Jochem Gietema
Machine Learning Science Lead @Onfido
Signup · Get unlimited contacts
WORK HISTORY
Machine Learning Science Lead @Onfido
London, GB
Leading a team of machine learning scientists working on document extraction with autoregressive multimodal models and efficient on-device ML. Contribute hands-on to research, model training, and training/inference optimisation (50% IC work).Team achievements- Large-scale autoregressive multimodal fine-tuning, leading to 20x sample efficiency for LoRA adapters- Multilingual vision-language model training from scratch, enabling us to explore new market opportunities- Latency and model size: decreased the latency of our vision-language model for document extraction by 4x, reducing the cost of GPU model serving- Model training, evaluation, and quantisation of our document classification model, resulting in a 50x decrease in model size so that it can run on-device.
EDUCATION
Vrije Universiteit Amsterdam (VU Amsterdam)
Bachelor of Arts (B.A.), Philosophy
Vrije Universiteit Amsterdam (VU Amsterdam)
Master of Laws (LLM), Criminal Law
CS Vincent van Gogh
Gymnasium
Vrije Universiteit Amsterdam (VU Amsterdam)
Master of Laws (LLM), Internet, Intellectual Property and IT
Vrije Universiteit Amsterdam (VU Amsterdam)
Bachelor of Laws (LL.B.)
SKILLS
ABOUT JOCHEM GIETEMA
Leading a team of machine learning scientists working on document extraction with autoregressive multimodal models and efficient on-device ML. Contribute hands-on to research, model training, and training/inference optimisation (50% IC work).I work for Onfido, a leading identity verification company used by companies such as Revolut, JP Morgan Chase, and DocuSign. The company was acquired last year by Entrust for 650M, making it one of the largest exits in the UK in recent years.Previously- Developed an open-source data visualisation tool Clusterfun, which is used at multiple leading computer vision companies in Europe.(https://github.com/gietema/clusterfun)- Co-authored a book about Bayesian Deep Learning - Enhancing Deep Learning with Bayesian Inference (Packt, 2023)(https://tinyurl.com/bdl-book)- Worked on document fraud detection at scale at Onfido (uncertainty estimation, active learning, few-shot learning).
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.