Nicolas Aspert

Related publications (24)

Learnable filter-banks for CNN-based audio applications

Nicolas Aspert, Laurent Colbois, Helena Peic Tukuljac, Benjamin Ricaud

We investigate the design of a convolutional layer where kernels are parameterized functions. This layer aims at being the input layer of convolutional neural networks for audio applications or applications involving time-series. The kernels are defined as one-dimensional functions having a band-pass filter shape, with a limited number of trainable parameters. Building on the literature on this topic, we confirm that networks having such an input layer can achieve state-of-the-art accuracy on several audio classification tasks. We explore the effect of different parameters on the network accuracy and learning ability. This approach reduces the number of weights to be trained and enables larger kernel sizes, an advantage for audio applications. Furthermore, the learned filters bring additional interpretability and a better understanding of the audio properties exploited by the network.

2022

, ,

This article provides feedback on a multidisciplinary and applied research project to design a data mapping and exploration tool on Twitter for journalists. The approach focuses on the observation and understanding of journalists' uses and a projection of the integration of the software in their daily practices and routines. Three use cases scenarios are explored, reflecting the diversity of the tool's appropriation. Each of them underlines the interest in designing a transparent interface that favors the user's understanding of the operations performed on the data, but also the need for users to have a minimal digital culture to produce journalistic content based on these digital traces.

2021

Spikyball Sampling: Exploring Large Networks via an Inhomogeneous Filtered Diffusion

Nicolas Aspert, Volodymyr Miz, Benjamin Ricaud

Studying real-world networks such as social networks or web networks is a challenge. These networks often combine a complex, highly connected structure together with a large size. We propose a new approach for large scale networks that is able to automatically sample user-defined relevant parts of a network. Starting from a few selected places in the network and a reduced set of expansion rules, the method adopts a filtered breadth-first search approach, that expands through edges and nodes matching these properties. Moreover, the expansion is performed over a random subset of neighbors at each step to mitigate further the overwhelming number of connections that may exist in large graphs. This carries the image of a “spiky” expansion. We show that this approach generalize previous exploration sampling methods, such as Snowball or Forest Fire and extend them. We demonstrate its ability to capture groups of nodes with high interactions while discarding weakly connected nodes that are often numerous in social networks and may hide important structures.

2020

What is Trending on Wikipedia? Capturing Trends and Language Biases Across Wikipedia Editions

Nicolas Aspert, Joëlle Hanna, Volodymyr Miz, Benjamin Ricaud, Pierre Vandergheynst

In this work, we propose an automatic evaluation and comparison of the browsing behavior of Wikipedia readers that can be applied to any language editions of Wikipedia. As an example, we focus on English, French, and Russian languages during the last four months of 2018. The proposed method has three steps. Firstly, it extracts the most trending articles over a chosen period of time. Secondly, it performs a semi-supervised topic extraction and thirdly, it compares topics across languages. The automated processing works with the data that combines Wikipedia's graph of hyperlinks, pageview statistics and summaries of the pages. The results show that people share a common interest and curiosity for entertainment, e.g. movies, music, sports independently of their language. Differences appear in topics related to local events or about cultural particularities. Interactive visualizations showing clusters of trending pages in each language edition are available online https://wiki-insights.epfl.ch/wikitrends

ACM2020

, , ,

Wikipedia is a rich and invaluable source of information. Its central place on the Web makes it a particularly interesting object of study for scientists. Researchers from different domains used various complex datasets related to Wikipedia to study language, social behavior, knowledge organization, and network theory. While being a scientific treasure, the large size of the dataset hinders pre-processing and may be a challenging obstacle for potential new studies. This issue is particularly acute in scientific domains where researchers may not be technically and data processing savvy. On one hand, the size of Wikipedia dumps is large. It makes the parsing and extraction of relevant information cumbersome. On the other hand, the API is straightforward to use but restricted to a relatively small number of requests. The middle ground is at the mesoscopic scale when researchers need a subset of Wikipedia ranging from thousands to hundreds of thousands of pages but there exists no efficient solution at this scale. In this work, we propose an efficient data structure to make requests and access subnetworks of Wikipedia pages and categories. We provide convenient tools for accessing and filtering viewership statistics or "pagecounts" of Wikipedia web pages. The dataset organization leverages principles of graph databases that allows rapid and intuitive access to subgraphs of Wikipedia articles and categories. The dataset and deployment guidelines are available on the LTS2 website \url{https://lts2.epfl.ch/Datasets/Wikipedia/}.

ACM2019

Process engineering and failure analysis of MEMS and MOEMS by Digital Holography Microscopy (DHM)

Nicolas Aspert, Etienne Cuche, Christian Depeursinge, Mihai Adrian Ionescu, Alexander Mehdaoui, Frédéric Montfort

Process engineering and failure analysis of MEMS and MOEMS require static and dynamical characterization of both their in-plane and out of plane response to an excitation. A remarkable characteristic of Digital Holography Microscopes (DHM) is the extremely short acquisition time required to grab the whole information necessary to provide 3D optical topography of the sample: a unique frame grab, without any vertical or lateral scan provides the information over the full field of view. First, it ensures DHM measurements to be insensitive to vibrations. Second, it opens the door to fast dynamical characterization of micro-systems. For periodic movement analysis, DHM can operate in stroboscopic mode with standard cameras. It enables precise characterization up to excitation frequencies of 100 kHz with recovery cycle of 10% simply by triggering properly the camera. Pulsed sources can be used for investigation of higher excitation frequencies. For non periodic movement analysis fast acquisition cameras and postponed treatment are used. DHM are therefore unique and very efficient tool for dynamical characterization of in- plane and out-of-plane response. In this paper we show the basics of the technology and illustrate process engineering and failure analysis using DHM with an example of in and out of plane characterization of movements of a variable capacitor using the stroboscopic mode of acquisition.

SPIE2007

, , , , , , ,

We present a procedure that compensates for phase aberrations in digital holographic microscopy by computing a polynomial phase mask directly from the hologram. The phase-mask parameters are computed automatically without knowledge of physical values such as wave vectors, focal lengths, or distances. This method enables one to reconstruct correct and accurate phase distributions, even in the presence of strong and high-order aberrations. Examples of applications are shown for microlens imaging and for compensating for the deformations associated with a tilted thick plate. Finally we show that this method allows compensation for the curvature of the specimen, revealing its surface defects and roughness. Examples of applications are shown for microlenses and metallic sphere imaging.

2006

Automatic procedure for aberrations compensation in digital holographic microscopy

Nicolas Aspert, Florian Charrière, Tristan Colomb, Etienne Cuche, Christian Depeursinge, Jonas Kühn, Mihaela Marian, Pierre Marquet, Frédéric Montfort

Digital Holographic Microscopy (DHM) is a powerful imaging technique allowing, from a single amplitude image (amplitude and phase), reflected acquisition (hologram), the reconstruction of the entire complex wave front by or transmitted through an object. Because holography is an interferometric technique, the reconstructed phase leads to a sub-wavelength axial accuracy (below lambda/100). Nevertheless, this accuracy is difficult to obtain from a single hologram. Indeed, the reconstruction process consisting to process the hologram with a digital reference wave (similar to classical holographic reconstruction) seems to need a-priori knowledge about the physical values of the setup. Furthermore, the introduction of a microscope objective (MO), used to improve the lateral resolution, introduces a wave front curvature in the object wave front. Finally, the optics of the set-up can introduce different aberrations that decrease the quality and the accuracy of the phase images. We propose here an automatic procedure allowing the adjustment of the physical values and the compensation for the phase aberrations. The method is based on the extraction of reconstructed phase values, along line profiles, located on or around the sample, in assumed to be flat area, and which serve as reference surfaces. The phase reconstruction parameters are then automatically adjusted by applying curve-fitting procedures on the extracted phase profiles. An example of a mirror and a USAF test target recorded with high order aberrations (introduced by a thick tilted plate placed in the set- up) shows that our procedure reduces the phase standard deviation from 45 degrees to 5 degrees.

SPIE2006

Digital Holographic Microscopy (DHM) for metrology and dynamic characterization of MEMS and MOEMS

Nicolas Aspert, Florian Charrière, Tristan Colomb, Etienne Cuche, Christian Depeursinge, Jonas Kühn, Pierre Marquet, Frédéric Montfort

Digital Holographic Microscopes (DHM) enables recording the whole information necessary to provide real time nanometric vertical displacement measurements with a single image acquisition. The use of fast acquisition camera or stroboscopic acquisition mode makes these new systems ideal tools for investigating the topography and dynamical behavior of MEMS and MOEMS. This is illustrated by the investigation of resonant frequencies of a dual axis micromirror. This enables the definition of the linear, non-linear, and modal resonance zones of its dynamical response.

SPIE2006

Digital Holographic Microscopy (DHM): Fast and robust 3D measurements with interferometric resolution for industrial inspection

Nicolas Aspert, Stéphane Bourquin, Etienne Cuche, Christian Depeursinge, Jonas Kühn, Pierre Marquet

The use of numerical procedure at a level never reach so far in optical imaging enables DHM to overcome two cumbersome alignment procedures of interferometric microscopes: (1) fine focus of the sample, and (2) sample tilt adjustment. In addition the technology enables to retrieve the full information from a single hologram acquisition. This enables very short acquisition time, smaller than typical propagation frequencies of ambient vibrations. These instruments can be operated without vibration insulation tables. Combined with the simplicity of the opto-mechanical design and the absence of moving parts, DHM are new measurement systems with a nanometer scale resolution ready to take place in production lines. DHM can be used to characterized shapes and surface roughness of a large variety of samples and their use for industrial inspection should definitely spread during the next years

Springer-Verlag2006