„Brillant, organisé et efficace. Il a un large spectre de connaissances théoriques et une grande aisance à développer des solutions en Data Science que ça soit en mode POC ou en mode industriel (Big Data). Ajouter à cela, un relationnel naturel et un sourire permanent. Vous avez là un Data Scientist qui fera gagner votre équipe !“
Lausanne, Waadt, Schweiz
Kontaktinformationen
7628 Follower:innen
500+ Kontakte
Info
Aktivitäten
-
It was a great pleasure to share insights about KOVE medical Ltd vision to help surgeons dramatically increase their success rate by reducing risks…
It was a great pleasure to share insights about KOVE medical Ltd vision to help surgeons dramatically increase their success rate by reducing risks…
Beliebt bei Maël F.
Berufserfahrung
Ausbildung
Veröffentlichungen
-
Open-Set Speaker Identification pipeline in live criminal investigations
Speaker recognition has many applications in conversational data, including in forensic science where Law Enforcement Agencies (LEAs) aim to assess the identity of a speaker on a specific recorded telephone call. However, speaker identification (SID) systems require initial enrollment data, whereas LEAs might start a case with text or video evidence, and few to no enrollment data. In this paper, we introduce the ROXANNE simulated dataset, a multilingual corpus of acted telephone calls following…
Speaker recognition has many applications in conversational data, including in forensic science where Law Enforcement Agencies (LEAs) aim to assess the identity of a speaker on a specific recorded telephone call. However, speaker identification (SID) systems require initial enrollment data, whereas LEAs might start a case with text or video evidence, and few to no enrollment data. In this paper, we introduce the ROXANNE simulated dataset, a multilingual corpus of acted telephone calls following a screenplay prepared by LEAs. We also present a process to build criminal networks from SID, by addressing practical constraints of these investigations. Our process reaches a speaker accuracy of 92.4% on the simulated data and a conversation accuracy of 84.9%. We finally offer some future directions for this work.
-
ROXSD: a Simulated Dataset of Communication in Organized Crime
Criminal investigations contain sensitive and confidential material and are nonpublic by nature. Access to investigation data is very limited and restricted to only selected groups of individuals. Even for research purposes, data typically cannot be accessed freely. Within criminal investigations, data is still processed manually to a large extent. Solutions provided for automation of this processing—or even of individual processing steps—can be assumed to have a significant impact on the work…
Criminal investigations contain sensitive and confidential material and are nonpublic by nature. Access to investigation data is very limited and restricted to only selected groups of individuals. Even for research purposes, data typically cannot be accessed freely. Within criminal investigations, data is still processed manually to a large extent. Solutions provided for automation of this processing—or even of individual processing steps—can be assumed to have a significant impact on the work of Law Enforcement Agencies (LEAs). Automation may effectively be key to handle large and complex amounts of data in an efficient manner under the typical operating conditions of LEAs.
-
ROXANNE Research Platform: Automate Criminal Investigations
Criminal investigations require manual intervention of several investigators and translators. However, the amount and the diversity of the data collected raises many challenges, and crossborder investigations against organized crime can quickly impossible to handle. We developed ROXANNE Research platform, an all-in-one platform which processes intercepted phone calls, runs state-of-the-art components such as speaker identification, automatic speech recognition or named entity detection, and…
Criminal investigations require manual intervention of several investigators and translators. However, the amount and the diversity of the data collected raises many challenges, and crossborder investigations against organized crime can quickly impossible to handle. We developed ROXANNE Research platform, an all-in-one platform which processes intercepted phone calls, runs state-of-the-art components such as speaker identification, automatic speech recognition or named entity detection, and builds a knowledge graph of the extracted information. Our aim for this work is to do a first step in the direction of an open research platform combining speech, text, and video processing algorithms with criminal network analysis for combating organized crime.
-
Improving Speaker Identification using Network Knowledge in Criminal Conversational Data
Criminal investigations rely on the collection of conversational data. The identity of speakers must be assessed in order to build or improve the accuracy of an existing criminal network. Investigators use social network analysis tools to identify the most central character and the different communities within the network. We introduce Crime Scene Investigation (CSI) television show as a potential candidate for criminal conversational data. We also introduce the metric of conversation accuracy…
Criminal investigations rely on the collection of conversational data. The identity of speakers must be assessed in order to build or improve the accuracy of an existing criminal network. Investigators use social network analysis tools to identify the most central character and the different communities within the network. We introduce Crime Scene Investigation (CSI) television show as a potential candidate for criminal conversational data. We also introduce the metric of conversation accuracy in the context of criminal investigations. In this paper, a speaker identification baseline is improved by re-ranking candidate speakers based on the frequency of previous interactions between speakers and the topology of the criminal network. The proposed method can be applied to conversations involving two or more speakers. We show that our approach outperforms the baseline speaker accuracy by 1.3% absolute (1.5% relative), and the conversation accuracy by 3.7% absolute (4.7% relative) on CSI data.
-
BertAA: BERT fine-tuning for Authorship Attribution
ICON2020
Identifying the author of a given text can be useful in historical literature, plagiarism detection, or police investigations. Authorship Attribution (AA) has been well studied and
mostly relies on a large feature engineering work. More recently, deep learning-based approaches have been explored for Authorship Attribution (AA). In this paper, we introduce BertAA, a fine-tuning of a pre-trained BERT language model with an additional dense layer and a softmax activation to perform authorship…Identifying the author of a given text can be useful in historical literature, plagiarism detection, or police investigations. Authorship Attribution (AA) has been well studied and
mostly relies on a large feature engineering work. More recently, deep learning-based approaches have been explored for Authorship Attribution (AA). In this paper, we introduce BertAA, a fine-tuning of a pre-trained BERT language model with an additional dense layer and a softmax activation to perform authorship classification. This approach reaches competitive performances on Enron Email, Blog Authorship, and IMDb (and IMDb62) datasets, up to 5.3% (relative) above current state-of-the-art approaches. We performed an exhaustive analysis allowing us to identify the strengths and weaknesses of the proposed method. In addition, we evaluate the impact of including additional features (e.g. stylometric and hybrid features) in an ensemble approach, improving the macro-averaged F1-Score by 2.7% (relative) on average. -
Detection of Similar Languages and Dialects Using Deep Supervised Autoencoder
Proceedings of the 17th International Conference on Natural Language Processing
Language detection is considered a difficult task especially for similar languages, varieties, and dialects. With the growing number of online content in different languages, the need for reliable and robust language detection tools also increased. In this work, we use supervised autoencoders with a bayesian optimizer for language detection and highlights its efficiency in detecting similar languages with dialect variance in comparison to other state-of-the-art techniques. We evaluated our…
Language detection is considered a difficult task especially for similar languages, varieties, and dialects. With the growing number of online content in different languages, the need for reliable and robust language detection tools also increased. In this work, we use supervised autoencoders with a bayesian optimizer for language detection and highlights its efficiency in detecting similar languages with dialect variance in comparison to other state-of-the-art techniques. We evaluated our approach on multiple datasets (Ling10, Discriminating between Similar Language (DSL), and Indo-Aryan Language Identification (ILI)). Obtained results demonstrate that SAE are higly effective in detecting languages, up to a 100% accuracy in the Ling10. Similarly, we obtain a competitive performance in identifying similar languages, and dialects, 92% and 85% for DSL ans ILI datasets respectively.
-
Graph2Speak: Improving Speaker Identification using Network Knowledge in Criminal Conversational Data
Criminal investigations mostly rely on the collection of speech conversational data in order to identify speakers and build or enrich an existing criminal network. Social network analysis tools are then applied to identify the most central characters and the different communities within the network. We introduce two candidate datasets for criminal conversational data, Crime Scene Investigation (CSI), a television show, and the ROXANNE simulated data. We also introduce the metric of conversation…
Criminal investigations mostly rely on the collection of speech conversational data in order to identify speakers and build or enrich an existing criminal network. Social network analysis tools are then applied to identify the most central characters and the different communities within the network. We introduce two candidate datasets for criminal conversational data, Crime Scene Investigation (CSI), a television show, and the ROXANNE simulated data. We also introduce the metric of conversation accuracy in the context of criminal investigations. By re-ranking candidate speakers based on the frequency of previous interactions, we improve the speaker identification baseline by 1.2% absolute (1.3% relative), and the conversation accuracy by 2.6% absolute (3.4% relative) on CSI data, and by 1.1% absolute (1.2% relative), and 2% absolute (2.5% relative) respectively on the ROXANNE simulated data.
Auszeichnungen/Preise
-
Dalle Molle Label for Quality of Life
Dalle Molle Foundation
-
Prize “Canne Blanche”
SZBlind
Innovation of the year for blind and visually impaired people in Switzerland
-
W.A. de Vigier Stiftung Winner
W.A. de Vigier Stiftung
Grant of 100’000 CHF to support the development of biped.ai
-
Future of Health Grant
Future of Health
Grant of 30'000 CHF to support the development of biped.ai
-
Biopole Campus Fund
Biopole
Grant of 30'000 CHF to support the development of biped.ai
-
FIT Digital grant
FIT
Grant of 20'000 CHF to support the development of biped.ai
-
VentureKick Winner
VentureKick
Passed all rounds of VentureKick start-up contest for biped, 150'000 CHF
-
Winner of the International Create Challenge
International Create Challenge
Won the International Create Challenge, an AI-based competition, with SoundMap, an assistive device for blind and visually impaired people. We won both the 1st prize and the special healthcare award. SoundMap is a smart wearable belt, equipped with a camera, able to provide real-time information on the surrounding environment of a person through Audio Augmented Reality (Audio AR).
Similarly to the way we easily identify the position of whistling birds, we aim to scan in real-time the…Won the International Create Challenge, an AI-based competition, with SoundMap, an assistive device for blind and visually impaired people. We won both the 1st prize and the special healthcare award. SoundMap is a smart wearable belt, equipped with a camera, able to provide real-time information on the surrounding environment of a person through Audio Augmented Reality (Audio AR).
Similarly to the way we easily identify the position of whistling birds, we aim to scan in real-time the surrounding environment and produce directional sounds (through audio AR) in the earphones connected to the device. Blind and visually impaired people are, therefore, able to map and understand their environment. -
Winner of the Hackathon "Hubs as a Service"
Grand Paris Seine & Oise
Over a weekend, we developed a system of mobile medical care to fight medical deserts. The project has been awarded the first prize among many strong teams and we had the chance to pitch our project in front of officials and local citizens, with 5'000 CHF prize.
-
Jury Prize Winner of the Start Lausanne Contest 2016
Start Lausanne
I took part in the entrepreneurial contest Start Lausanne, and went through the different steps : Elevator Pitch, Business Model and Business Plan. The whole contest lead me to the final, in which I won the Jury Prize as well as 5'000 CHF to launch Wanago.
-
3x Regional Javelin Champion
French Athletics Federation
Junior Javelin Champion of the Rhône-Alpes region. I have since then kept practicing regularly javelin.
-
Volleyball Swiss National Champion
Swiss Volleyball Federation
Captain and team player of the team regional junior team that became Swiss National Champion.
Sprachen
-
Français
Muttersprache oder zweisprachig
-
Anglais
Muttersprache oder zweisprachig
-
Allemand
Fließend
Erhaltene Empfehlungen
3 Personen haben Maël F. empfohlen
Jetzt anmelden und ansehenWeitere Aktivitäten von Maël F.
-
NOA by biped robotics marks successful market entry in Europe and the US 🎉 🙌 Navigation, Obstacle, Artificial intelligence (NOA) is the new brand…
NOA by biped robotics marks successful market entry in Europe and the US 🎉 🙌 Navigation, Obstacle, Artificial intelligence (NOA) is the new brand…
Beliebt bei Maël F.
-
VdTech rocks : Lausanne-based biped robotics specializes in innovative AI-powered navigation devices designed to enhance mobility for blind and…
VdTech rocks : Lausanne-based biped robotics specializes in innovative AI-powered navigation devices designed to enhance mobility for blind and…
Beliebt bei Maël F.
-
Couldn’t have said it better myself - oh wait…. Alt text provided by AI
Couldn’t have said it better myself - oh wait…. Alt text provided by AI
Beliebt bei Maël F.
-
In August 2012 I had my #PhD defense talk. My topic was stereo image depth estimation with low compute resources. I used Curiosity, which has just…
In August 2012 I had my #PhD defense talk. My topic was stereo image depth estimation with low compute resources. I used Curiosity, which has just…
Beliebt bei Maël F.
-
My company @podsqueeze_app makes close to 200K ARR yet I am still making less than my salary as a developer in 2020! 🤔 Or am I? If we only…
My company @podsqueeze_app makes close to 200K ARR yet I am still making less than my salary as a developer in 2020! 🤔 Or am I? If we only…
Beliebt bei Maël F.
-
Come visit the biped booth at AER International Conference for a glance to the future of O&M powered by AI.
Come visit the biped booth at AER International Conference for a glance to the future of O&M powered by AI.
Beliebt bei Maël F.
Weitere ähnliche Profile
Weitere Mitglieder, die Maël F. heißen
Es gibt auf LinkedIn 1 weitere Person, die Maël F. heißt.
Weitere Mitglieder anzeigen, die Maël F. heißen