About
About Me
Hey I’m an AI researcher based in London. I have expertise mainly in RL based agents and methods, along with experiance in TTS and lenguage models based on large transformers due to my work in Amazon AGI division.
Education
Undergraduate (Computer Science) + Masters (Machine Learning)
I completed my Undergraduate and Masters at the University of Essex. This included a focus on NLP (Natural Language Processing) and general AI methods such as Neural Networks, in both supervised and RL setups. Another area of intrest at this time was mobile application development and general programming.
I gained extensive experiance with development tools such as git and docker at this time, along with an interest in Linux and distributed computing frameworks such as Kubernetes.
PhD (Deep Reinforcement Learning)
During my PhD I focused on Deep Reinforcement learning with a focus on making AI agents able to better generalise there policies to changing environments. This was included changes to the environment setup during inference as compared to training, as well as leanrning policies that were better suited to multi-agent environments where the other agents policies are unkown during training time.
Research Experience
Signal AI
Alongside my Masters year in computer science, I worked part time on visualisations of Natural Language Processing (NLP) data.
University of Essex
Worked on a project with Essex County Council, using traffic based data to provide real time information to commuters.
Microsoft Research Cambridge
Researched deep multi-agent reinforcement learning. Evaluating the use of Win or Learn Fast (WoLF) within deep RL. Introduced WoLF-PPO within a resulting publication as lead authour.
Queen Mary University of London
Researched deep learning approaches for population prediction. Focused on forward model learning for modeling population sentiment based on simulation.
Amazon AGI
Conducting research in TTS (Text to Speech). This includes using deep learning methods for modeling more natural and expressive speech. Recently working on speech-to-speech models.
Areas of Interest
Publications
Clyde: A Deep Reinforcement Learning DOOM Playing Agent - Dino S. Ratlicffe, Sam Devlin, Udo Kruschwitz And Luca Citi - AAAI 2017
Win Or Learn Fast Proximal Policy Optimisation - Dino S. Ratcliffe, Katja Hofmann And Sam Devlin - IEEE Conference On Games 2019
Cross-Lingual Style Transfer With Conditional Prior VAE And Style Loss - Dino S. Ratcliffe, You Wang, Alex Mansbridge, Penny Karanasou, Alexis Moinet And Marius Cotescu - Interspeech 2022