Ankit Vani [he/him]
PhD candidate at Mila, Université de Montréal
E-mail: ![[firstname][dot][lastname]@umontreal.ca](email.png)
Github: ankitkv
I am a PhD candidate at Mila, Université de Montréal, under the supervision of Aaron Courville.
I am interested in studying the emergence of complex and interesting phenomena from simple rules. This includes trying to uncover the principles that drive intricate behavior in biological intelligence, as well as building systems that operate on proposed simple principles towards emerging complex phenomena like language and understanding. Presently, I am pursuing the emergence of systematicity in deep neural networks, a critical trait for artificial intelligence to generalize in the real world.
Education
- Ph.D., Machine Learning, Université de Montréal, ongoing
- M.S., Computer Science, New York University, 2017
- B.E., Computer Engineering, University of Pune, 2013
Research Interests
- Systematic generalization
- Language emergence
- Compositionality and disentanglement
- Representation learning
Publications and Preprints
- Samuel Lavoie, Christos Tsirigotis, Max Schwarzer, Ankit Vani, Michael Noukhovitch, Kenji Kawaguchi, Aaron Courville. Simplicial Embeddings in Self-Supervised Learning and Downstream Classification. International Conference on Learning Representations (ICLR), 2023. [pdf]
- Arian Hosseini, Ankit Vani, Dzmitry Bahdanau, Alessandro Sordoni, Aaron Courville. On the Compositional Generalization Gap of In-Context Learning. Fifth BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, 2022. [pdf]
- Hattie Zhou, Ankit Vani, Hugo Larochelle, Aaron Courville. Fortuitous Forgetting in Connectionist Networks. International Conference on Learning Representations (ICLR), 2022. [pdf]
- Ankit Vani, Max Schwarzer, Yuchen Lu, Eeshan Dhekane, Aaron Courville. Iterated learning for emergent systematicity in VQA. International Conference on Learning Representations (ICLR), 2021. [pdf]
- Jose Gallego, Ankit Vani, Max Schwarzer, Simon Lacoste-Julien. GAIT: A Geometric Approach to Information Theory. International Conference on Artificial Intelligence and Statistics (AISTATS), 2020. [pdf]
- Ankit Vani, Yacine Jernite, David Sontag. Grounded Recurrent Neural Networks. arXiv:1705.08557 Preprint, 2017. [pdf]
- Ramakrishna B Bairi, Ankit Vani, Pooja Ahuja, Ganesh Ramakrishnan. Categorising videos using a personalised category catalogue. Proceedings of the 2nd ACM IKDD Conference on Data Sciences, 2015. [pdf]
Research Reports
- Ankit Vani. Adversarial Discrete Sequence Generation. November 2017. [pdf] (advisor: Rob Fergus)
- Ankit Vani. Adversarial Objectives for Text Generation. December 2016. [pdf] (advisor: Kyunghyun Cho)
- Ankit Vani and Vighnesh Birodkar. Challenges with Variational Autoencoders for Text. December 2016. [pdf] (advisor: David Sontag)
double l[]={2.237245047823e-312,3.01793166842199e-307};main(){(1[l]+=*l)?printf(l+1),1[l]=-*l,main():puts(l);}