Full text
| |
| Author(s): |
Thalita Firmo Drumond
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação |
| Defense date: | 2016-08-29 |
| Examining board members: |
Fernando José Von Zuben;
Guilherme Palermo Coelho;
Romis Ribeiro de Faissol Attux
|
| Advisor: | Fernando José Von Zuben |
| Abstract | |
The advances in information technology are promoting the generation and storage of an ever-increasing amount of data, produced by virtually all fields of human activity. So, there are distinguished demands for the analysis and knowledge extraction of these data, aiming at better understanding the involved processes, detecting trends and supporting decisions in the most varied fields. In this context, the study and development of scalable and efficient data mining tools and machine learning frameworks are highly desirable. One of such techniques is co-clustering, which aims to cluster simultaneously objects and their attributes, locating groups presenting some internal coherence pattern. This family of algorithms has been commonly applied to gene expression data, in order to identify groups of genes with coherent expression patterns under a group of conditions. There are also several works using co-clustering for collaborative filtering, an approach commonly used for recommender systems, which aims to suggest objects or contents, such as movies or books, that might interest a user. In this work, different co-clustering techniques were explored in two different application domains. First, a collaborative filtering framework, using a co-clustering-based matrix factorization technique, was extended through a robust and scalable ensemble approach using random projections for dimensionality reduction and approximate nearest neighbors. The agregation of each technique was studied individually, with experiments on real-world datasets commonly used in the litterature. The proposed ensemble was also compared to traditional and state-of-the-art techniques, exhibiting competitive results. Secondly, a contiguous co-clustering technique was applied to different neuronal brain activity time-series, looking for coherent temporal activity patterns between brain regions. Patterns found were used to construct a dynamic brain mapping, expressed by connectivity patterns that evolve with time. Those obtained functional maps are relevant for visualization of the numerous patterns found by the co-clustering algorithm, allowing the discrimination between patients and controls (AU) | |
| FAPESP's process: | 14/11125-1 - Co-clustering in ensembles of collaborative filters |
| Grantee: | Thalita Firmo Drumond |
| Support Opportunities: | Scholarships in Brazil - Master |