Post-doctoral position:
Predicting memorability of media content A post-doctoral research position is proposed in Technicolor’s Imaging Science Lab in Rennes Research & Innovation Center, France. The position is open for a PhD graduate on an 18 month, full time basis. Context Technicolor, a worldwide technology leader in the media and entertainment sector, is at the forefront of digital innovation. Our R&I conducts basic and applied research to foster innovation within all Technicolor Business Units (BUs). It thus aims at providing BUs with cutting-edge technologies fitting their needs as well as exploring longer term avenues toward new technologies and business opportunities. The Rennes center is the largest center of all Technicolor R&I facilities. It is located in middle of the a science park that gathers large corporate research centers (e.g., Orange, Canon) as well as major French academic research centers (e.g., INRIA, CNRS). The selected candidate will integrate a research team named “Content Processing” gathering more than 20 researchers, engineers, PhDs, and Postdocs coming from 10 different countries. The main line of research in the team is to understand, organize, and enhance content, both professional and usergenerated. Project The understanding of media contents such as images or videos remains a challenging research topic in machine learning and affective computing. Some recent research in this domain targets the modeling of more subjective criteria such as emotion, interestingness, and memorability of such content [1-9]. The project in which the postdoctoral fellow will be involved focuses on the memorability aspect, as knowing what makes an image or a video memorable enables high-potential applications in e.g., education, advertising, selective encoding or content database management. State-of-the-art approaches mainly concentrated on image memorability prediction. So far, they have explored both low-level image features with varied attributes and semantic features [5-8], as well as higher-level visual attention model [9] for the task. Human memorability has also been studied for a long time in Psychology [2, 4, 10] and some of the statements found in this axis of research probably deserve to be taken into account, within modern machine learning paradigms, to better improve memorability modeling and prediction. Indeed, memorability modeling is still at its learning stage, at least when multimodal content is concerned. More generally, the postdoctoral researcher will work on perceptual understanding of multimodal data (i.e., audio, visual, and possibly text associated with the video). Based on the further understanding of “what makes a content memorable?”, the development and implementation of innovative algorithms to model and predict memorability scores, for different types of multimodal content, is envisioned. In this context, multimodal fusion and temporal modeling of data, through machine learning techniques such as deep learning, will be explored. The successful postdoc will also be expected to present scientific results at international conferences and journals, as well as to generate intellectual property (patents). Technicolor R&D France 975 Avenue des Champs Blancs CS 17616 35576 Cesson-Sévigné, France www.technicolor.com
Skills & Requirements The candidate should hold a PhD related to signal processing/machine learning, affective computing, computer vision, or connected fields. Ideally the candidate has good knowledge of audio/image processing and machine learning techniques such as deep learning. Good programming skills (Python/Matlab, C++, Linux) and fluency in English are also required. Applicants should submit a curriculum vitae, a list of publications, a statement of research interests and preferably a recommendation letter. Contacts and application For more information, please contact Dr. Ngoc Q. K. Duong (email below). Applications should be sent to
[email protected] and
[email protected] References [1] L. Mai and G. Schoeller, “Emotions, attitudes and memorability associated with TV commercials”, Journal of Targeting, Measurement and Analysis for Marketing, 2009. [2] U. Rimmele, L. Davachi, R. Petrov, S. Dougal, and E. A. Phelps, “Emotion Enhances the Subjective Feeling of Remembering, Despite Lower Accuracy for Contextual Details,” Psychology Association, 2011. [3] C.H. Demarty, M. Sjoberg, B. Ionescu, T.T. Do, H. Wang, N. Q. K. Duong and F. Lefebvre, “MediaEval 2016 Predicting Media Interestingness Task,” Proc. of the MediaEval 2016 Workshop, Netherlands, Oct. 20-21, 2016. [4] S. Hamann, “Cognitive and neural mechanisms of emotional memory”, Trends in cognitive science, 2001 [5] Z. Bylinskii, P. Isola, C. Bainbridge, A. Torralba, and A. Oliva, “Intrinsic and extrinsic effects on image memorability”, Vision Research, pp. 165-178, 2015. [6] T. Konkle, T. F. Brady, G. A. Alvarez, and A. Oliva, “Scene memory is more detailed than you think the role of categories in visual long-term memory”, Psychological Science, 21(11), pp.15511556, 2010. [7] A. Khosla, A. S. Raju, A. Torralba, and A. Oliva, “Understanding and predicting image memorability at a large scale”, Proc. ICCV, 2015 [8] A. Khosla, A. Sarma, and R. Hamid, “What makes an image popular?”, Proc. WWW, 2014 [9] B. Celikkale, A. T. Erdem, and E. Erdem, “Visual attention driven spatial pooling for image memorability”, Proc. IEEE CVPR Workshop, 2013 [10] Y. Sakuta and J. Gyoba., “Affective impressions and memorability of color-form combinations”, Journal of General Psychology, 2006.
Technicolor R&D France 975 Avenue des Champs Blancs CS 17616 35576 Cesson-Sévigné, France www.technicolor.com