The France-Stanford Center for Interdisciplinary Studies and the France-Berkeley Fund published the list of their 2015 call laureates, of which 3 Inria awardees..
France-Stanford Center for Interdisciplinary Studies 2015-2016 awardees:
- Visiting Student Researcher Fellowship in Computer Science:Gabin Personeni, LORIA/INRIA, Nancy, France
The widespread use of patient Electronic Health Records in hospitals generates large volumes of data offering exciting opportunities for novel discoveries in medicine. Indeed, we consider these records for a secondary use that is to constitute patient cohorts useful to experiment complex biomedical hypotheses. For instance, these medical records are used to monitor the safety of drugs and to alert about drugs that are prescribed but should be withdrawn from the market.Beside these patient records, data describing the current state of human knowledge about diseases, genetics and drug mechanisms has been recently published. This “state-of-the-art” data can be used in conjunction with medical records to help guiding knowledge discovery.
However, medical records are encoded in a hospital-specific manner, making them hard to interconnect with state-of-the-art data. Our research project is to integrate medical records data from the Stanford Hospital (available in a database named INTREPID) with the state-of-the-art biological knowledge, using an appropriate informatics toolbox: the semantic web. Once combined, these two datasets will be used to gain new insight on why distinct populations react differently to drugs and, when possible, to explain the biological mechanisms that may explain these differences.
- Collaborative Projects Grant Recipients in Engineering: Inertial Sensors Based Analysis of Gait on Children with Spastic Cerebral Palsy: Jessica Rose, Department of Orthopaedic Surgery, Stanford University and Christine Azevedo-Coste, DEMAR Inria Research Team/LIRMM, Montpellier, France
Analysis of walking abnormalities is an important clinical assessment used for treatment of gait disorders in children with cerebral palsy (CP). Camera-‐ based motion capture, the current gold standard, enables practitioners to perform gait analyses with high accuracy. However, the technology can only be used in the laboratory where motion capture is constrained to a limited space and incurs significant expense. Mobile systems are now possible using light-‐weight wearable sensors known as inertial measurement units (IMU). These sensor-‐based systems have potential to provide a more efficient, mobile alternative for movement analysis and can offer real-‐time feedback to patients for more effective rehabilitation. We propose an interdisciplinary collaboration between the Department of Orthopedic Surgery at Stanford University and the Institut National de Recherche en Informatique et Automatique (INRIA) in Montpellier. Our aim is to quantitatively assess walking problems associated with CP, using wearable technology. Despite their small size, ease-‐of-‐use, robust design and low-‐cost, numerous recognized technical issues make the use of IMUs relatively complex. Through a series of experiments we will combine our efforts and complementary skills to propose an IMU sensor system and software to extract meaningful gait parameters for rehabilitation of children with CP.
Source: http://francestanford.stanford.edu/
France Berkeley Fund 2015 awardees:
Graph-NN: Computing and Manipulating Very Large Dynamic Graphs of Nearest Neighbors: Michael Franklin, Department of Electrical Engineering and Computer Sciences, University of California, Berkeley and Laurent Amsaleg, LINKMEDIA Inria Research Team, IRISA-CNRS, Université Rennes I, Rennes.
The objectives of Graph-NN are to use the distributed graph-dedicated GraphX system from the AMPLab to design new k-nn graph algorithms for multimedia that (i) work at very large scale and (ii) are dynamic. K-nn graphs constructed over very large datasets will allow to investigate the extent with which high-dimensional indexing, image/video/audio recognition etc. are improved. Working in this direction allows us to already anticipate few challenges. A first challenge is related to constructing the graph. Because GraphX distributes computations, any partitioning of the points across CPUs is likely to significantly increase communication costs due of the high-dimensionality of the data.
Scale makes things more complicated as we typically want to construct a k-nn graph over more than 107-9 items with k ≈ 1000. Existing solutions typically construct graphs few orders of magnitude smaller. It is also very challenging to consider dynamic k-nn graphs where, e.g., a photo/video collection grows as the time goes by. There, neighbors may disappear from the horizon of a point while other may appear, depending on the evolution of neighborhoods. Adding dynamicity to GraphX is challenging.
