Andrey Filchenkov

Valeria Efimova, Artyom Chebykin, Ivan Jarsky et al.

Neural style transfer draws researchers' attention, but the interest focuses on bitmap images. Various models have been developed for bitmap image generation both online and offline with arbitrary and pre-trained styles. However, the style transfer between vector images has not almost been considered. Our research shows that applying standard content and style losses insignificantly changes the vector image drawing style because the structure of vector primitives differs a lot from pixels. To handle this problem, we introduce new loss functions. We also develop a new method based on differentiable rasterization that uses these loss functions and can change the color and shape parameters of the content image corresponding to the drawing of the style image. Qualitative experiments demonstrate the effectiveness of the proposed VectorNST method compared with the state-of-the-art neural style transfer approaches for bitmap images and the only existing approach for stylizing vector images, DiffVG. Although the proposed model does not achieve the quality and smoothness of style transfer between bitmap images, we consider our work an important early step in this area. VectorNST code and demo service are available at https://github.com/IzhanVarsky/VectorNST.

Valeria Efimova, Ivan Jarsky, Ilya Bizyaev et al.

Generative Adversarial Networks (GAN) have motivated a rapid growth of the domain of computer image synthesis. As almost all the existing image synthesis algorithms consider an image as a pixel matrix, the high-resolution image synthesis is complicated.A good alternative can be vector images. However, they belong to the highly sophisticated parametric space, which is a restriction for solving the task of synthesizing vector graphics by GANs. In this paper, we consider a specific application domain that softens this restriction dramatically allowing the usage of vector image synthesis. Music cover images should meet the requirements of Internet streaming services and printing standards, which imply high resolution of graphic materials without any additional requirements on the content of such images. Existing music cover image generation services do not analyze tracks themselves; however, some services mostly consider only genre tags. To generate music covers as vector images that reflect the music and consist of simple geometric objects, we suggest a GAN-based algorithm called CoverGAN. The assessment of resulting images is based on their correspondence to the music compared with AttnGAN and DALL-E text-to-image generation according to title or lyrics. Moreover, the significance of the patterns found by CoverGAN has been evaluated in terms of the correspondence of the generated cover images to the musical tracks. Listeners evaluate the music covers generated by the proposed algorithm as quite satisfactory and corresponding to the tracks. Music cover images generation code and demo are available at https://github.com/IzhanVarsky/CoverGAN.

Maria Dziuba, Ivan Jarsky, Valeria Efimova et al.

Nowadays, there are many diffusion and autoregressive models that show impressive results for generating images from text and other input domains. However, these methods are not intended for ultra-high-resolution image synthesis. Vector graphics are devoid of this disadvantage, so the generation of images in this format looks very promising. Instead of generating vector images directly, you can first synthesize a raster image and then apply vectorization. Vectorization is the process of converting a raster image into a similar vector image using primitive shapes. Besides being similar, generated vector image is also required to contain the minimum number of shapes for rendering. In this paper, we focus specifically on machine learning-compatible vectorization methods. We are considering Mang2Vec, Deep Vectorization of Technical Drawings, DiffVG, and LIVE models. We also provide a brief overview of existing online methods. We also recall other algorithmic methods, Im2Vec and ClipGEN models, but they do not participate in the comparison, since there is no open implementation of these methods or their official implementations do not work correctly. Our research shows that despite the ability to directly specify the number and type of shapes, existing machine learning methods work for a very long time and do not accurately recreate the original image. We believe that there is no fast universal automatic approach and human control is required for every method.

Arip Asadulaev, Vitaly Shutov, Alexander Korotin et al.

We present a novel algorithm for domain adaptation using optimal transport. In domain adaptation, the goal is to adapt a classifier trained on the source domain samples to the target domain. In our method, we use optimal transport to map target samples to the domain named source fiction. This domain differs from the source but is accurately classified by the source domain classifier. Our main idea is to generate a source fiction by c-cyclically monotone transformation over the target domain. If samples with the same labels in two domains are c-cyclically monotone, the optimal transport map between these domains preserves the class-wise structure, which is the main goal of domain adaptation. To generate a source fiction domain, we propose an algorithm that is based on our finding that adversarial attacks are a c-cyclically monotone transformation of the dataset. We conduct experiments on Digits and Modern Office-31 datasets and achieve improvement in performance for simple discrete optimal transport solvers for all adaptation tasks.

Arip Asadulaev, Alexander Panfilov, Andrey Filchenkov

Adversarial examples are transferable between different models. In our paper, we propose to use this property for multi-step domain adaptation. In unsupervised domain adaptation settings, we demonstrate that replacing the source domain with adversarial examples to $\mathcal{H} Δ\mathcal{H}$-divergence can improve source classifier accuracy on the target domain. Our method can be connected to most domain adaptation techniques. We conducted a range of experiments and achieved improvement in accuracy on Digits and Office-Home datasets.

Arip Asadulaev, Alexander Panfilov, Andrey Filchenkov

It was shown that adversarial examples improve object recognition. But what about their opposite side, easy examples? Easy examples are samples that the machine learning model classifies correctly with high confidence. In our paper, we are making the first step toward exploring the potential benefits of using easy examples in the training procedure of neural networks. We propose to use an auxiliary batch normalization for easy examples for the standard and robust accuracy improvement.

Arip Asadulaev, Rostislav Korst, Alexander Korotin et al.

We propose a novel algorithm for offline reinforcement learning using optimal transport. Typically, in offline reinforcement learning, the data is provided by various experts and some of them can be sub-optimal. To extract an efficient policy, it is necessary to \emph{stitch} the best behaviors from the dataset. To address this problem, we rethink offline reinforcement learning as an optimal transportation problem. And based on this, we present an algorithm that aims to find a policy that maps states to a \emph{partial} distribution of the best expert actions for each given state. We evaluate the performance of our algorithm on continuous control problems from the D4RL suite and demonstrate improvements over existing methods.

Natalia Khanzhina, Alexey Lapenok, Andrey Filchenkov

According to recent studies, commonly used computer vision datasets contain about 4% of label errors. For example, the COCO dataset is known for its high level of noise in data labels, which limits its use for training robust neural deep architectures in a real-world scenario. To model such a noise, in this paper we have proposed the homoscedastic aleatoric uncertainty estimation, and present a series of novel loss functions to address the problem of image object detection at scale. Specifically, the proposed functions are based on Bayesian inference and we have incorporated them into the common community-adopted object detection deep learning architecture RetinaNet. We have also shown that modeling of homoscedastic aleatoric uncertainty using our novel functions allows to increase the model interpretability and to improve the object detection performance being evaluated on the COCO dataset.

Qi Yang, Aleksandr Farseev, Andrey Filchenkov

Human personality traits are the key drivers behind our decision-making, influencing our life path on a daily basis. Inference of personality traits, such as Myers-Briggs Personality Type, as well as an understanding of dependencies between personality traits and users' behavior on various social media platforms is of crucial importance to modern research and industry applications. The emergence of diverse and cross-purpose social media avenues makes it possible to perform user personality profiling automatically and efficiently based on data represented across multiple data modalities. However, the research efforts on personality profiling from multi-source multi-modal social media data are relatively sparse, and the level of impact of different social network data on machine learning performance has yet to be comprehensively evaluated. Furthermore, there is not such dataset in the research community to benchmark. This study is one of the first attempts towards bridging such an important research gap. Specifically, in this work, we infer the Myers-Briggs Personality Type indicators, by applying a novel multi-view fusion framework, called "PERS" and comparing the performance results not just across data modalities but also with respect to different social network data sources. Our experimental results demonstrate the PERS's ability to learn from multi-view data for personality profiling by efficiently leveraging on the significantly different data arriving from diverse social multimedia sources. We have also found that the selection of a machine learning approach is of crucial importance when choosing social network data sources and that people tend to reveal multiple facets of their personality in different social media avenues. Our released social multimedia dataset facilitates future research on this direction.

Igor Kuznetsov, Andrey Filchenkov

Episodic memory lets reinforcement learning algorithms remember and exploit promising experience from the past to improve agent performance. Previous works on memory mechanisms show benefits of using episodic-based data structures for discrete action problems in terms of sample-efficiency. The application of episodic memory for continuous control with a large action space is not trivial. Our study aims to answer the question: can episodic memory be used to improve agent's performance in continuous control? Our proposed algorithm combines episodic memory with Actor-Critic architecture by modifying critic's objective. We further improve performance by introducing episodic-based replay buffer prioritization. We evaluate our algorithm on OpenAI gym domains and show greater sample-efficiency compared with the state-of-the art model-free off-policy algorithms.

Aleksandr Farseev, Qi Yang, Andrey Filchenkov et al.

In this technical demonstration, we showcase the World's first personality-driven marketing content generation platform, called SoMin.ai. The platform combines deep multi-view personality profiling framework and style generative adversarial networks facilitating the automatic creation of content that appeals to different human personality types. The platform can be used for the enhancement of the social networking user experience as well as for content marketing routines. Guided by the MBTI personality type, automatically derived from a user social network content, SoMin.ai generates new social media content based on the preferences of other users with a similar personality type aiming at enhancing the user experience on social networking venues as well diversifying the efforts of marketers when crafting new content for digital marketing campaigns. The real-time user feedback to the platform via the platform's GUI fine-tunes the content generation model and the evaluation results demonstrate the promising performance of the proposed multi-view personality profiling framework when being applied in the content generation scenario. By leveraging content generation at a large scale, marketers will be able to execute more effective digital marketing campaigns at a lower cost.

Gideon Stein, Andrey Filchenkov, Arip Asadulaev

Since the publication of the original Transformer architecture (Vaswani et al. 2017), Transformers revolutionized the field of Natural Language Processing. This, mainly due to their ability to understand timely dependencies better than competing RNN-based architectures. Surprisingly, this architecture change does not affect the field of Reinforcement Learning (RL), even though RNNs are quite popular in RL, and time dependencies are very common in RL. Recently, Parisotto et al. 2019) conducted the first promising research of Transformers in RL. To support the findings of this work, this paper seeks to provide an additional example of a Transformer-based RL method. Specifically, the goal is a simple Transformer-based Deep Q-Learning method that is stable over several environments. Due to the unstable nature of Transformers and RL, an extensive method search was conducted to arrive at a final method that leverages developments around Transformers as well as Q-learning. The proposed method can match the performance of classic Q-learning on control environments while showing potential on some selected Atari benchmarks. Furthermore, it was critically evaluated to give additional insights into the relation between Transformers and RL.

Qi Yang, Aleksandr Farseev, Andrey Filchenkov

Nowadays, social networks play a crucial role in human everyday life and no longer purely associated with spare time spending. In fact, instant communication with friends and colleagues has become an essential component of our daily interaction giving a raise of multiple new social network types emergence. By participating in such networks, individuals generate a multitude of data points that describe their activities from different perspectives and, for example, can be further used for applications such as personalized recommendation or user profiling. However, the impact of the different social media networks on machine learning model performance has not been studied comprehensively yet. Particularly, the literature on modeling multi-modal data from multiple social networks is relatively sparse, which had inspired us to take a deeper dive into the topic in this preliminary study. Specifically, in this work, we will study the performance of different machine learning models when being learned on multi-modal data from different social networks. Our initial experimental results reveal that social network choice impacts the performance and the proper selection of data source is crucial.

Arip Asadulaev, Igor Kuznetsov, Gideon Stein et al.

The outcome of Jacobian singular values regularization was studied for supervised learning problems. It also was shown that Jacobian conditioning regularization can help to avoid the ``mode-collapse'' problem in Generative Adversarial Networks. In this paper, we try to answer the following question: Can information about policy conditioning help to shape a more stable and general policy of reinforcement learning agents? To answer this question, we conduct a study of Jacobian conditioning behavior during policy optimization. To the best of our knowledge, this is the first work that research condition number in reinforcement learning agents. We propose a conditioning regularization algorithm and test its performance on the range of continuous control tasks. Finally, we compare algorithms on the CoinRun environment with separated train end test levels to analyze how conditioning regularization contributes to agents' generalization.

Arip Asadulaev, Igor Kuznetsov, Andrey Filchenkov

It is important to develop mathematically tractable models than can interpret knowledge extracted from the data and provide reasonable predictions. In this paper, we present a Linear Distillation Learning, a simple remedy to improve the performance of linear neural networks. Our approach is based on using a linear function for each class in a dataset, which is trained to simulate the output of a teacher linear network for each class separately. We tested our model on MNIST and Omniglot datasets in the Few-Shot learning manner. It showed better results than other interpretable models such as classical Logistic Regression.

Evgeny Zamyatin, Andrey Filchenkov

Generative adversarial networks (GANs) has gained tremendous popularity lately due to an ability to reinforce quality of its predictive model with generated objects and the quality of the generative model with and supervised feedback. GANs allow to synthesize images with a high degree of realism. However, the learning process of such models is a very complicated optimization problem and certain limitation for such models were found. It affects the choice of certain layers and nonlinearities when designing architectures. In particular, it does not allow to train convolutional GAN models with fully-connected hidden layers. In our work, we propose a modification of the previously described set of rules, as well as new approaches to designing architectures that will allow us to train more powerful GAN models. We show the effectiveness of our methods on the problem of synthesizing projections of 3D objects with the possibility of interpolation by class and view point.

Valeria Efimova, Andrey Filchenkov, Anatoly Shalyto

Many algorithms for data analysis exist, especially for classification problems. To solve a data analysis problem, a proper algorithm should be chosen, and also its hyperparameters should be selected. In this paper, we present a new method for the simultaneous selection of an algorithm and its hyperparameters. In order to do so, we reduced this problem to the multi-armed bandit problem. We consider an algorithm as an arm and algorithm hyperparameters search during a fixed time as the corresponding arm play. We also suggest a problem-specific reward function. We performed the experiments on 10 real datasets and compare the suggested method with the existing one implemented in Auto-WEKA. The results show that our method is significantly better in most of the cases and never worse than the Auto-WEKA.

Ivan Smetannikov, Ilya Isaev, Andrey Filchenkov

One of the classical problems in machine learning and data mining is feature selection. A feature selection algorithm is expected to be quick, and at the same time it should show high performance. MeLiF algorithm effectively solves this problem using ensembles of ranking filters. This article describes two different ways to improve MeLiF algorithm performance with parallelization. Experiments show that proposed schemes significantly improves algorithm performance and increase feature selection quality.