Louis Dorard
My work is in Machine Learning, in particular in online learning, bandit problems and searching large spaces. I am also interested in applications of ML techniques to the web (personalisation of content), and to music (composition, performance, music information retrieval).
In my PhD, I am studying bandit algorithms as a way to focus the exploration of large spaces and to make search as quick as possible. I am especially interested in the use of probabilistic models (Gaussian Processes) for the search of tree-structured spaces, such as sequences of possible actions to take in an environment (for planning in Markov Decision Processes).
I am applying these techniques to:
- Content-Based Image Retrieval with Relevance Feedback
- Sequence Labelling and performing music automatically, by having the computer "label" music notes with performance parameters (such as loudness and duration) in a way that renders the music expressively.
Please read my research summary for more information. I am also the main organiser of the 2011 Exploration & Exploitation Challenge and associated workshop at ICML (http://explo.cs.ucl.ac.uk/).
Supervisors: John Shawe-Taylor
In my PhD, I am studying bandit algorithms as a way to focus the exploration of large spaces and to make search as quick as possible. I am especially interested in the use of probabilistic models (Gaussian Processes) for the search of tree-structured spaces, such as sequences of possible actions to take in an environment (for planning in Markov Decision Processes).
I am applying these techniques to:
- Content-Based Image Retrieval with Relevance Feedback
- Sequence Labelling and performing music automatically, by having the computer "label" music notes with performance parameters (such as loudness and duration) in a way that renders the music expressively.
Please read my research summary for more information. I am also the main organiser of the 2011 Exploration & Exploitation Challenge and associated workshop at ICML (http://explo.cs.ucl.ac.uk/).
Supervisors: John Shawe-Taylor
less
Uploads
Papers by Louis Dorard
Algorithms designed to handle the exploration-exploitation dilemma were initially motivated by problems with rather small numbers of actions. But the ideas they were based on have been extended to cases where the number of actions to choose from is much larger than the maximum possible number of plays. Several problems fall into this setting, such as information retrieval with relevance feedback, where the system must learn what a user is looking for while serving relevant documents often enough, but also global optimisation, where the search for an optimum is done by selecting where to acquire potentially expensive samples of a target function. All have in common the search of large spaces.
In this thesis, we focus on an algorithm based on the Gaussian Processes probabilistic model, often used in Bayesian optimisation, and the Upper Confidence Bound action-selection heuristic that is popular in bandit algorithms. In addition to demonstrating the advantages of the GP-UCB algorithm on an image retrieval problem, we show how it can be adapted in order to search tree-structured spaces. We provide an efficient implementation, theoretical guarantees on the algorithm’s performance, and empirical evidence that it handles large branching factors better than previous bandit-based algorithms, on synthetic trees.
Algorithms designed to handle the exploration-exploitation dilemma were initially motivated by problems with rather small numbers of actions. But the ideas they were based on have been extended to cases where the number of actions to choose from is much larger than the maximum possible number of plays. Several problems fall into this setting, such as information retrieval with relevance feedback, where the system must learn what a user is looking for while serving relevant documents often enough, but also global optimisation, where the search for an optimum is done by selecting where to acquire potentially expensive samples of a target function. All have in common the search of large spaces.
In this thesis, we focus on an algorithm based on the Gaussian Processes probabilistic model, often used in Bayesian optimisation, and the Upper Confidence Bound action-selection heuristic that is popular in bandit algorithms. In addition to demonstrating the advantages of the GP-UCB algorithm on an image retrieval problem, we show how it can be adapted in order to search tree-structured spaces. We provide an efficient implementation, theoretical guarantees on the algorithm’s performance, and empirical evidence that it handles large branching factors better than previous bandit-based algorithms, on synthetic trees.