Re: Post-doc positions in data mining
Date: July 20, 2012 09:27AM
Another post-doc position in data mining that was advertised in my emails:
==========================================
A post-doc position is about to be available at INRIA Rennes, in the frame of the 2012 INRIA national post-doc campaign.
Applications should be submitted by means of the INRIA recruitment platform: left button on the page http://tinyurl.com/c7xhus3.
INRIA project-team DREAM
INRIA Centre de Rennes, Bretagne Atlantique/IRISA, France
Duration : 12 to 16 months
Beginning : before the end of year 2012
Post-doctoral position
Mining multi-scale and multi-variate environmental data for knowledge discovery
Context
In many scentific domains, sensors are used for observing systems and for studying the evolution of these systems' behavior in time and/or space.
In the agro-environmental domain, more and more sensors are recording the manifestations and the evolution of natural phenomena.
The data, recorded as time series, are used to design or to confirm scientific theories that explain the behavior of eco-systems.
For scientists, the difficulty of data analysis rises with the size of recorded data.
They are particularly desiring tools that could facilitate the emergence of interesting features from recorded data, e.g. a relevant regularity or divergence.
Problem
Interesting phenomena emerge more or less obviously in times series with respect to the selected abstraction level: some phenomenon may appear clearly when data are abstracted week by week whereas it is difficult to observe the same phenomenon at the day or month level.
Moreover, the optimal abstraction level evolves during time with the evolution of the context in which the measures are recorded.
In addition, when several sensors record different aspects of the same phenomenon, the measures are often correlated.
These correlations should be displayed to the scientists, such as the temporal causality, e.g. "when the value of some variable rises the value of some other variable decreases with a delay of 3 to 5 days".
Objective
The goal of this project is to design new methods for extracting multi-scale and multi-variate temporal patterns coming from different sensors.
This goal rises some difficult issues. What are the main scales present in the data? What are the relationships between these scales? What are the relationships between the different variables? How to model these relationships and reason with this model?
Tasks
• analyze symbolic and numerical representation proposals for multi-scale time series,
• analyze machine learning and temporal data mining methods of multi-scale multi-variate time series,
• propose algorithms for the simultaneous extraction of multi-scale and multi-variate temporal patterns from several time series,
• implement a prototype of the proposed algorithms,
• assess the proposal on artificial data and on a real dataset provided by INRA recording water quality at the exit of a watershed. The results will be discussed with expert in the agro-environmental domain.
Profile
• PhD in computer science, preferably with a speciality in data mining or symbolic or statistical machine learning
• knowledge in time series analysis, if possible
• good programming skills
Bibliography
[1] Euzenat J., An algebraic approach to granularity in time representation, Proc. 2nd IEEE international workshop on temporal representation and reasoning (TIME), pp 147-154, 1995.
[2] Castro N., Azevedo P., Multiresolution Motif Discovery in Time Series, in Proceedings of the SIAM International Conference on Data Mining (SDM 2010), 2010, pp. 665-676.
[3] Thomas Guyet; René Quiniou. Extracting temporal patterns from interval-based sequences, in International Joint Conference on Artificial Intelligence (IJCAI), Jul 2011, Barcelone, Spain
[4] Shahar Y, Musen MA., Knowledge-based temporal abstraction in clinical domains. Artif. Intell. Med. 1996 Jul;8(3):267-98.
Keywords
machine learning ; data mining ; temporal patterns ; multi-scale ; multi-variate ; time series
Contacts
René Quiniou (INRIA/IRISA - rene.quiniou@inria.fr) Thomas Guyet (Agrocampus Ouest/IRISA - thomas.guyet@agrocampus-ouest.fr) Alice Aubert (INRA - Alice.Aubert@rennes.inra.fr)