Octavian Ganea

UPDATE: I will join NYU Courant and NYU Tandon as a tenure-track assistant professor in fall 2022. My research focus will be on geometric deep learning inspired by problems in structural biology, drug discovery and others. If you want to build the next generation of drug discovery technologies, please reach out! I am looking for postdocs, PhD students and visiting researchers!

I am a postdoctoral researcher at CSAIL-MIT working with prof. Tommi Jaakkola and prof. Regina Barzilay on AI solutions for drug discovery and structural biology using geometric and physical inductive biases. I am part of the Machine Learning for Pharmaceutical Discovery and Synthesis consortium, the Abdul Latif Jameel Clinic for Machine Learning in Health, and the DARPA Accelerated Molecular Discovery program. I am also a member of the ELLIS society. I received my PhD from the Data Analytics Lab at ETH Zurich under the supervision of prof. Thomas Hofmann working on non-Euclidean representation learning for graphs, hierarchical data, and natural language processing.

profile photo Picture taken from the summit of Zinalrothorn, with the famous Weisshorn in the background.

CV  /  Google Scholar  /  Publications  /  Awards  /  Invited Talks  /  Supervision  /  Teaching  /  Service  /  Misc & Mountains

Research

I model and embed geometry in machine learning. My recent research aims to accelerate drug discovery and material generation, for which geometrical modeling is a core pillar. My deep learning solutions explicitly incorporate geometrical and physical first principles that naturally constrain the 3D structures and the interactions of molecules, proteins, or materials. Euclidean symmetries and other laws governing single and multibody systems are injected in my models to increase quality, efficiency and user trust. My research has wide implications in other related areas such as robotics or 3D graphics.

Geometric characteristics of real world data sometimes go beyond our 3D intuitions. In my past research, I also challenged fundamental geometrical assumptions of representation learning. I advocate for going beyond the traditional Euclidean space to prevent the loss of important structural information for certain ubiquitous types of data. I created principled and efficient deep learning and embedding methods rooted in Riemannian geometry, for instance by leveraging the power and flexibility of hyperbolic and elliptic spaces. My models have been improving machine learning solutions in various areas such as computer vision or natural language processing.



Publications

Featured Publications on Euclidean 3D Models for Chemistry/Biology


EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction
H. Stark*, O. Ganea*, L. Pattanaik, R. Barzilay, T. S. Jaakkola.

Full paper at ICML 2022: International Conference on Machine Learning.

[Paper] [Pitch] [Data + Code]

Independent SE(3)-Equivariant Models for End-to-End Rigid Protein Docking
O. Ganea*, X. Huang*, C. Bunne, Y. Bian, R. Barzilay, T. S. Jaakkola, A. Krause.

Spotlight at ICLR 2022 (International Conference on Learning Representations)
Top 15 among 3326 ICLR submissions (top 0.4 %) ranked by average review score.

[Paper] [Slides + Video] [Poster] [Data + Code] [MIT News coverage]

GeoMol: Torsional Geometric Generation of Molecular 3D Conformer Ensembles
O. Ganea*, L. Pattanaik*, C. Coley, R. Barzilay, K. Jensen, W. Green, T. Jaakkola

Full paper @ NeurIPS 2021: Conference on Neural Information Processing Systems.

Spotlight (top 3% of all submitted papers).

[Paper] [Poster] [Slides] [Data + Code] [MIT News coverage]

Crystal Diffusion Variational Autoencoder for Periodic Material Generation
T. Xie*, X. Fu*, Octavian-Eugen Ganea* (equal first author), R. Barzilay, T. Jaakkola .

Full paper at ICLR 2022 (International Conference on Learning Representations)

[Paper] [Poster] [Data + Code]

Featured Publications on Foundations of Non-Euclidean Machine Learning

Hyperbolic Neural Networks
O. Ganea*, G. Bécigneul*, T. Hofmann.

Full paper @ NeurIPS 2018: Conference on Neural Information Processing Systems.

Spotlight (top 4% of all submitted papers).

[Paper] [Video] [Poster] [Data + Code]

Hyperbolic Entailment Cones for Learning Hierarchical Embeddings
O. Ganea, G. Bécigneul, T. Hofmann.

Full paper and Oral talk at ICML 2018: International Conference on Machine Learning.

[PDF] [Slides] [Poster] [Data + Code]

Poincaré GloVe: Hyperbolic Word Embeddings
A. Tifrea*, G. Bécigneul*, O. Ganea*.

Full paper @ ICLR 2019: International Conference on Learning Representations.

[Paper] [Data + Code] [Poster]

Constant Curvature Graph Convolutional Networks
G. Bachmann, G. Bécigneul, O. Ganea

Full paper @ ICML 2020: International Conference on Machine Learning.

[Paper] [Slides]

Featured Publications on Natural Language Processing

Breaking the Softmax Bottleneck via Learnable Monotonic Pointwise Non-linearities
O. Ganea, S. Gelly, G. Bécigneul, A. Severyn.

Full paper and Oral talk at ICML 2019: International Conference on Machine Learning.

[Paper] [Slides]

Deep Joint Entity Disambiguation with Local Neural Attention
O. Ganea, T. Hofmann.

Full paper @ EMNLP 2017: Conference on Empirical Methods in Natural Language Processing.

[PDF] [Slides] [Poster] [Data + Code]

Other Publications (chronologically)

Computationally Tractable Riemannian Manifolds for Graph Embeddings
C. Cruceru, G. Bécigneul, O. Ganea

Full paper @ AAAI 2021: The Thirty-Fifth AAAI Conference on Artificial Intelligence.

[Paper] [Poster] [Code + Data]

Message Passing Networks for Molecules with Tetrahedral Chirality
L. Pattanaik, O. Ganea, I. Coley, K. Jensen, W. Green, C. Coley.

Paper @ NeurIPS'20 Machine Learning for Molecules Workshop

[Paper] [Video + Slides] [Data + Code]

Optimal Transport Graph Neural Networks
B. Chen*, Gary Bécigneul*, O. Ganea* (equal first author) , R. Barzilay, T. Jaakkola

Preprint.

[Paper] [Slides] [Data + Code]

Mixed-curvature Variational Autoencoders
O. Skopek, O. Ganea, G. Bécigneul.

Full paper @ ICLR 2020: International Conference on Learning Representations.

[Paper] [Code + Data] [Slides + Video]

Hierarchical Image Classification using Entailment Cone Embeddings
A. Dhall, A. Makarova, O. Ganea, D. Pavllo, M. Greeff, A. Krause

Paper @ DiffCVML 2020: Intl. Workshop on Differential Geometry in Computer Vision and Machine Learning

[Paper] [Slides] [Blog post] [Code + Data]

Riemannian Adaptive Optimization Methods
G. Bécigneul, O. Ganea

Full paper @ ICLR 2019: International Conference on Learning Representations.

[PDF] [External Code] [Poster]

End-to-end Neural Entity Linking
N. Kolitsas*, O. Ganea*, T. Hofmann.

Full paper @ CoNLL 2018: Conference on Natural Language Learning.

[PDF] [Data + Code]

Probabilistic Bag-Of-Hyperlinks Model for Entity Linking
O. Ganea, M. Ganea, A. Lucchi, C. Eickhoff, T. Hofmann.

Full paper @ WWW 2016: International World Wide Web Conference. Oral talk.

[PDF] [Slides] [Poster] [Data + Code] [Online system (Gerbil - D2KB)] [Comparison with existing systems (Jan 2018)]

Web2Text: Deep Structured Boilerplate Removal
T. Vogels, O. Ganea, C. Eickhoff .

Long paper @ ECIR 2018: European Conference on Information Retrieval.

[PDF] [Slides] [Source Code]



Awards
2021 - Outstanding Reviewer Award - NeurIPS 2021 Thirty-fifth Conference on Neural Information Processing Systems
2021 - Top 3% of all submitted papers - NeurIPS 2021 Spotlight paper at Conference on Neural Information Processing Systems
2019 - Fellowship Grant - Institute for Advanced Study for the special-year program 2019 - 2020 in Machine Learning led by Sanjeev Arora. (declined)
2018 - Top 4% of all submitted papers - NeurIPS 2018 Spotlight paper at Conference on Neural Information Processing Systems
2013 - Top 5% - Google Code Jam algorithms contest
2010 - Excellence scholarship 30,000$ - Dinu Patriciu foundation (Romania) - for master studies at EPFL
2010, 2009, 2008 - Excellence Teaching Diplomas, awarded three times by the Prime-minister of Romania and/or Ministry of Education and Research in Romania for my teaching activity for international mathematical contests and olympiads, e.g., my students won 5 gold and silver medals at IMO.
2009 - 1st prize - International Mathematical Contest for University Students IMC (www.imc-math.org) - the equivalent of IMO for BSc students
2008 - Gold Medal - South Eastern Mathematical Olympiad for University Students -- SEEMOUS - Athens, Greece
2008 - 2nd prize - International Mathematical Contest for University Students IMC (www.imc-math.org)
2007 - 2nd prize - International Mathematical Contest for University Students IMC (www.imc-math.org)
2007 - Silver Medal - South Eastern Mathematical Olympiad for University Students -- SEEMOUS - Cyprus
2005 - Silver Medal (3 rd place) - Tuymada International Olympiad in Mathematics, Yakutsk, Russia
2005 - Silver Medal - Balkan Mathematical Olympiad -- Iasi, Romania
2006, 2005, 2004, 2003, 2002 - Gold Medal, Romanian National Mathematical Olympiad


Invited Talks
May 2022: Moderna Inc., hosts: Eric Ma, James Ross
April 2022: Keynote talk, Deep Generative Models for Highly Structured Data (ICLR 2022 Workshop)
April 2022: Boston Area Group for Informatics and Modeling (BAGIM), host: Brian Kelley
March 2022: Invited speaker, Workshop on Functional Inference and Machine Intelligence
March 2022: Mila - Quebec AI Institute and Valence Discovery (video recording)
March 2022: New York University, host departments: CS, CDS, CSE, ECE
March 2022: Relay Therapeutics, host: Patrick Walters
March 2022: University of Waterloo, hosts: Lila Kari, Olga Veksler, Ian Goldberg
February 2022: Pfizer Inc., host: Vishnu Sresht
February 2022: Amazon Research UK, host: Tom Diethe
February 2022: Sanofi S.A., host: Yu Qiu
February 2022: University of Illinois at Urbana-Champaign, CSLS conference
January 2022: Entos AI, host: Tom Miller
January 2022: Microsoft Research UK (Cambridge) and Netherlands (Amsterdam), hosts: Max Welling, Rianne van den Berg
January 2022: Intel AI Lab, host: Mariano Phielipp
November 2021: Imperial College London, host: Michael Bronstein
November 2021: Mila - Quebec AI Institute, host: Jian Tang
June 2021: Microsoft Research New England, host: David Alvarez-Melis
April 2021: Machine Learning for Pharmaceutical Discovery and Synthesis Consortium, https://mlpds.mit.edu/
October 2020: MIT, guest lecture: 6.867 Machine Learning lecture
July 2020: Georgia Institute of Technology, host: Le Song
July 2020: Northeastern University, hosts: Tina Eliassi-Rad, Dmitri Krioukov, Rose Yu
December 2020: Bowdoin College, guest lecture, host: Jennifer Taback
April 2020: Machine Learning for Pharmaceutical Discovery and Synthesis Consortium, https://mlpds.mit.edu/
March 2020: IBM Global Business Services, host: Lucia Stavarache
<
February 2020: University of Massachusetts Amherst, host: Andrew McCallum



January 2020: Relational.ai, host: Nikos Vasilakis

October 2019: Aggregate Intellect - https://aisc.ai.science, host: Amir Feizpour
October 2019: MIT - Massachusetts Institute of Technology, prof. Tommi Jaakkola’s group
October 2018: Google Brain, Zurich, host: Sylvain Gelly
April 2018: ETH Zurich, Machine Learning Seminar
May 2017: Google Research, Mountain View, California
March 2017: ETH Zurich, Machine Learning Seminar


Mentoring and Research Supervision

I proposed and (co-)supervised several student research projects, some being later pushed into research publications. MSc thesis is 6 months long, BSc thesis is 4-6 months long:

Andreea Musat: Modeling molecular interactions using geometric deep learning
MSc thesis, MIT and Technical University of Munich, January - July 2022
Victor Armegioiu: Mdeling molecular interactions using geometric deep learning
MSc thesis, MIT and Technical University of Munich, January - July 2022
Hannes Stark: End-to-end Geometric Drug Binding
Research internship, MIT and Technical University of Munich, Sep 2021 - present
Julia Bala: Generic Riemannian Embedding Spaces
UROP project, MIT, January - June 2021
Xinyuan Huang: Modeling Protein Complexes
MSc thesis (jointly with Charlotte Bunne and Yatao Bian), MIT and ETH Zurich, January - July 2021
Panayiotou Panayiotis: Permutation Invariant Graph Generation via Optimal Transport
MSc thesis, MIT and ETH, April - October 2020
Octav Dragoi: Permutation Invariant Graph Generation and Optimization
MSc thesis, MIT and TU Munich, April - October 2020
Bachmann Gregor: Constant Curvature Graph Neural Networks, (paper @ICML'20)
MSc thesis (jointly with Gary Becigneul), ETH, March - September 2019
Calin Cruceru: Matrix Graph Embeddings, (paper @ AAAI'21)
MSc thesis (jointly with Gary Becigneul), ETH, April - October 2019
Ondrej Skopek: Mixed-curvature Variational Autoencoders, (paper @ICLR'20)
MSc thesis (jointly with Gary Becigneul), ETH, March - September 2019
Andreas Bloch: Mixed-curvature Recommender Systems,
MSc thesis (jointly with Gary Becigneul), ETH, March - September 2019
Ankit Dhall: Hierarchical Image Captioning using Entailment Cone Representations, (paper @DiffCVML'20)
MSc thesis (jointly with Andreas Krause and Anastasia Makarova), ETH, March - September 2019
Philipp Wirth: Language Models with External Advisers,
MSc thesis (jointly with Gary Becigneul), ETH, March - September 2019
Jovan Andonov: Neural Ordinary Differential Equations for Language Modeling,
MSc thesis (jointly with Gary Becigneul and Paulina Grnarova), ETH, March - September 2019
Alexandru Tifrea: Hyperbolic Word Embeddings, (paper @ ICLR'19)
MSc thesis (jointly with Gary Becigneul), ETH, April - October 2018
Kolitsas Nikolaos: End-to-end Neural Entity Linking, (paper @ CONLL'18)
MSc thesis, ETH, October 2017 - April 2018
Junlin Yao: Detecting Medication and Adverse Drug Events from Electronic Health Records,
MSc thesis (jointly with Carsten Eickhoff), ETH, September 2017 - March 2018
Igor Petrovski: Hyperbolic Sentence Embeddings,
MSc thesis (jointly with Gary Becigneul), ETH, October 2017 - April 2018
Valentin Trivonov: Sparse and Interpretable Embeddings, (paper @ EMNLP'18 Workshop)
MSc thesis (jointly supervised with Anna Potapenko), ETH, December 2017 - May 2018
Andreas Hess: Reinforcement Learning for Question Answering with Semi-structured Tables,
MSc thesis, ETH, December 2016 - May 2017
Yifan Su: Deep Structured Prediction for Joint Entity Linking and Coreference Resolution,
MSc thesis (jointly supervised with Aurelien Lucchi), ETH, July - December 2016
Till Haug: Convolutional and Recursive Neural Networks for Question Answering on Semi-structured Tables , (paper @ ECIR'18)
BSc thesis (jointly supervised with Paulina Grnarova), ETH, August 2016 - January 2017
Severin Bahman: Memory Networks for Entity Linking,
Research project (jointly supervised with Aurelien Lucchi), ETH, March - July 2016
Andreas Georgiadis: Learning Sentence and Entity Representations for Question Answering,
Research project (jointly supervised with Aurelien Lucchi), ETH, March - July 2016
Thijs Vogels: Structured Prediction for Web Page Content Extraction, (paper @ ECIR'18)
Research project (jointly supervised with Carsten Eickhoff), ETH, Sep 2015 - May 2016
Monteiro Freire Ribeiro Joao Pedro: Unsupervised Knowledge Base Fact Prediction using Matrix Word Embeddings,
BSc thesis (jointly supervised with Paulina Grnarova), ETH, Nov 2015 - May 2016


Teaching
2020: MIT, guest lecture: 6.867 Machine Learning lecture -- (sole instructor)
2020: lectures for International Mathematics Olympiad preparations for Romania’s team - website -- (sole instructor)
2008-2018: Teaching Assistant for the lectures:

Deep Learning (2017, 2018 - ETHZ)

Computational Intelligence Lab: 2015, 2016, 2017, 2018 (head TA) - ETHZ

Information Retrieval (2014, 2015,2016 - ETHZ)

Advanced Algorithms (2011 - EPFL)

Graph Theory (2011 - EPFL)

Concurrency (2011 - EPFL)

Numerical Methods (2008-2009 - UPB)

2007 - 2010: Lecturer for Mathematics Olympiads and Competitions and Member of the Romanian national selection committee for the International Mathematical Olympiad (IMO). I taught mathematics lectures to the candidates aiming to represent Romania at the IMO. Several of my students won prizes and medals at international contests, including 5 gold and silver medals at IMO (Stefan Ivanovici, Octav Dragoi, Ioana-Maria Tamas). -- (sole instructor)


Service
Co-founder of OpenConsulting.AI, a free AI consulting platform for societal non-profit projects.
Conference Reviewer: ICML 2022, NeurIPS 2021 (Outstanding Reviewer Award), NeurIPS 2020, ICML 2020, NeurIPS 2019, AAAI 2020, ACL 2018, EMNLP 2019, EMNLP 2018.

Journal reviewer: JMLR 2022
Research mentor at London Geometry and Machine Learning Summer School, logml.ai, July 2021.


Besides Research (MISC)
Completed the longest via ferrata in Switzerland (Leukerbad) - difficulty level K5-6. video
Climbed 10 four-thousand meter summits.
Videos: Zinalrothorn (4221m),    Mönch (4107m),    Mont Blanc traverse (via 3 monts),    Dufourspitze (4634m)
Finished 6 mountain and 1 road marathons.
Media coverage: Forbes "30 under 30", Romania’s 2013 edition

website credits