UNFEX · Utilising and enhancing unsupervised feature extraction
6РП — Действия „Мария Кюри“
- Период
- 2005-11-01 → 2007-10-31
- Финансиране от ЕС
- 152 809 €
- Участници
- 1
- Схема
- EIF
Линиите свързват координатора с партньорите.
Накратко на български
Методите за автоматично извличане на характеристики анализират сложни данни и текстови връзки, за да разберат дали един текст следва логически от друг. Това помага на специалистите да оценят качеството на резултатите от съвременните алгоритми за обработка на данни.
Кратко обяснение, генерирано от езиков модел по текста на CORDIS. Оригиналът е по-долу.
Резултати накратко
Final Activity Report Summary - UNFEX (Utilising and enhancing unsupervised feature extraction)
During my fellowship I developed and analysed new model selection methods for unsupervised feature extraction like nonlinear dimensionality reduction (NLDR). Those are techniques which are able to find simple descriptions of high-dimensional data, as they appear in all branches of science and industry. The results of my work can help practitioners to assess the quality of results obtained from state-of-the-art NLDR algorithms. Furthermore, I proposed and evaluated a new approach to the problem of textual entailment which tries to automatically answer the question whether one piece of text entails some other piece of text. For this, I combined ideas from logic and probability to obtain a calculus that tries to mimic human reasoning (which is often only approximately sound).
Текст от CORDIS, на английски · Данни: CORDIS, © Европейски съюз
Цел на проекта
All data-driven research areas depend eminently on intelligent data analysis. This project will advance the state of the art in unsupervised data analysis by addressing two important issues:- For practitioners it is very difficult to decide which of the numerous methods for unsupervised feature extraction is appropriate for given data. This project will develop practical tools for model selection to utilise the plenty of approaches.- New paradigms in feature extraction lead to special features with special properties.To this end we will focus on developing methods that extract statistically independent features, hereby aiming to solve the problem of nonlinear ICA. Nonlinear independent features give a particularly succinct description of data. Methods for nonlinear ICA that can deal with strongly nonlinearly mixed data might become the basis for new sophisticated approaches to density estimation in high dimensions.
Оригинален текст от CORDIS (на английски).
Участници
- UNIVERSITY OF EDINBURGH · EDINBURGHКоординаторОбединеното кралство
Връзки
Данни: CORDIS, © Европейски съюз