In NIST speaker recognition evaluation 2016
This document presents the system submission for the group composed of MIT Lincoln Laboratory, Johns Hopkins University (JHU), Laboratoire de Recherche et de Développement de l’EPITA (LRDE) and Universidad Autónoma de Madrid (ATVS). The primary submission is a combination of four systems focused on i-vector systems. Two secondary submissions are also included
In
Digitization of the real world using real sensors has many drawbacks; in particular, we loose “well-composedness” in the sense that two digitized objects can be connected or not depending on the connectivity we choose in the digital image, leading then to ambiguities. Furthermore, digitized images are arrays of numerical values, and then do not own any topology by nature, contrary to our usual modeling of the real world in mathematics and in physics. Loosing all these properties makes difficult the development of algorithms which are “topologically correct” in image processing: e.g., the computation of the tree of shapes needs the representation of a given image to be continuous and well-composed; in the contrary case, we can obtain abnormalities in the final result. Some well-composed continuous representations already exist, but they are not in the same time $n$-dimensional and self-dual. In fact, $n$-dimensionality is crucial since usual signals are more and more 3-dimensional (like 2D videos) or 4-dimensional (like 4D Computerized Tomography-scans), and self-duality is necessary when a same image can contain different objects with different contrasts. We developed then a new way to make images well-composed by interpolation in a self-dual way and in $n$-D; followed with a span-based immersion, this interpolation becomes a self-dual continuous well-composed representation of the initial $n$-D signal. This representation benefits from many strong topological properties: it verifies the intermediate value theorem, the boundaries of any threshold set of the representation are disjoint union of discrete surfaces, and so on.
In Proceedings of the 13th international conference on latent variable analysis and signal separation (LVA-ICA)
Spectral Umixing (SU) in hyperspectral remote sensing aims at recovering the signatures of the pure materials in the scene (endmembers) and their abundances in each pixel of the image. The usual SU chain does not take spectral variability (SV) into account, and relies on the estimation of the Intrinsic Dimensionality (ID) of the data, related to the number of endmembers (NOE) to use. However, the ID can be significantly overestimated in difficult scenarios, and sometimes does not correspond to the desired scale and application dependent NOE. Spurious endmembers are then frequently extracted and included in the model. We propose an algorithm for SU incorporating SV, using collaborative sparsity to discard the least explicative endmembers in the whole image. We compute an algorithmic regularization path for this problem to select the optimal set of endmembers using a statistical criterion. Results on simulated and real data show the interest of the approach.
In
In this article we explain monads so they can be understood to the Lisp programmer. We base the explanation on a very clean explanation presented in the Scala programming language. We then proceed to re-present the concepts using mostly simple Common Lisp concepts. We do not attempt to justify the motivation behind the definitions, and we do not attempt to give any examples of applications. Most notably, we do not attempt to explain the connection monads have to modeling side effects.
In
Given a directed acyclic graph (DAG) and two arbitrary nodes, find maximal common joins of the two nodes. In this technical report I suggest an algorithm for efficiently calculating the minimal set of nodes which derive from a pair of nodes.
In Proceedings of the 2nd international workshop on robust reading conference (IWRR-ECCV)
Current text segmentation evaluation protocols are often incapable of properly handling different scenarios (broken/merged/partial characters). This leads to scores that incorrectly reflect the segmentation accuracy. In this article we propose a new evaluation scheme that overcomes most of the existent drawbacks by extending the EvaLTex protocol (initially designed to evaluate text detection at region level). This new unified platform has numerous advantages: it is able to evaluate a text understanding system at every detection stage and granularity level (paragraph/line/word and now character) by using the same metrics and matching rules; it is robust to all segmentation scenarios; it provides a qualitative and quantitative evaluation and a visual score representation that captures the whole behavior of a segmentation algorithm. Experimental results on nine segmentation algorithms using different evaluation frameworks are also provided to emphasize the interest of our method.
In Proceedings of the 23st international conference on pattern recognition (ICPR)
Many text segmentation methods are elaborate and thus are not suitable to real-time implementation on mobile devices. Having an efficient and effective method, robust to noise, blur, or uneven illumination, is interesting due to the increasing number of mobile applications needing text extraction. We propose a hierarchical image representation, based on the morphological Laplace operator, which is used to give a robust text segmentation. This representation relies on several very sound theoretical tools; its computation eventually translates to a simple labeling algorithm, and for text segmentation and grouping, to an easy tree-based processing. We also show that this method can also be applied to document binarization, with the interesting feature of getting also reverse-video text.
In Proceedings of the thirteenth international colloquium on theoretical aspects of computing (ICTAC)
We present an algorithm to build an automaton from a rational expression. This approach introduces support for extended weighted expressions. Inspired by derived-term based algorithms, its core relies on a different construct, rational expansions. We introduce an inductive algorithm to compute the expansion of an expression from which the automaton follows. This algorithm is independent of the size of the alphabet, and actually even supports infinite alphabets. It can easily be accommodated to generate deterministic (weighted) automata. These constructs are implemented in Vcsn, a free-software platform dedicated to weighted automata and rational expressions.
In Proceedings of the 14th international symposium on automated technology for verification and analysis (ATVA’16)
Checking liveness properties with partial-order reductions requires a cycle proviso to ensure that an action cannot be postponed forever. The proviso forces each cycle to contain at least one fully expanded state. We present new heuristics to select which state to expand, hoping to reduce the size of the resulting graph. The choice of the state to expand is done when encountering a dangerous edge. Almost all existing provisos expand the source of this edge, while this paper also explores the expansion of the destination and the use of SCC-based information.
In Proceedings of the 14th international symposium on automated technology for verification and analysis (ATVA’16)
We present Spot 2.0, a C++ library with Python bindings and an assortment of command-line tools designed to manipulate LTL and $\omega$-automata in batch. New automata-manipulation tools were introduced in Spot 2.0; they support arbitrary acceptance conditions, as expressible in the Hanoi Omega Automaton format. Besides being useful to researchers who have automata to process, its Python bindings can also be used in interactive environments to teach $\omega$-automata and model checking.
Copyright (c) 2022, LRE; all rights reserved.
Template by Bootstrapious. Ported to Hugo by DevCows.
