Keywords
===
Multimodal merging, active perception, developmental learning, social robotics, psychophysics.
Subject
===
In the AMPLIFIER (Active Multisensory Perception & LearnIng For InteractivE Robots) project, we study multimodal perception in humans with psychophysics experiments and modeling to improve human robot interactions for social robots. We adopt a constructivist and sensori-motor approach which considers the perception as the result of mastering sensori-motor contingencies that are learned incrementally all life long [6, 7, 8]. We will study multiple research questions: 1- How and when fusing data coming from various sensors? 2- How this merging process evolves with age? 3- How active perception (i.e. active sampling of data in the environment, e.g. by moving the eyes to sample visual data) influences the multimodal perception (and reciprocally)? This project (2018-2022) involves members from Lyon 1 University (LIRIS, CRNL), Univ. Grenoble Alpes (LJK, Gipsa-lab, LPNC) and Hoomano, a start-up located in Lyon.
Multi-sensory merging is a key feature to generate a consistent perception of the environment [9]. Moreover, it allows better detection and discrimination of stimuli [11] and has been argued to be Bayes optimal [2]. In machine learning, multimodal merging is often performed with classical methods developed for computer vision, natural language processing, ... Depending on the model, the fusion is done at dierent stages of the processing chain [1]. However, the impact of the fusion is not well understood and usually is task dependent.
Instead of simply considering actions performed to achieve goals and sustain a physical state (pragmatic action), active perception proponents also consider actions performed to select or obtain more information from the environment (epistemic action) [3, 4]. Active perception through sensori-motor regularities helps to overcome the complexity of sensory flows by focusing on the predictability of the
flows modification caused by the action [7]
The PhD candidate will work within the AMPLIFIER team to test the hypothesis that active perception is a key element in multimodal merging (especially for weighting the relevant information in perception) and how it can co-evolve with sensori-motor regularities learning. The approach entails a computer science orientation, yet in strong interaction with psychophysic/cognitive science and robotics. The candidate will have to ground its developments in the neuroscience and psychophysics literature, and contribute to the modeling of the psychophysical data (probably via student supervision). His/Her main objective will be to propose and implement a bio-inspired model for active multisensory integration and learning (based on the previous work of the advisors [5, 10] using neural fields, a dynamical model of neuronal activity) that will be tested on a social robot toward the end of the thesis.
Profile
===
Ideally, the candidate would have the following skills:
• background in artificial intelligence / cognitive science / computational neuroscience / machine learning (or equivalent)
• good programming skills (especially in Python)
• interest in neuroscience and/or psychophysics
• ability to work in a multi-disciplinary and multi-site team
• previous experience in a scientific environment
• good reporting/documentation skills
• good written/oral English skills
• autonomy
Any of these skills will be a plus:
• programming skills in web technologies
• previous experience with robots (especially Nao and Pepper)
• previous experience with machine learning/artificial intelligence algorithms
• previous experience in neuroscience and/or psychophysics
Localization
===
18 months at LIRIS laboratory, Lyon, France and 18 months at LJK laboratory, Grenoble, France. The precise split between the two locations is not fixed and can be negotiated.
The student will also have to work at the Hoomano office in Lyon for the robotic applications.
Duration
===
3 years (standard PhD duration in France) with an ideal starting date in September/October 2018 (can be negotiated).
Remuneration
===
Around 1685€/month gross salary. Funding is guaranteed for 3 years (obtained on a regional call - ARC AuRA). Additionally to his research, the candidate can also give lessons at universities in Lyon and/or Grenoble with additional remuneration.
Advisors
===
• Mathieu Lefort: associate professor at SMA group, LIRIS laboratory, Lyon
• Jean-Charles Quinton: associate professor at SVH team, LJK laboratory, Grenoble
Application
===
To apply, please send a CV and application letter to Mathieu Lefort (mathieu.lefort@liris.cnrs.fr) and Jean-Charles Quinton (jean-charles.quinton@imag.fr). Candidates can apply until the 17th of June. Interviews will be done the week after, aiming a final decision before the end of June.
If you have any question regarding this position, please send an email to Mathieu Lefort.
References
===
[1] Pradeep K Atrey, M Anwar Hossain, Abdulmotaleb El Saddik, and Mohan S Kankanhalli. Multimodal fusion for multimedia analysis: a survey. Multimedia systems , 16(6):345-379, 2010.
[2] Marc O Ernst and Martin S Banks. Humans integrate visual and haptic information in a statistically optimal fashion. Nature , 415(6870):429-433, 2002.
[3] Karl Friston, Jérémie Mattout, and James Kilner. Action understanding and active inference. Biological cybernetics , 104(1-2):137-160, 2011.
[4] David Kirsh and Paul Maglio. On distinguishing epistemic from pragmatic action. Cognitive science , 18(4):513-549, 1994.
[5] Mathieu Lefort, Yann Boniface, and Bernard Girau. Somma: Cortically inspired paradigms for multimodal processing. In IJCNN 2013 , pages 1-8. IEEE, 2013.
[6] Matteo Mossio and Dario Taraborelli. Action-dependent perceptual invariants: From ecological to sensorimotor approaches. Consciousness and cognition , 17(4):1324-1340, 2008.
[7] J Kevin O'Regan and Alva Noë. A sensorimotor account of vision and visual consciousness. Behavioral and brain sciences , 24(5):939-973, 2001.
[8] Jean Piaget. La naissance de l'intelligence chez l'enfant, volume 370. Delachaux et Niestlé Neufchâtel, Switzerland, 1977.
[9] Barry E Stein and M Alex Meredith. The merging of the senses. The MIT Press, 1993.
[10] Nicola Catenacci Volpi, Jean Charles Quinton, and Giovanni Pezzulo. How active perception and attractor dynamics shape perceptual categorization: A computational model. Neural Networks, 60:1-16, 2014.
[11] Robert B Welch. Intersensory interactions. Handbook of perception and human performance. Sensory processes and perception, 1986.