[iva] 1 PhD position in affective computing at the Grenoble Alps University

[Fabien Ringeval]

1 PhD position in affective computing at the Grenoble Alps University

ATOS France and the Grenoble Informatics Laboratory (LIG) invite applications for a fully funded PhD position on "Weakly-supervised learning of human affective behaviors from multimodal interactions with a chatbot". The PhD will be co-supervised by Jean-Phillippe Vigne (ATOS) and Béatrice Bouchot (ATOS), Pr. Laurent Besacier (LIG) and Dr. Fabien Ringeval (LIG).


Thesis description
==================
The thesis targets three main objectives: 
1) the development of a weakly-supervised learning methodology for the semi-automatic annotation of affective information from speech and text produced by humans while interacting with a chatbot
2) the development of a module that performs a robust fusion of inputs’ representations (speech + text) in order to infer attributes of affect in varying noisy conditions
3) an evaluation of the system’s robustness in different contexts of interaction with the chat-bot.

Recent advances in deep learning have shown promising results in many applications of affective computing [Picard-95], where ones of the most dominant tasks consist in quantifying attributes of human emotion, such as arousal, valence, or dominance [Russel-80], time-continuously from signals recorded by sensors [Wöllmer-08]. Long Short-Term Memory Recurrent Neural Networks (LSTM-RNN) [Gers-99] have successfully been employed to model long-range contextual dependencies between attributes of affect and speech data [Eyben-10, Eyben-12, Ringeval-15], and convolutional neural networks (CNNs) have shown promising results for learning useful information from the raw signals when combined with LSTM-RNN in the so-called “end-to-end” framework [Trigeorgis-16]. Recently, semi-supervised [Schmitt-16, Ghosh-16] and unsupervised [Cummins-18] methods of representation learning have shown the interest of exploiting resources from other domains in order to deal with the issue of data scarcity, which is of paramount importance for methods based on deep learning, as they need as many examples as possible to generalise well on expressions of affect produced ‘in-the-wild’ [Ringeval-18].

In this thesis, weakly-supervised methods based on deep learning will be exploited to perform semi-automatic annotation of human affective behaviour from speech and text – either typed on a keyboard, or automatically retrieved from speech by an ASR. Context-aware novelty detection [Marki-15] based on deep LSTM auto-encoders will be used to detect novel affective content, and semi-supervised learning methods [Zhang-18] will be employed to enrich the model while following a curriculum learning [Lotfian-18]. Data exploited to build and evaluate the system will rely on the data collected during the project but also on existing publicly available datasets of emotion, including people with various culture, language, age, education, but also featuring different environments and contexts of interaction. Data automatically retrieved from social platforms like YouTube channels will be considered for automatically enriching the model in a ‘virtuous circle’ fashion.

The envisioned starting date is December 2018.


Requirements
============
We are looking for one candidate with a strong focus on deep learning for affective computing with the following profile:
+ Master’s degree with background in Machine Learning, Speech Processing, Affective Computing                 
+ Excellent programming skills (Python, Java, C/C++), knowledge of Keras/TensorFlow/Torch would be ideal
+ Ability to work independently and be self-motivated         
+ Excellent communication skills in English           
 

Applying
========
To apply, please email your application to: fabien.ringeval at imag.fr <mailto:fabien.ringeval at imag.fr>, laurent.besacier at imag.fr <mailto:laurent.besacier at imag.fr>, jean-philippe.vigne at atos.net <mailto:jean-philippe.vigne at atos.net> and beatrice.bouchot at atos.net <mailto:beatrice.bouchot at atos.net>.
 
The application should consist of a single pdf file including:                  
+ a curriculum vitae showing academic records with tracks related to the themes of the thesis               
+ transcript of marks according to M1-M2 profile or last 3 years of engineering school      
+ statement letter expressing your interest in the position and your profile relevance             
+ contact and recommendation letter of at least one university referent                   
 
Incomplete applications will not be processed. Potential candidates will be invited for an interview with the supervisors.
 

Conditions of employment     
========================
You will be hired on a fixed-term contract (3 years contract – CIFRE) at ATOS, a global leader in digital transformation.
 

Working at Grenoble (ATOS/LIG)       
==================================
You will be integrated in two teams with academic and industrial profiles: the GETALP team of the LIG, recognised for its research activities in the fields of speech and language processing, and the team Cognitive Intelligence from ATOS, who is specialised in Artificial Intelligence (AI) for the development of chatbots.

ATOS is a leader in digital services with pro forma annual revenue of circa € 12 billion and circa 100,000 employees in 73 countries, serving a global client base. ATOS R&D team has a very active innovation spirit backed by a culture of Intellectual Property. Together these have led to numerous disruptive developments, including more than 1,500 patents. ATOS Grenoble (1000 collaborators) is focusing on AI, working with a variety of clients to implement solutions where they create value. ATOS leadership in Cloud technology, Cybersecurity and High-performance computing, along with our partnerships with major AI companies (e.g., Google), help us provide clients with the resources, expertise and support they need.

The LIG is one of the largest laboratories in Computer Science in France. It is structured as a Joint Research Center (Unité Mixte de Recherche) founded by the CNRS, the Grenoble Institute of Technology (Grenoble INP), the INRIA Grenoble Rhône-Alpes, and the Grenoble Alps University (UGA), which has recently been ranked as France’s number one university in eleven disciplines, including Computer Science & Engineering, in the latest Shanghai Academic Subject Rankings of World Universities 2017. The LIG hosts 17 research teams and three teams providing administrative and technical supports, which represent an overall of 450 collaborators including 205 permanent researchers, 143 PhD students, and 35 persons in supporting teams as identified in 2016.

The city of Grenoble is located on a plateau at the foot of the French Alps and is advertised to be the “Capital of the Alps” due to its immediate proximity to the mountains. The IMAG building hosting the LIG is located on a landscaped campus of 175 hectares, which straddles Saint-Martin-d’Hères and Gières, and welcome around 40,000 students and researchers working in various research institutions. Thanks to this campus, UGA has been ranked as the eighth most beautiful universities in Europe by the Times Higher Education magazine in 2018. Overall, Grenoble as a city is the largest research center in France after Paris with 22,800 researchers.     
  

References
==========

[Cummins-18]         Nicholas Cummins, Shahin Amiriparian, Gerhard Hagerer, Anton Batliner, Stefan Steidl, and Björn Schuller, An image-based deep spectrum feature representation for the recognition of emotional speech, in Proceedings of ACM MM 2017, pp. 478–484, October 2017, ACM.
[Eyben-10]              Florian Eyben, Martin Wöllmer, Alex Graves, Björn Schuller, Ellen Douglas-Cowie and Roddy Cowie, On-line emotion recognition in a 3-D activation-valence-time continuum using acoustic and linguistic cues, Journal on Multimodal User Interfaces 3(1-2):7–19, March 2010, Springer Nature.
[Eyben-12]              Florian Eyben, Martin Wöllmer, and Björn Schuller, A multi-task approach to continuous five-dimensional affect sensing in natural speech, ACM Transactions on Interactive Intelligent Systems (TiiS) - Special Issue on Affective Interaction in Natural Environments 2(1):6, March 2012, ACM.
[Gers-99]                  Felix A. Gers, Jürgen Schmidhuber, and Fred Cummins, Learning to forget: Continual prediction with LSTM, in Proceedings of ICANN 1999, pp. 850–855, ENNS.
[Ghosh-16]              Sayan Ghosh, Eugene Laksana, Louis-Philippe Morency and Stefan Scherer, Representation learning for speech emotion recognition, in Proceedings of Interspeech 2016, pp. 3603–3607, September 2016, ISCA.
[Lotfian-18]             Reza Lotfian and Carlos Busso, Curriculum learning for speech emotion recognition from crowdsourced labels, arXiv:1805.10339, May 2018.
[Marki-15]                Erik Sayan Ghosh, Eugene Laksana, Louis-Philippe Morency and Stefan Scherer, Representation learning for speech emotion recognition, in Proceedings of Interspeech 2016, pp. 3603–3607, September 2016, ISCA.
[Picard-95]               Rosalind W. Picard, Affective Computing, MIT Press.
[Ringeval-15]          Fabien Ringeval, Florian Eyben, Eleni Kroupi, Anil Yuce, Jean-Philippe Thiran, Touradj Ebrahimi, Denis Lalanne, and Björn Schuller, Prediction of asynchronous dimensional emotion ratings from audiovisual and physiological data, Pattern Recognition Letters, 66:25–30, November 2015, Elsevier.
[Ringeval-18]          Fabien Ringeval, Björn Schuller, Michel Valstar, Roddy Cowie, Heysem Kaya, Maximilian Schmitt, Shahin Amiriparian, Nicholas Cummins, Denis Lalanne, Adrien Michaud, Elvan Ciftçi, Hüseyin Güleç, Albert Ali Salah, and Maja Pantic, AVEC 2018 Workshop and challenge: Bipolar disorder and cross-cultural affect recognition, in Proceedings of AVEC’18, ACM MM, October 2018, ACM.
[Russel-80]               James A. Russel, A circumplex model of affect, Journal of personality and social psychology, 39(6):1161–1178, December 1980, APA.
[Schimtt-16]             Maximilian Schmitt, Fabien Ringeval, and Björn Schuller, At the border of acoustics and linguistics: Bag-of-Audio-Words for the recognition of emotions in speech. In Proceedings Interspeech 2016, pp. 495–499, San Fransisco (CA), USA, September 2016, ISCA.
[Trigeorgis-16]        George Trigeorgis, Fabien Ringeval, Raymond Brueckner, Erik Marchi, Mihalis Nicolaou, Björn Schuller and Stefanos Zafeiriou, Adieu features? End-to-end speech emotion recognition using a deep convolutional recurrent network, in Proceedings ICASSP 2016, pp. 5200–5204, Shanghai, China, April 2016, IEEE.
[Wöllmer-08]           Martin Wöllmer, Florian Eyben, Stephan Reiter, Björn Schuller, Cate Cox, Ellen Douglas-Cowie, and Roddy Cowie, Abandoning emotion classes-towards continuous emotion recognition with modelling of long-range dependencies, in Proceedings of Interspeech 2008, Brisbane, Australia, pp. 597–600, ISCA.
[Zhang-18]              Zixing Zhang, Jin Han, Jun Deng, Xinzhou Xu, Fabien Ringeval, and Björn Schuller, Leveraging unlabelled data for emotion recognition with enhanced collaborative semi-supervised learning, IEEE Access, 6, April 2018, IEEE.

 

 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.uni-bielefeld.de/mailman2/unibi/public/iva-list/attachments/20180817/e5d52266/attachment-0001.html>