Advances in Machine Learning
Lead Research Organisation:
University College London
Department Name: Computer Science
Abstract
Machine learning and artificial intelligence have already begun to have a transformational economic impact through automation of jobs and improved decision making. There remain fundamental challenges to enhance current machine learning techniques, particularly in scaling them up to even larger problems as we tackle bigger challenges in AI. My project will aim to develop novel methods of representing and reasoning with learned information. Potential impacts would include better and more meaningful search, more natural computer interfaces, tools for scientific research and many others.
Develop new ways to represent, process and combine symbolic information with non-symbolic processing methods.
The Research fits well with the EPSRC's research area of Artificial Intelligence Technologies.
Develop new ways to represent, process and combine symbolic information with non-symbolic processing methods.
The Research fits well with the EPSRC's research area of Artificial Intelligence Technologies.
Organisations
People |
ORCID iD |
| David Barber (Primary Supervisor) | |
| Raza Habib (Student) |
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/N509577/1 | 01/10/2016 | 24/03/2022 | |||
| 1782562 | Studentship | EP/N509577/1 | 01/10/2016 | 12/01/2021 | Raza Habib |
| Description | We have a developed a new algorithm for performing more efficient Markov Chain Monte Carlo. We combine recent advances in deep artificial neural networks with traditional probabilistic machine learning and our algorithm will likely have benefit in both AI and scientific applications. Machine learning deals with the challenge of drawing conclusions and inferences from data. A large variety of algorithms in this domain come under the umbrella of Bayesian machine learning, where the goal is to assign probabilities to possible future predictions given a model and some data. Unfortunately, for most interesting and complex models exact computation is intractable and approximations are needed. The quality of the approximate answer depends on the efficiency of the algorithm used and we have introduced a novel framework for construction these algorithms more efficiently called Auxiliary Variational Markov Chain Monte Carlo. In addition to the novel MCMC algorithm we have also built a system combining probabilistic learning with speech synthesis allowing us to create a state of the art speech synthesiser that is able to control the emotional expression of synthesised speech. |
| Exploitation Route | The MCMC algorithm is already being used by physicists at the university of Warwick and has opened a new research avenue for producing better MCMC algorithms. The speech synthesis algorithm is being used by google and has been cited in follow up papers. |
| Sectors | Aerospace, Defence and Marine,Creative Economy,Digital/Communication/Information Technologies (including Software),Education,Government, Democracy and Justice,Culture, Heritage, Museums and Collections |
| URL | https://github.com/AVMCMC/AuxiliaryVariationalMCMChttps://tts-demos.github.io/ |
| Title | Auxiliary Variational Sampler |
| Description | We have open sourced software to draw samples from posterior distributions that arise in bayesian statistical analysis. The software should aid researchers in statistical methods to develop new algorithms and statisticians to perform data analysis. |
| Type Of Technology | Software |
| Year Produced | 2019 |
| Open Source License? | Yes |
| Impact | Other researches have used the software in their analysis and are building on new developments. |
| URL | https://github.com/AVMCMC/AuxiliaryVariationalMCMC |