[CLOSED] [DPL_2016_DM_014] Inferring Device Failures from Online Discussions

Technicolor maintains large datasets of failed and returned hardware equipment (Internet gateways, STBs etc.), including dates and cause of failures. While this dataset provides interesting statistics, it does not allow anticipating return of hardware nor provides insight on how these failures are perceived by end-users.

The objective of this training is to develop a model for the prediction of hardware returns, using the observed online user activity. We would like to verify if the online discussions reflect the failures observed in the returned hardware. We would like to answer the following questions: Can we anticipate and correlate the trends in online discussions with the trends in the returned devices? How are the problems and failures perceived by the end-user? Can we model the behaviour of online users? In addition to the model, a service crawling (monitoring) the web for triggering alerts is a possible training outcome.

Skills: Analytical and modelling capabilities (a machine learning background is a plus)

Good programming skills (python, java, …)

Keywords: Web crawling, social networks, online chatter, device failures, modelling, machine learning.

This internship is located in Rennes, France. If interested, please apply at stage.rennes@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [DPL_2016_DM_015] Guess who is sitting on the couch

Electronic Program Guide (EPG) usually proposed on internet or DVB stream hold poor information on the programs. It is often quite limited so it can be augmented using third party information such as IMDB or Wikipedia.

An analysis of the resulting programs description, using natural language processing (NLP) and combined with TV usage data will allow to determine the household composition and separating the profiles of its members.

The intern will implement a profile separation algorithm to determine households compositions based on TV usage analysis and to generate an explicit profile for each of the household member.

Skills : Statistics, Data Mining, development (Python or Java, R), curiosity!

Keywords : user analytics and profiling, Statistics, Natural Language Processing (NLP)

This internship is located in Rennes, France.  If interested, please apply at stage.rennes@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [DPL_2016_DM_016] acoustic gait/staircase person recognition

Technicolor is leveraging is position within the IoT arena for some years now and in this context collecting data from different sensors and analyse them becomes a key topic to address for the company.

In particular learning about the habit of elderly people and sending appropriate notifications to medical staff or relatives is considered as a promising area.

Based on our current collaboration with geriatric centres, GAIT is a parameter that reveals many aspects of elderly people disorder and a common approach is to use wearable IMS(Inertial Motion Sensor) but a problems emerges in doing it that way: its weak acceptance ratio from the concerned people and then it’s possibly failure on a wide deployment perspective. Another approach is to consider to not wear any sensor, that’s the aim of the problematic proposed in this internship that will focus on evaluating an acoustic based approach.

The expectations of this internship are the following:

  • To constitute a realistic audio database in collecting with a smartphone different people GAIT/staircases.
  • To find the best pre-processing method to extract relevant acoustic features
  • To compare different classifiers performances for which the main criteria are precision and recall ratios
  • To give argued recommendations on the choice of classifiers and more globally on the results performed by this approach

The challenge relies on finding the best compromise between the number of features vector and the acoustic recognition efficiency.

Skills :

- Audio data processing, machine learning, feature extraction, development skills.

- Good knowledge of English enabling easy reading of scientific literature and drafting of documentation

- Python, Python notebook 2.7/3.4, sciKitLearn, Librosa, IDE Eclipse or Spyder

Keywords: machine learning, feature extraction, sound recognition, GAIT recognition.

This internship is located in Rennes, France. If interested, please apply at stage.rennes@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [DPL_2016_DM_017] Mining activities of the daily living from sensor data for Smart Home applications

The increasing aging population has inspired many machine learning researchers to find innovative solutions for assisted living. A problem often encountered in this context is Activity Recognition (AR) that aims to identify the actions carried out by a person given a set of sensor observations. The challenge here is to infer higher-level behaviour (activities) from a combination of low-level input data (sensors) for assisting the seniors in their daily life (automation) but also creating alerts (falls, declining activity).

After analysing and describing the existing datasets, the intern will apply machine learning algorithms in order to recognize or detect activities of the daily living. He or she will constitute a survey, choose approaches, and propose innovative solutions.

Depending on obtained results and innovative ideas, this work may lead to patents and to research publication.

Skills

Machine learning and data mining; development skills (Python or R); strong interest in research, English mandatory

Appreciates working with a team spirit; curiosity.

Keywords: data mining, machine learning, sensors, activity recognition.

This internship is located in Rennes, France. If interested, please apply at stage.rennes@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.

[CLOSED] [DPL_2016_DM_076] Transfer learning applied to activity Recognition

Summary - Digital life is now effective: connected devices are mushrooming, bringing the promise of making our daily life easier and connected. Today, digital services tend to get smarter, capturing the end-user context and proposing service personalization. The user context is a complex asset that can be broken down into categories: ‘Who’, ‘Where’, ‘Doing what’, ‘With Whom’.

In this study we will focus on the ‘Doing what’ category and consider Activity Recognition (AR) based on wearable multi-sensor platforms. A previous work has shown that AR model trained on data collected from a small number of individuals may not perform well when applied to other users.

The goal of this internship will be to investigate techniques such as transfer learning or active learning in order to define a training procedure able to adapt a generic model to any end-user and able to evolve across time according to changes in the user behavior.

Skills : Machine learning, data mining tools (R, SciPy, Weka), Java, Python, working knowledge of English

Keywords : Ambient Intelligence, Machine Learning, Body Sensing, Activity Recognition, Wearable Computing, Transfer Learning

This internship is located in Rennes, France. If interested, please apply at stage.rennes@technicolor.com by sending us your resume and a cover letter with the internship reference in the email subject line.