Research

My main research interests are statistical machine learning and machine vision.  My research projects are summarised here - for more details, see my publications.

Machine Vision

• Object Recognition
  
  See the MSR Cambridge object recognition page for further information on object recognition projects I have been involved in - including LiveObject - a realtime recognizer for around 15 object classes. Previous projects include:
  

  • LOCUS: Learning Object Classes with Unsupervised Segmentation
    LOCUS addresses the problem of learning a model of an object class (e.g. horses, cars) from a 'bucket' of images each containing an object in that class, without requiring any human annotation of the images. It discovers the location and pose of the object in each image and also gives a segmentation of each object.
  • Object Categorization by Learned Universal Visual Dictionary
    Using labelled data, we learn a dictionary of textons which is both compact and discriminative, allowing real time object recognition. See videos of the interactive demo.
    
• Generative Affine Localisation and Tracking
  
  Using a layered generative model and VMP, we are able to localise objects undergoing affine motion in a video sequence, whilst simultaneously learning their shape and appearance.

Bioinformatics

• Integrating Functional Genomics data with Genetics to Help Understand Human Disease
  
  Common diseases such as cardiovascular disease, cancer, obesity, diabetes and psychiatric illnesses are caused by a combination of multiple genetic and environmental factors.  Understanding how the genetic factors interact with each other and with the environment would allow better prevention, diagnosis and treatment of these diseases.  We are collaborating with researchers at the Wellcome Trust Sanger Institute to apply genome-level machine learning with both genetic and functional genomic data, to study the effects of multiple genes jointly.
  

Machine Learning

• Infer.NET and Csoft
  
  Infer.NET is a powerful framework for factor graph inference which exploits the capabilities of the .NET framework, such as generics and reflection. Csoft is an extension to the C# language that adds probabilistic modelling and inference capabilities and hence provides a convenient way to invoke Infer.NET from within ordinary C# code.
  
  
• Variational Message Passing (VMP) and VIBES
  
  During my Ph.D., I created the Variational Message Passing algorithm which allows variational inference to be applied automatically to any conjugate-exponential Bayesian network. You can read my publications on VMP or try it out for yourself using the VIBES inference engine.
  
  
• Escaping local minima through hierarchical model selection
  
  Using hierarchical models to avoid local minima in complex generative models.

You can also read about past projects.