Публикации

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.

Let X_1, ... ,X_n be i.i.d. sample in R^p with zero mean and the covariance matrix S. The problem of recovering the projector onto the eigenspace of S from these observations naturally arises in many applications. Recent technique from [Koltchinskii and Lounici, 2015b] helps to study the asymptotic distribution of the distance in the Frobenius norm between the true projector P_r on the subspace of the r-th eigenvalue and its empirical counterpart \hat{P}_r in terms of the effective trace of S. This paper offers a bootstrap procedure for building sharp confidence sets for the true projector P_r from the given data. This procedure does not rely on the asymptotic distribution of || P_r - \hat{P}_r ||_2 and its moments, it applies for small or moderate sample size n and large dimension p. The main result states the validity of the proposed procedure for finite samples with an explicit error bound on the error of bootstrap approximation. This bound involves some new sharp results on Gaussian comparison and Gaussian anti-concentration in high dimension. Numeric results confirm a nice performance of the method in realistic examples. 

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.