About Me
PhD student in Machine Learning at Mila, with Prof Jian Tang. I am broadly interested in how learning can be improved through the use of graph representations, having previously worked on neural algorithmic reasoners for implicit planning and applications to biotechnology, focusing on drug discovery.
Publications
Neural Algorithmic Reasoners are Implicit Planners
Deac, A., Veličković, P., Milinković, O., Bacon, PL., Tang, J. and Nikolić, M.NeurIPS 2021 Spotlight talk
How to transfer algorithmic reasoning knowledge to learn new algorithms?
Xhonneux, L.-P. A. C., Deac, A., Veličković, P., Tang, J.NeurIPS 2021
Large-scale graph regression with very deep GNNs
Addanki, R., Battaglia, P. w., Budden, D., Deac, A., Godwin, J., Li, W. L. S., Sachez-Gonzalez, A., Stott, J., Shantanu, T. and Veličković, P.KDD Cup 2021
Large-scale node classification with bootstrapping
Addanki, R., Battaglia, P. w., Budden, D., Deac, A., Godwin, J., Keck, T., Sachez-Gonzalez, A., Stott, J., Shantanu, T. and Veličković, P.KDD Cup 2021
Neural message passing for joint paratope-epitope prediction
Del Vecchio, A., Deac, A., Liò, P. and Veličković, P.Computational Biology Workshop at ICML 2021
XLVIN:eXecuted Latent Value Iteration Nets
Deac, A., Veličković, P., Milinković, O., Bacon, PL., Tang, J. and Nikolić, M.Deep Reinforcement Learning Workshop at NeurIPS 2020, Learning Meets Combinatorial Algorithms Workshop at NeurIPS 2020
We use a latent value iteration executor on a state graph derived using self-supervised learning to design an implicit-planner within deep reinforcement learning.
Graph neural induction of value iteration
Deac, A., Bacon, PL. and Tang, J.Graph Representation Learning and Beyond Workshop at ICML 2020
We propose a GNN-executor aimed at modelling the value iteration (VI) algorithm, across arbitrary environment models, with direct supervision on the intermediate steps of VI.
Structured Multi-View Representations for Drug Combinations
Liu*, S., Deac*, A., Zhu, Z. and Tang, J.Machine Learning for Molecules Workshop at NeurIPS 2020
We introduce a multi-view framework for drug combinations which leverages information from the drugs’ chemical structure, while also matching the sets of the drugs’ target proteins.
Empowering Graph Representation Learning with Paired Training and Graph Co-Attention
Deac, A., Huang, YH., Veličković, P., Liò, P. and Tang, J.Pre-print
We use graph co-attention in a paired graph training system for graph classification and regression.
Drug-Drug Adverse Effect Prediction with Graph Co-Attention
Deac, A., Huang, YH., Veličković, P., Liò, P. and Tang, J.ICML 2019 Workshop on Computational Biology
We propose a neural network architecture able to set state-of-the-art results on the drug-drug interaction (DDI) task—using the type of the side-effect and the molecular structure of the drugs alone—by leveraging a co-attentional mechanism.
Attentive cross-modal paratope prediction
Deac, A., Veličković, P. and Sormanni, P.Journal of Computational Biology (2019)
ICML 2018 Workshop on Computational Biology (contributed talk)
We use self and cross-modal attention to predict binding probabilities of antibody residues, obtaining state-of-the-art performance as well as new qualitative insights.
Education
Graph representation learning with applications to drug discovery, supervised by Prof Jian Tang.
Murray Edwards College
Honours Pass *with Distinction*
For my dissertation project, I have developed a novel conditional graph-variational autoencoder architecture for targeted drug design.
Murray Edwards College
For my dissertation project, I leveraged neural network architectures to analyse which amino acids participate in antibody-antigen interactions.
Experience
Montréal team
I worked at the intersection of graph representation learning and reinforcement learning with Doina Precup.
Generative Chemistry Team
I analyzed the effectiveness of learning molecular representations conditioned on the information from the target proteins for the task of binding affinity prediction in a low-data scenario.
Ads Quality Team
My project was focused on developing novel methodologies for keyword scoring.
Supervised by Prof Jian Tang
I worked on graph-based neural networks for molecule generation and drug-drug side-effect prediction.
Google Assistant Team
My project consisted of implementing a new notifications feature end-to-end. My focus was on reminders in particular, using C++ on the back end side and Java/Android for front end.
Google Hangouts Meet Team
I worked on improving the testing infrastructure of the Android application. The project’s goal was to do UI automation testing, which included working with Java, Python, Android, Espresso, dependency injection and Dagger.
Scholarships and Awards
Google Intern Award for Grace Hopper Celebration
2019Full travel award for the 2019 Grace Hopper Celebration conference in Orlando, United States
Google Prize for the Best Part III Research Project
2019Best Computer Science Master's research project at the University of Cambridge for 2018-19
Best presentation prize at Oxbridge Women in Computer Science Conference
2019Awarded for my work on predicting drug-drug interactions (DDIs)
Rising star prize at Oxbridge Women in Computer Science Conference
2018Awarded for my work on antibody-antigen interaction prediction
Paula Browne Scholarship from Murray Edwards College
2018Awarded to up to four undergraduates per academic year