Адрес: г. Москва, Кочновский проезд, д. 3 (недалеко от станции метро "Аэропорт").
Телефон: +7(495) 772-95-90 *22668
Кузнецов Сергей Олегович
Заместитель руководителя департамента
Громов Василий Александрович
Антропова Лариса Ивановна
Sign languages are the main way for people from deaf community to communicate with other people. In this paper, we have compared several real-time sign language dactyl recognition systems using deep convolutional neural networks. Our system is able to recognize words from natural language gestured using signs for each letter. We evaluate our approach on American (ASL) and Russian (RSL) sign languages. For ASL, we trained on dataset prepared by Massey University, Institute of Information and Mathematical Sciences, for RSL we collect our own dataset, which we aim to enlarge together with RSL community in Russia. The results showed 100% accuracy for ASL Massey dataset, while RSL recognition quality is behind sufficient quality due to much more complex nature of real-world RSL dataset.
It has recently been shown that first-order- and datalog-rewritability of ontology-mediated queries (OMQs) with expressive ontologies can be checked in NExpTime using a reduction to CSPs. In this paper, we present a case study for OMQs with Boolean conjunctive queries and a fixed ontology consisting of a single covering axiom 𝐴 -> 𝐹 v 𝑇, A -> F v T, possibly supplemented with a disjointness axiom for T and F. The ultimate aim is to classify such OMQs according to their data complexity: AC0, L, NL, P or coNP. We report on our experience with trying to distinguish between OMQs in P and coNP using the reduction to CSPs and the Polyanna software for finding polymorphisms.
An increasing number of algorithms in deep reinforcement learning area creates new challenges for environments, particularly, for their comprehensive analysis and searching application areas. The key purpose of this article is to provide an extensible environment for researches. We consider a Match-3 game, which has simple gameplay, but challenging game design for engaging players. The article provides metrics for evaluation of agents and corresponding baselines in different scenarios.
In this work, we study the effect of combining existent improvements for Deep Q-Networks (DQN) in Markov Decision Processes (MDP) and Partially Observable MDP (POMDP) settings. Combinations of several heuristics, such as Distributional Learning and Dueling architectures improvements, for MDP are well-studied. We propose a new combination method of simple DQN extensions and develop a new model-free reinforcement learning agent, which works with POMDP and uses well-studied improvements from fully observable MDP. To test our agent we choose the VizDoom environment, which is old first person shooter, and the Health Gathering scenario. We prove that improvements used in MDP setting may be used in POMDP setting as well and our combined agents can converge to better policies. We develop an agent with combination of several improvements showing superior game performance in practice. We compare our agent with Recurrent DQN using Prioritized Experience Replay and Snaphot Ensembling agent and get approximately triple increase in per episode reward.
We present a study on co-authorship network representation based on network embedding together with additional information on topic modeling of research papers and new edge embedding operator. We use the link prediction (LP) model for constructing a recommender system for searching collaborators with similar research interests. Extracting topics for each paper, we construct keywords co-occurrence network and use its embedding for further generalizing author attributes. Standard graph feature engineering and network embedding methods were combined for constructing co-author recommender system formulated as LP problem and prediction of future graph structure. We evaluate our survey on the dataset containing temporal information on National Research University Higher School of Economics over 25 years of research articles indexed in Russian Science Citation Index and Scopus. Our model of network representation shows better performance for stated binary classification tasks on several co-authorship networks.
Depth map super-resolution is a challenging computer vision problem. In this paper, we present two deep convolutional neural networks solving the problem of single depth map super-resolution. Both networks learn residual decomposition and trained with specific perceptual loss improving sharpness and perceptive quality of the upsampled depth map. Several experiments on various depth super-resolution benchmark datasets show state-of-art performance in terms of RMSE, SSIM, and PSNR metrics while allowing us to process depth super-resolution in real time with over 25-30 frames per second rate.
The progress of deep learning models in image and video processing leads to new artificial intelligence applications in Fashion industry. We consider the application of Generative Adversarial Networks and Neural Style Transfer for Digital Fashion presented as Virtual fashion for trying new clothes. Our model generate humans in clothes with respect to different fashion preferences, color layouts and fashion style. We propose that the virtual fashion industry will be highly impacted by accuracy of generating personalized human model taking into account different aspects of product and human preferences. We compare our model with state-of-art VITON model and show that using new perceptual loss in deep neural network architecture lead to better qualitative results in generating humans in clothes.
Co-authorship networks represent a graph, in which vertices are authors, and edges represent research papers written in co-authorship. Every paper could generate several edges in such a graph, if a number of coauthors is greater than two. Co-authorship networks play important role in understanding the structure of research collaborations usually resulted in joint research papers. Moreover, when analyzing university ranking and research staff publishing activity, coauthorship network may help identifying both, efficient research communities and also people, who lack proper collaborators while having poor research results. Our paper is devoted to the visualization and interpretation of the former sets using as an example co-authorship network of National Research University Higher School of Economics (HSE), Moscow, Russia, while we also discuss the possible solutions for recommending collaborators for the latter set of researchers with low academic profile. Our paper is a case study for our university, which can be extended to larger co-authorship networks using research indexing services.