Research Overview

The SML carries out research in the following key areas of machine learning:

  • Document Analysis and Understanding:

    The information revolution has led to an the explosive growth of unstructured documents available for analysis. Analysis and a deep understanding of these documents is crucial for automating standard tasks (tagging, classifying, and extracting information from documents) as well as for performing more intelligent search and organizing search results.

  • Computer Vision:

    Interpreting images and tracking objects in video is a core area of machine learning requiring highly specialized techniques and algorithms. The ability to automatically perform such vision tasks has a wide range of applications from autonomous mobile systems to automated scene analysis for traffic monitoring, security, or medical applications.

  • Reinforcement Learning and Planning:

    Programming a machine to achieve complex tasks requires a considerable investment of resources. Reinforcement learning (RL) aims to create machines that autonomously learn to perform complex tasks by interacting directly with the environment. Applications include robotics, operations planning, resource management, logistics, and operation research.

  • Kernel Methods and Statistics:

    Kernels are systems that describe similarities between objects, for example, the similarity of two faces or the similarity between the pen strokes used to write a letter or character. In combination with support vector machines and convex optimization techiques, kernel methods permit efficient and powerful learning techniques that can be applied to text classification, optical character recognition, bioinformatics, and natural language processing.

  • Sequential Prediction and Information Theory:

    The task of predicting future observations from previous observations is at the core of many tasks in machine learning and more broadly, artificial intelligence. Information theory provides a powerful tool for making such predictions by averaging predictors according to a measure of their (algorithmic) information content. Applications include online classification, pattern recognition, and time-series predictions (e.g., the weather or the stock market).

  • Topic Modelling:

    Latent topic models for text, web and image data are an unsupervised probabilistic model of the words or links in a document or blobs in an image. They are intended to extract semantic topics based on word/link co-occurrences.