Deep Learning

CIMRL: Combining IMitation and Reinforcement Learning for Safe Autonomous Driving
A novel approach to bring two worlds - Imitation and Reinforcement Learning - together for Safe Planning.
Jonathan Booher, Khashayar Rohanimanesh, Junhong Xu, Aleksandr Petiushko Александр Петюшко
PDF Cite DOI arXiv
CIMRL: Combining IMitation and Reinforcement Learning for Safe Autonomous Driving
Multi-Constraint Safe RL with Objective Suppression for Safety-Critical Applications
A novel approach to tackle the multiple constraints challenge in Safe RL.
Zihan Zhou, Jonathan Booher, Wei Liu, Aleksandr Petiushko Александр Петюшко, Animesh Garg
PDF Cite arXiv SymposiumWeb
Multi-Constraint Safe RL with Objective Suppression for Safety-Critical Applications
Mutual modality learning for video action classification
Mutual Modality (RGB, Optical Flow, RGBDiff, etc) Learning for Video Recognition.
Stepan Komkov, Maksim Dzabraev, Aleksandr Petiushko Александр Петюшко
PDF Cite Code Source Document DOI arXiv
Mutual modality learning for video action classification
Many Heads but One Brain: Fusion Brain - a Competition and a Single Multimodal Multitask Architecture
Multi-task multi-modal challenge, baseline, and methodology.
Daria Bakshandaeva, Denis Dimitrov, Vladimir Arkhipkin, Alex Shonenkov, Mark Potanin, Denis Karachev, Andrey Kuznetsov, Anton Voronov, Vera Davydova, Elena Tutubalina, Aleksandr Petiushko Александр Петюшко
PDF Cite Code Dataset Source Document DOI arXiv
Many Heads but One Brain: Fusion Brain - a Competition and a Single Multimodal Multitask Architecture
Nonparametric Uncertainty Quantification for Single Deterministic Neural Network
Elegant way of measuring both aleatoric and epistemic uncertainties for neural nets.
Nikita Kotelevskii, Aleksandr Artemenkov, Kirill Fedyanin, Fedor Noskov, Alexander Fishkov, Artem Shelmanov, Artem Vazhentsev, Aleksandr Petiushko Александр Петюшко, Maxim Panov
PDF Cite Slides Source Document arXiv OpenReview Conf
Nonparametric Uncertainty Quantification for Single Deterministic Neural Network
Smoothed Embeddings for Certified Few-Shot Learning
Geometrical approach for certified few-shot learning via randomized smoothing of templates.
Mikhail Pautov, Olesya Kuznetsova, Nurislam Tursynbek, Aleksandr Petiushko Александр Петюшко, Ivan Oseledets
PDF Cite Poster Slides Source Document arXiv OpenReview Conf
Smoothed Embeddings for Certified Few-Shot Learning
Mathematics and Theory in Deep Learning with Aleksandr Petiushko
Fireside chat about the importance of Math and Theory for ML/DL roles.
Nov 9, 2022 12:00 AM DLS Podcasts
Aleksandr Petiushko Александр Петюшко
YT Podcast Google Podcasts Apple Podcasts Spotify Spotify for Podcasters Yandex Music In Russian
Mathematics and Theory in Deep Learning with Aleksandr Petiushko
Certified Robustness via Randomized Smoothing over Multiplicative Parameters of Input Transformations
A novel method of certified robustness using Rayleigh dostribution for smoothing when transformations consist not additive but multiplicative group.
Nikita Muravev, Aleksandr Petiushko Александр Петюшко
PDF Cite Video Source Document DOI arXiv
Certified Robustness via Randomized Smoothing over Multiplicative Parameters of Input Transformations
Quadric Hypersurface Intersection for Manifold Learning in Feature Space
A novel approach to use the so-called “Manifold Hypothesis” in order to solve the outlier detection problem by quadrics on top of FaceID embeddings.
Fedor Pavutnitskiy, Sergei O. Ivanov, Evgeniy Abramov, Viacheslav Borovitskiy, Artem Klochkov, Viktor Vyalov, Anatolii Zaikovskii, Aleksandr Petiushko Александр Петюшко
PDF Cite Code Video Source Document arXiv
Quadric Hypersurface Intersection for Manifold Learning in Feature Space
CC-Cert: A Probabilistic Approach to Certify General Robustness of Neural Networks
A theoretical approach to estimate the certified robustness tested on different image transformations.
Mikhail Pautov, Nurislam Tursynbek, Marina Munkhoeva, Nikita Muravev, Aleksandr Petiushko Александр Петюшко, Ivan Oseledets
PDF Cite Poster Source Document DOI arXiv
CC-Cert: A Probabilistic Approach to Certify General Robustness of Neural Networks