BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Machine Learning
DTSTART:20170221T140000
DTEND:20170221T160000
DTSTAMP:20260407T134009Z
UID:cf6072eaa2932a055fdf86ed402eb0471738ea4bb340c76222717e1c
CATEGORIES:Conferences - Seminars
DESCRIPTION:Honnet Pinheiro Canévet\nPierre-Edouard Honnet\, Pedro Olivei
 ra Pinheiro\, and Olivier Canévet just completed their PhD theses at Idia
 p (Martigny) affiliated with EPFL. Their research is in the field of Artif
 icial Intelligence (Speech for Pierre-Edouard\, and Computer Vision for Pe
 dro and Olivier). They present their research in the frame of a small "Mac
 hine Learning" seminar.\n\n-----------------------------------------------
 -----------------------\n\nSpeaker: Pierre-Edouard Honnet\nTitle: Intonati
 on Modelling for Speech Synthesis and Emphasis\nPreservation\nPresentation
  in English\n\nSpeech-to-speech translation is a framework which recognise
 s speech in\nan input language\, translates it to a target language and sy
 nthesises\nspeech in this target language. This presentation will deal wit
 h\naspects of speech-to-speech translation which are lost in traditional\n
 systems. Motivated by the Swiss multilingual context\, the development\nof
  an intonation model will be presented\, and some of its application\nfor 
 speech synthesis. The model is physiologically plausible\, and an\nautomat
 ic extraction method is proposed to retrieve its parameters. In\na convers
 ation scenario\, it is interesting to be able to preserve word\nemphasis\,
  which indicates what in the sentence is important\, or what\nis the impli
 cit message of the speaker. Following\, we apply the model\nto word emphas
 is synthesis\, using random forest to predict word level\nintonation conto
 urs.\n\n------------------------------------------------------------------
 ----\n\nSpeaker: Pedro Oliveira Pinheiro\nTitle: Large-Scale Image Segment
 ation with Convolutional Neural Networks\nPresentation in English\n\nObjec
 t recognition is one of the most important problems in computer\nvision. A
  main challenge is the problem of variability: objects can\nappear across 
 huge variations in pose\, appearance\, illumination and\nocclusion\, and a
  visual system need to be robust to all these\nchanges. In this presentati
 on\, I will show how we can leverage\ninformation of large-scale datasets 
 to deal with pixel-level\nrecognition. We aim to algorithms that require t
 he least amount of\nfeature engineering and are easy to scale.\n\n--------
 --------------------------------------------------------------\n\nSpeaker:
  Olivier Canévet\nTitle: Object Detection with Active Sample Harvesting\n
 Présentation en français\n\nEn vision par ordinateur (computer vision) o
 u en apprentissage\nautomatique (machine learning)\, que ce soit pour entr
 aîner un\nclassifieur d'images ou un détecteur d'objets\, la phase\nd'ap
 prentissage se résume à trouver une frontière de décision optimale\nen
 tre les classes. En pratique\, les exemples d'apprentissage n'ont pas\ntou
 s la même importance. Certains sont aisément classifiés\, tandis que\nd
 'autres\, proches de la frontière ou mal classés\, sont ceux qui ont de\
 nl'importance. Cependant\, la plupart des méthodes d'apprentissage\nséle
 ctionnent les exemples et les images de manière uniforme\, et leur\naccor
 dent la même importance. Le but de nos travaux a été de mettre au\npoin
 t des méthodes pour trouver efficacement les exemples\nd'apprentissage le
 s plus informatifs\, sans jamais accéder à la\ntotalité de l'ensemble d
 'apprentissage.
LOCATION:Zeuzier https://www.google.com/maps/place/EPFL+Valais+Wallis/?ref
 =zeuzier
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR