Data Analysis and Machine Learning
Gradient-based adversarial attacks on categorical sequence models via traversing an embedded world
Ivan Fursov, Alexey Zaytsev, Nikita Klyuchnikov, Andrey Kravchenko and Evgeny Burnaev
The authors show the importance of threats by over trust in deep learning models for validation of categorical sequences, which may lead to fraudulent money transactions, medical fraud, and the spreading of fictional texts.
Deep Learning on Point Clouds for False Positive Reduction at Nodule Detection in Chest CT Scans
Ivan Drokin and Elena Ericheva
The authors proposed novel representation of scanned data (point cloud) and PointNet-based architecture for reduction of false positives in medical classification. All 3 reviewers voted for “strong accept” score.
Natural Language Processing
Do topics make a metaphor? Topic modeling for metaphor identification and analysis in Russian
Yulia Badryzlova, Anastasia Nikiforova and Olga Lyashevskaya
For an interesting combination of linguistic research questions (the nature of metaphor) and advanced NLP methods (topic modeling).
Social Network Analysis
Detecting Automatically Managed Accounts in Online Social Networks: Graph Embedding Approach
Ilia Karpov, Ekaterina Glazkova
The authors introduce a unique open dataset of bot accounts from a Russian social network that have been selected according to a rigorous definition. They also show that bots differ in their level of complexity, and that different machine learning approaches are needed to detect these different types. The proposed algorithms demonstrate a very high quality.
Theoretical Machine Learning and Optimization
Fast Approximation Algorithms for Stabbing Special Families of Line Segments with Equal Disks
In the paper, novel polynomial-time constant-factor approximation algorithms for several geometric settings of the well known Hitting Set combinatorial optimization problem are proposed. The algorithms have state-of-the-art expected performance guarantees and can be applied for solving relevant practical problems in operations research.
Checking Conformance between Colored Petri Nets and Event Logs
Julio Cesar Carrasquel, Khalil Mecheraoui, Irina Lomazova
The authors proposed an algorithm for conformance checking between colored Petri nets and event logs. The paper describes a solid theoretical approach broadening the existing conformance checking techniques.