# Eugene Kharitonov  [](https://github.com/eugene-kharitonov/) [](https://scholar.google.com/citations?user=8PE1wjQAAAAJ&hl=en) [](https://twitter.com/n0mad_0) [my_mail] Hi there! I am working at FB AI Research Paris, where I am trying to find a balance between Research and Engineering. I have overly diverse interests: I am excited about understanding the internal workings of deep learning models, I study communicating neural agents, and build tools for applying NLP directly on speech [(FB AI blogpost on that!)](https://ai.facebook.com/blog/textless-nlp-generating-expressive-speech-from-raw-audio). I also worked on other topics, such as federated learning, search engine evaluation, and information retrieval. ## Sample of my current interests (full list of [publications](https://scholar.google.com/citations?user=8PE1wjQAAAAJ&hl=en)) * _Generative spoken language modeling from raw audio_. Kushal Lakhotia<sup>✝</sup>, Eugene Kharitonov<sup>✝</sup>, Wei-Ning Hsu, Yossi Adi, Adam Polyak, Benjamin Bolte, Tu-Anh Nguyen, Jade Copet, Alexei Baevski, Adelrahman Mohamed, and Emmanuel Dupoux. TACL 2021. [[arxiv]](https://arxiv.org/abs/2102.01192) * _What they do when in doubt: a study of inductive biases in seq2seq learners_. Eugene Kharitonov<sup>✝</sup>, Rahma Chaabouni<sup>✝</sup>. ICLR 2021. [[arxiv]](https://arxiv.org/abs/2006.14953) * _Entropy Minimization In Emergent Languages_. Eugene Kharitonov, Rahma Chaabouni, Diane Bouchacourt, Marco Baroni. ICML 2020. [[arxiv]](https://arxiv.org/abs/1905.13687) * _Compositionality and Generalization in Emergent Languages_. Rahma Chaabouni<sup>✝</sup>, Eugene Kharitonov<sup>✝</sup>, Diane Bouchacourt, Emmanuel Dupoux, Marco Baroni. ACL 2020. [[arxiv]](https://arxiv.org/abs/2004.09124) * _EGG: a platform for doing research in Language Emergence_. Eugene Kharitonov, Rahma Chaabouni, Diane Bouchacourt, Marco Baroni. EMNLP 2019 Demo. [[github]](https://github.com/facebookresearch/EGG) [[2 min screencast]](https://vimeo.com/345470060)[[arxiv]](https://arxiv.org/abs/1907.00852) ## Other notable things * _Learning Sensitive Combinations of A/B Test Metrics_. Eugene Kharitonov, Alexey Drutsa, Pavel Serdyukov. WSDM 2017. [[dl.acm]](https://dl.acm.org/doi/abs/10.1145/3018661.3018708) * octopi: Character-level language models + arithmetic coding = compression ♥ [[github]](https://github.com/n0mad/octopi) ## Education * PhD in CS, University of Glasgow, 2016 * MSc Applied Math & Physics, Moscow Institute of Physics and Technology (MIPT), 2009 * Yandex School of Data Analysis, 2009