Michael Joannou

Machine Learning Research Scientist @Yoti

London, GB

I lead the Face Age Estimation project and provide technical knowledge and direction on Face Anti-Spoofing and Face-Matching projects:• Developed a SOTA age-estimation model that reduced error by 13%(Computer Vision).• My most recent age-estimation algorithm placed 1st internationally on the National Institute for Standards & Technology evaluation for ‘Child Online Safety’. It placed 2nd in the world for 18-30 year olds on ‘mugshot’ images.(Computer Vision).• Reduced model latency by utilising light-weight DNN architectures and producing C++ models with cppflow.• Developed Out-of-Distribution (OOD) detection methods to provide uncertainty measures about model predictions.• Finetune LLMs/VLMs (LLaVA, Llama) and Vision Transformers (CLIP) for vision tasks.• Train Transformer vision encoders and align with text embeddings from language model encoders (LLM)• Training Contrastive Language Image Pretraining (CLIP) models for vision tasks.• Training Vision Language Models (Vision Instruction Tuning; LLaVA).• Developing complex input pipelines for efficient training of computer vision models using multiple GPUs (train speed increased x5).• Creating custom machine learning model architectures to improve performance.• Management and use of Database (SQL; PostgreSQL).• Presenting results and directions to the Head of R&D and CTO weekly.• Articulating complex findings and deliverables to a broad audience within the organization.

University of Plymouth

Master's degree, Robotics

University of Birmingham

PhD, Artificial Intelligence and Cognitive Science

Cardiff University / Prifysgol Caerdydd

Bachelor of Engineering (BEng) with a Year in Industry, Mechanical Engineering