12 Nov- Seminari Ariadna Quattoni
El proper dijous 12 de novembre tenim programat un seminari a càrrec d'Ariadna Quattoni a l'aula S208 de l'Edifici Omega del Campus Nord de la UPC.
Aquests són els detalls del seminari:
Títol Learning Data Representations Using "Partial Supervision"
Ponent Ariadna Quattoni
Lloc Omega-S208
Dia 12 Novembre 2009
Horari 10:00h - Café
10:15h - Presentació
Abstract Classical machine learning approaches for inferring compact data representations (e.g. dimensionality reduction techniques) rely solely on unsupervised data. Given a high dimensional dataset these techniques discover lower dimensional representations that preserve relevant statistics of the data.
In some instances however, some form of "partial supervision" is readily available which can guide the search for a good representation. For example, imagine that our goal is to classify newspaper images into topics. In this case, we can use image captions as "partial supervision" to learn a representation usefull for topic classification. Similar "partial supervision" approaches have been used to induce representations for NLP tasks such as tagging and word-sense
disambiguacion.
In this talk I will first briefly review the classical unsupervised approaches to dimensionality reduction. I will then describe two "partial supervision" approaches and illustrate them in the context of real applications. Hopefully these methods will be also applicable to your NLP problems.
Creiem que serà especialment interessant.