Hit papers significantly outperform the citation benchmark for their cohort. A paper qualifies
if it has ≥500 total citations, achieves ≥1.5× the top-1% citation threshold for papers in the
same subfield and year (this is the minimum needed to enter the top 1%, not the average
within it), or reaches the top citation threshold in at least one of its specific research
topics.
Describing Videos by Exploiting Temporal Structure
2015593 citationsLi Yao, Atousa Torabi et al.PolyPublie (École Polytechnique de Montréal)profile →
Countries citing papers authored by Nicolas Ballas
Since
Specialization
Citations
This map shows the geographic impact of Nicolas Ballas's research. It shows the number of citations coming from papers published by authors working in each country. You can also color the map by specialization and compare the number of citations received by Nicolas Ballas with the expected number of citations based on a country's size and research output (numbers larger than one mean the country cites Nicolas Ballas more than expected).
This network shows the impact of papers produced by Nicolas Ballas. Nodes represent research fields, and links connect fields that are likely to share authors. Colored nodes show fields that tend to cite the papers produced by Nicolas Ballas. The network helps show where Nicolas Ballas may publish in the future.
Co-authorship network of co-authors of Nicolas Ballas
This figure shows the co-authorship network connecting the top 25 collaborators of Nicolas Ballas.
A scholar is included among the top collaborators of Nicolas Ballas based on the total number of
citations received by their joint publications. Widths of edges
represent the number of papers authors have co-authored together.
Node borders
signify the number of papers an author published with Nicolas Ballas. Nicolas Ballas is excluded from
the visualization to improve readability, since they are connected to all nodes in the network.
Wang, Jianyu, et al.. (2020). SloMo: Improving Communication-Efficient Distributed SGD with Slow Momentum. International Conference on Learning Representations.9 indexed citations
Assran, Mahmoud, Joshua Romoff, Nicolas Ballas, Joëlle Pineau, & Michael Rabbat. (2019). Gossip-based Actor-Learner Architectures for Deep Reinforcement Learning. arXiv (Cornell University). 32. 13299–13309.2 indexed citations
6.
Jastrzȩbski, Stanisław, Zachary Kenton, Nicolas Ballas, et al.. (2019). On the Relation Between the Sharpest Directions of DNN Loss and the SGD Step Length. Edinburgh Research Explorer.4 indexed citations
7.
Jastrzȩbski, Stanisław, Zachary Kenton, Nicolas Ballas, et al.. (2018). DNN's Sharpest Directions Along the SGD Trajectory.. arXiv (Cornell University).1 indexed citations
8.
George, Thomas, César Laurent, Xavier Bouthillier, Nicolas Ballas, & Pascal Vincent. (2018). Fast Approximate Natural Gradient Descent in a Kronecker Factored Eigenbasis. Neural Information Processing Systems. 31. 9550–9560.9 indexed citations
9.
Jastrzȩbski, Stanisław, Zachary Kenton, Devansh Arpit, et al.. (2018). Finding Flatter Minima with SGD. International Conference on Learning Representations.3 indexed citations
10.
Jastrzȩbski, Stanisław, et al.. (2018). SGD Smooths The Sharpest Directions.1 indexed citations
Krueger, David, Nicolas Ballas, Stanisław Jastrzȩbski, et al.. (2017). Deep Nets Don't Learn via Memorization. PolyPublie (École Polytechnique de Montréal).23 indexed citations
Rankless uses publication and citation data sourced from OpenAlex, an open and comprehensive
bibliographic database. While OpenAlex provides broad and valuable coverage of the global
research landscape, it—like all bibliographic datasets—has inherent limitations. These include
incomplete records, variations in author disambiguation, differences in journal indexing, and
delays in data updates. As a result, some metrics and network relationships displayed in
Rankless may not fully capture the entirety of a scholar's output or impact.