Nils Reimers

Career

Reimers was a doctoral researcher in the UKP Lab at TU Darmstadt from 2014 to 2018, completing a PhD in deep learning and natural language processing under Iryna Gurevych, and stayed on as a postdoctoral researcher until 2021, leading research projects and lecturing on deep learning for NLP. In 2021 Reimers joined Hugging Face as a research scientist and built its neural search research team.³

Reimers joined Cohere in 2022 as Director of Machine Learning and Principal Scientist, leading the team that trains large language models for semantic search and text classification. Reimers's homepage now gives the title VP of AI Search, with a team working on dense and sparse embeddings, neural re-ranking and query generation.³¹

Research and open-source work

Reimers created the Sentence Transformers library, first developed at the UKP Lab and released under the Apache-2.0 license, and with Gurevych wrote the 2019 Sentence-BERT paper. In 2020 Reimers and Gurevych showed how to make monolingual sentence embedding models multilingual through knowledge distillation, and Reimers co-authored Augmented SBERT, a data augmentation method for bi-encoders.⁵⁴⁶⁷

Much of Reimers's later work concerns evaluating and adapting neural search. Reimers co-authored BEIR (2021), a benchmark for zero-shot evaluation of retrieval models, GPL (2021), a generative pseudo-labeling method for adapting dense retrievers to new domains without labeled data, and MTEB (2022), a benchmark covering many text embedding tasks.⁸⁹¹⁰

Work and contributions

• Sentence Transformers Open-source code

  Reimers's vita describes Reimers as the creator of sentence-transformers. The Apache-2.0 repository's README cites the Sentence-BERT papers by Reimers and Gurevych and says the project was originally developed by the UKP Lab at TU Darmstadt.³⁵

• Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks Published research

  First author, with Iryna Gurevych, of the EMNLP 2019 paper that introduced SBERT, a modification of BERT using siamese and triplet networks to derive sentence embeddings.⁴

Timeline

1. 2014Began doctoral research in the UKP Lab at TU Darmstadt.³
2. 2018Completed a PhD in natural language processing and became a postdoctoral researcher at TU Darmstadt.³
3. Aug 2019Posted the Sentence-BERT paper with Iryna Gurevych.⁴
4. Apr 2020Posted a method for making sentence embeddings multilingual using knowledge distillation.⁶
5. 2021Joined Hugging Face as a research scientist to build a neural search research team.³
6. Apr 2021Co-authored BEIR, a benchmark for zero-shot evaluation of information retrieval models.⁸
7. Dec 2021Co-authored GPL, a method for unsupervised domain adaptation of dense retrieval.⁹
8. 2022Joined Cohere as Director of Machine Learning and Principal Scientist.³
9. Oct 2022Co-authored MTEB, the Massive Text Embedding Benchmark.¹⁰

Other documented roles

Roles the sources document that are not confirmed as current.

• Research Scientist, Hugging Face³
• Post-Doctoral Researcher, Technical University of Darmstadt (UKP Lab)³

Where they work

• Cohere 5 people in the wiki

Related people

• Aidan GomezAlso at Cohere
• Jay AlammarAlso at Cohere
• Joelle PineauAlso at Cohere
• Nick FrosstAlso at Cohere
• Gabe PereyraEvaluation and benchmarks · Language models
• Yuxin WuML frameworks and libraries · Language models
• Su Lin BlodgettEvaluation and benchmarks · Language models
• Kate SouleLanguage models · Evaluation and benchmarks
• AI ExplainedEvaluation and benchmarks · Language models
• Valentina PyatkinLanguage models · Evaluation and benchmarks