The big data approach to violin making

We are pleased to announce the online seminar “The big data approach to violin making” that will be held by Sebastian Gonzalez, a postdoc researcher of our lab.

The seminar will discuss how machine learning can help us understand the relation between the shape of a violin and its vibrational properties.

Join the seminar on Friday November 27th at 4.30pm CET in Sebastian’s Webex room.

Abstract

The shape of the violin is probably one of its most important characteristics, and the one upon which the violin maker has complete control. However, traditional violin making is based more in tradition than understanding, and a definitive scientific study of the specific relations between shape and vibrational behaviour is still missing. Using standard neural networks and big data methodologies, we show that the vibrational behaviour of violin tops can be predicted from its geometric/material parameters. The results demonstrate the relations between shape and vibrations and show the value that artificial intelligence methods have for classical violin making. Furthermore, we quantify correlations between vibrational modes and thicknesses (explaining the so-called `plate tuning’) in violin making, and shed light on their complexity at the same time that we present a predictive tool for said relation.

Biography

Sebastian Gonzalez, originally from Chile, obtained his PhD from UTwente in the Netherlands in 2013 in computational physics, where he did a thesis on clustering and segregation of granular materials. While finishing his PhD he met a family of violin makers in the north of Germany and started learning the craft. After 4 years learning he opened a guitar repair shop in Valparaiso, Chile, while doing a postdoc in the simulation of active colloids. He moved to Cremona shortly before the pandemic hit the world and he is still trying to finish his first violin, pictured here in the workbench.

Fifth Place in the FORCE Machine Learning competition

The FORCE Machine Learning competition has come to an end!

After reaching the 2nd place in the public prediction leaderbord, our team classified 5th in the final leaderboard on a private dataset.

Our approach is based on boosted trees with the particularity of training different models depending on the actual features present in the current test well. The final model is obtained by soft voting of the best models in validation.

The competition was very close and the results of the top 7 teams are almost equivalent. We are very happy with our result in both leaderboards. Taking part in this competition was a lot of fun and very informative. Thanks to Peter Bormann, Peder Aursand and all the organizers. And thanks to all the teams that made the competition so exciting and that have now shared their solutions. We are convinced that the competition has initiated many positive dynamics that will continue for a long time!

See the codes of the top-10 solutions on GitHub!

MAE Seminar – Italian Red Spruce and Traditional Mandolins

It is our pleasure to announce an interesting seminar on Musical Acoustics. The seminar will be offered by Martino Quintavalla, a luthier and researcher in materials engineering. The seminar will concern the choice of wood in musical instruments: Italian Red Spruce and traditional mandolins.

November 12, 2020, 2:30 pm. Approximate duration: 1.5 hr
Webex platform at the following link: https://politecnicomilano.webex.com/meet/fabio.antonacci

Abstract

Stringed musical instruments and mandolins in particular are part of worldwide culture, dating back to ancient times. During the last two centuries, mandolin had a great development and diffusion all over the world and italian mandolin tradition still keeps going on. In spite of this, scientific research on these instruments is still at an early stage.
In a research project founded by the World Wood Day Foundation in 2019, Martino Quintavalla together with Federico Gabrielli, mandolin luthier since more than 30 years and Claudio Canevari, teacher at the Civica Scuola di Liuteria di Milano and researcher, undertook the challenge to investigate some peculiar aspects of mandolin acoustics and resonance wood characterization. The Acoustic emission spectra of several mandolins were recorded, as well as their vibration modal shapes aiming at establishing a connection between instrument design and perceived sound quality. Likewise, several red spruce samples have been characterized to determine a way to design soundboards and choose those specimens allowing for the best acoustic performance. The project has ended with the construction of three experimental mandolins, nominally identical in the aesthetics but with completely different soundboards and sounds.

Biography

Martino Quintavalla received his PhD in materials engineering in 2016 at Politecnico di Milano and, since then, he dedicates his activity to the characterization and use of materials in several engineering fields as a post-doc researcher. In the last years, he has dedicated his research activity to musical instruments acoustics. He is author of several scientific publications and a professional luthier.

Strumenti e Metodi per la Valutazione e la Progettazione Acustica di Ambienti d’Ascolto

Giovedì 12 Novembre 2020
Ore 10.00 – 12.00

Abstract

Vengono presentate tecniche avanzate per la valutazione accurata del comportamento acustico di ambienti di ascolto e una breve panoramica delle tecniche esistenti o in corso di sviluppo per la simulazione predittiva del comportamento acustico dell’ambiente utilizzabili in fase di progetto. Infine vengono presentati due casi di studio di grande interesse e rilievo per la città di Cremona: l’Auditorium Arvedi, sul quale sono state applicate a posteriori alcune delle tecniche presentate; e il Palazzo Magio-Grasselli, nuova sede dell’Istituto Superiore di Studi Musicali “Claudio Monteverdi”, per il quale è in corso di progettazione un recupero acustico-funzionale in presenza di forti vincoli storico-architettonici.

Relatori

Augusto Sarti, docente del Politecnico di Milano,è fondatore e coordinatore scientifico del Musical Acoustics Lab e del Sound and Music Computing Lab del Politecnico di Milano. È fonda- tore e coordinatore della Laurea Magistrale in Music and Acou- stic Engineering presso il Politecnico di Milano. Ha promosso, coordinato, o contribuito a oltre 30 progetti europei nell’area dell’elaborazione di segnali multimediali. È co-autore di oltre 350 pubblicazioni scientifiche e brevetti. I suoi interessi di ricerca sono nell’area dell’elaborazione di segnali audio/acustici e acustica com- putazionale. È membro eletto dell’EURASIP Board of Directors, e IEEE Senior Member.

Fabio Antonacci, è ricercatore presso Politecnico di Milano, si oc- cupa di elaborazione di segnali acustici, con particolare riferimento a schiere di microfoni ed altoparlanti, modellazione della propa- gazione in ambienti chiusi. E’ co-autore di oltre 120 pubblicazioni su riviste scientifiche internazionali e articoli in atti di conferenze internazionali. Attualmente è membro dei comitati tecnici “Audio and Acoustics Signal Processing” dell’IEEE e “Acoustics, Speech and Music Signal Processing” dell’EURASIP e del comitato edito- riale dell’EURASIP.

Il seminario è gratuito ed è proposto con modalità on line sulla piat-taforma Teams. La procedura di iscrizione al seminario, disponibile a questo link è da completare entro e non oltre lunedì 9 Novembre. Solo chi sarà iscritto potrà partecipare.
La partecipazione al seminario darà diritto a 2 cfp agli Architetti iscritti all’Ordine.

Si comunica che il Corso di Formazione Permanente verrà erogato come “Evento riunione Teams”.
Informativa privacy sul sito www.polo-cremona.polimi.it

2nd place at Xeek FORCE Challenge on Machine Predicted Lithology

The FORCE Machine Predicted Lithology challenge has come to an end.

The ISPL team composed by Luca Bondi, Edoardo Daniele Cannas, Vincenzo Lipari and Francesco Picetti with the help of Paolo Bestagini has reached the second place in the Prediction leaderboard. Even working on a custom desk!
Let’s see what is going to happen in the final leaderboard on a different private dataset!

https://www.linkedin.com/posts/polimi-ispl_xeek-activity-6724648240710475776-cK8l

2nd place at FORCE challenge working remotely

IEEE Access “Article of the Week” on Camera Model Identification

The result of a fruitful collaboration between ISPL and Anderson Rocha‘s group at Universidade Estadual de Campinas is now IEEE Access “Article of the Week”[1].

If Interested in open-set camera model identification, go check it out!

[1] [doi] P. R. M. Júnior, L. Bondi, P. Bestagini, S. Tubaro, and A. Rocha, “An In-Depth Study on Open-Set Camera Model Identification,” IEEE Access, 2019.
[Bibtex]
@article{junior2019depth,
author = {P. R. M. {J\'unior} and L. {Bondi} and P. {Bestagini} and S. {Tubaro} and A. {Rocha}},
doi = {10.1109/ACCESS.2019.2921436},
groups = {forensics},
journal = {IEEE Access},
title = {An In-Depth Study on Open-Set Camera Model Identification},
year = {2019},
Bdsk-Url-1 = {https://doi.org/10.1109/ACCESS.2019.2921436}}

DeepFake Detection Challenge

DeepFake techniques enable realistic AI-generated videos of people doing and saying fictional things. These videos can be used maliciously as a source of misinformation, harassment and persuasion, thus badly impacting on the society. ISPL is actively working to fight this threat by developing state-of-the-art DeepFake detectors.

Check out our paper [1] accepted at ICPR 2020, which granted us the Top-2% ranking in the Facebook DeepFake Detection Challenge.

ispl deepfake
Face embedding projection

[1] Unknown bibtex entry with key [bonettini2020video]
[Bibtex]

ISPL involved in PREMIER project

Group photo of the kick-off meeting of PREMIER project
PREMIER kick-off meeting took place in Cima Sappada, Italy, in conjunction with the GTTI MMSP annual thematic meeting.

PREserving Media trustworthiness in the artificial Intelligence ERa

ISPL is involved in PREMIER project, funded by the Italian Ministry of Education, University, and Research.

The objective of PREMIER is to devise new techniques capable of distinguishing fake from original videos. The project will develop detectors based on deep learning, overcoming the main shortcomings of existing methodologies.

PREMIER project will contribute to boost video forensics technology and provide citizens with advanced publicly available tools to assess the authenticity of their own videos.

Check out the official web page for news and updates on the project!