Post-Doc Position - LEAR team, INRIA, Grenoble, France
Large-scale image classification
The LEAR team at INRIA Grenoble is looking for a post-doctoral researcher with a specialization in Computer Vision and Machine Learning.
Activities
A current trend in image classification is to learn classifiers for many different object/image classes (10,000's), to reach the variety of classes recognizable by a human. See for example the Pascal/Imagenet challenge.
The starting point of the post-doc will be to investigate the relationship of large-scale supervised image classification with large-scale image retrieval, to leverage recent advances of the LEAR team in large-scale image retrieval using nearest-neighbor search on vectors [2, 3]. Indeed, the so-called "attribute" features allow to bridge the two tasks, by building features using an image retrieval algorithm for the purpose of image classifcation (see eg. [1, 6]).
An important characteristic to take into account when dealing with large image databases is the high level of noise. Within one category, images may vary in appearance and shape, and can also be mislabelled or ambiguously labelled. A rigorous machine learning framework to handle these different sources of noise is still lacking. In particular, designing machine learning algorithms robust to label noise or efficiently performing inference on the correct label from ambiguous ones is still an open issue.
Then a major challenge would be to design algorithms that scale linearly with the volume of the data, eventually outperforming algorithms learned without attribute-level features and with less training data. A promising line of research would be to combine the strengths of both online learning [4] and learning with sparsity-enforcing regularization penalty [5], in order to be able to train the learning with a large number of examples and large number of features.
Skills and Profile
- PhD degree (preferably in Computer Vision or Machine Learning)
- Solid programming skills; the project involves programming in C, Python and Matlab
- Solid mathematics knowledge (especially linear algebra and machine learning)
- Creative and highly motivated
- Fluent in written and spoken English
- Prior knowledge in the areas of online/large-scale learning is an asset
The position is offered at the "Rhone-Alpes" Research Unit of INRIA, located near Grenoble and Lyon, in France. The Unit includes more than 600 people, within 26 research teams and 10 support services. Grenoble is a lively city which hosts many foreign students and researchers. Located in the heart of the French Alps its direct surroundings offer great outdoor recreation including skiing, cycling, and hiking. Paris can be reached from Grenoble in 3h by train.
Contact: matthijs DOT douze AT inria DOT fr, zaid DOT harchaoui AT inria DOT fr, cordelia DOT schmid AT inria DOT fr
References
[1] Torresani, L.; Summer, M. , Fitzgibbon, A. Efficient Object Category Recognition Using Classemes ECCV, 2010
[2] Boiman, O.; Shechtman, E., Irani, M., In Defense of Nearest-Neighbor Based Image Classification, CVPR 2008
[3] Jegou, H.; Douze, M; Schmid, C; Product quantization for nearest neighbor search, PAMI 2011
[4] Bottou, L; Chapelle, O; DeCoste, D; Weston, J; Eds. Large-Scale Kernel Machines, MIT Press 2007
[5] Sra, S; Nowozin, S; Wright, S. J; Optimization for Machine Learning, MIT Press 2010.
[6] Douze, M; Ramisa, A; Schmid, C.; Combining attributes and Fisher vectors for efficient image retrieval, CVPR 2011