetrical Translation

Surprisingly, most state-of-the-art automatic machine translation systems are non-symmetric. If we have a sentence in language A, and translate this to language B, then re-translate back to sentence A, we don't get the same sentence (in particular the same meaning) back. This project aims to make an automatic translation system based on a central understanding of what a sentence actually means so that translational symmetry is preserved.

You will learn : advanced Machine Learning techniques, Bayesian techniques, Generative Models, Latent Variable Models, Natural Language Processing.

Suitability : the project would be suitable for a student with strong

mathematical abilities and an interest in advanced machine learning.