Tomáš Mikolov is a Czech computer scientist recognized for his contributions to Machine learning and Neural network language models. He is best known as the lead author of the 2013 paper that introduced word2vec, a technique for learning word embeddings from text, which became a foundational tool in Natural language processing and Deep learning. His later research includes work on fastText and cross-lingual word representations.
Mikolov received his PhD from Brno University of Technology and held research positions at Microsoft Research, Google Brain, and Facebook AI Research. In 2020, he joined the Czech Institute of Informatics, Robotics and Cybernetics at the Czech Technical University in Prague, where he leads a research group focused on scalable mathematical models.
Career
During his doctoral studies at Brno University of Technology, Mikolov spent time at Johns Hopkins University, with a visit arranged through the support of Sanjeev Khudanpur and Frederick Jelinek. He also spent several months in Yoshua Bengio's machine-learning laboratory at the Université de Montréal, where he worked on Recurrent neural network language models.
After completing his PhD in 2012, Mikolov joined Google Brain, part of Google's AI research ecosystem. In 2014, he moved to Facebook AI Research (FAIR), where he worked on Natural language processing and broader Artificial intelligence research. His time at FAIR included collaborations on efficient text classification and multilingual embeddings.
In 2020, Mikolov returned to the Czech Republic and joined the Czech Institute of Informatics, Robotics and Cybernetics at the Czech Technical University in Prague. He established a research group investigating mathematical models that can increase in complexity, aiming to improve scalability and efficiency in Machine learning. In 2025, he co-founded BottleCap AI, a Prague-based company developing efficient foundation models.
Research Contributions
Mikolov's early work focused on applying recurrent neural networks to language modelling. His 2012 doctoral dissertation, Statistical Language Models Based on Neural Networks, examined neural-network approaches to predicting text sequences, building on earlier work in Sequence-to-Sequence (Seq2Seq) learning and language-model training.
At Google, he led the development of word2vec, a method for learning word embeddings from large text collections. The approach used shallow neural networks to produce dense vector representations of words, capturing semantic and syntactic relationships. A follow-up paper, Distributed Representations of Words and Phrases and their Compositionality, introduced techniques such as negative sampling and subsampling of frequent words, which improved training efficiency and quality. This paper received the 2023 NeurIPS Test of Time Award.
While at Facebook AI Research, Mikolov co-authored work on fastText, including methods for text classification and word representations based on character subwords. This allowed the model to handle out-of-vocabulary words and morphologically rich languages. He also worked on mapping word representations between languages, enabling the construction of bilingual dictionaries and improving statistical-machine-translation systems without parallel corpora.
Impact and Legacy
Word2vec became a cornerstone of modern Deep learning research, influencing later developments in Transformer (architecture) architectures and Large language model training. The embeddings produced by word2vec were widely adopted for tasks such as sentiment-analysis, information-retrieval, and Machine translation. Mikolov's emphasis on efficient, scalable methods anticipated the shift toward practical Generative AI systems.
His work on fastText contributed to efficient text classification in resource-constrained settings, and his cross-lingual embedding research opened avenues for low-resource language processing. As of the early 2020s, his publications remained among the most cited in the field of Machine learning.
Current Work
At the Czech Institute of Informatics, Robotics and Cybernetics, Mikolov's group explores models that can grow in complexity, potentially addressing limitations of fixed-capacity neural networks. BottleCap AI, co-founded in 2025, focuses on developing efficient foundation models, continuing his interest in reducing computational costs while maintaining performance. His ongoing research aims to bridge theoretical insights with practical applications in Artificial intelligence.