I'm Thomas van Dongen. I am currently working as head of AI engineering at Springer Nature. I am one of the founding members of The Minish Lab where we develop open-source machine learning packages.
My research interests include:
- 🚤 Small, fast models: Making CPU-friendly models that can run on any device.
- 🧩 Word Embeddings: Focusing on static embeddings to balance performance and resource usage.
- ⚡ Efficient Nearest Neighbors: Optimizing ANN/KNN methods for scalable semantic search.
- 🔍 Recommenders: Developing smarter systems to improve recommendations and information retrieval, focussed on the scientific publishing space.
I'm currently working on:
- model2vec: a library for creating state-of-the-art static embedding models by distilling sentence transformers.
- vicinity: a library for fast and lightweight nearest neighbor search, with flexible indexing backends.
- semhash: a library for lightweight text deduplication, outlier detection, and representative filtering.
- tokenlearn: a library for pre-training static embedding models.
- model2vec-rs: a Rust port of Model2Vec.
Info: