| dc.contributor | LUIS ENRIQUE SUCAR SUCCAR | |
| dc.creator | LINDSEY JENNIFER FIEDLER CAMERAS | |
| dc.date | 2013 | |
| dc.date.accessioned | 2023-07-25T16:21:10Z | |
| dc.date.available | 2023-07-25T16:21:10Z | |
| dc.identifier | http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/229 | |
| dc.identifier.uri | https://repositorioslatinoamericanos.uchile.cl/handle/2250/7805450 | |
| dc.description | Probabilistic graphical models are a common choice for modeling domains
that operate under uncertainty. Within this framework, Bayesian networks have
become very popular due to their highly intuitive representation that makes their
interpretation easy for someone unfamiliar with computational models. Temporal
Nodes Bayesian Networks extend this concept by including a dynamic component
to the model. Specifically, they model the relationships existing between
events and when they occur. Recently, an algorithm for inducing a temporal
nodes Bayesian network from a database was defined. This algorithm, denominated
LIPS (Learning Interval, Parameters and Structure), implemented a novel
strategy for learning the dynamic aspects of the model, while using traditional
Bayesian network learning techniques for the structure and the parameters. Unfortunately,
the results obtained by the LIPS algorithm depend heavily on the
amount of data available for learning, and just as with many other traditional
machine learning algorithms, learning from scarce data sets leads to unreliable
models that produce poor results. Transfer learning is a paradigm that compensates
for scarce data sets by borrowing information from other models that have
already been learned. These auxiliary models differ from the target learning task,
but hold some degree of similarity with it, such that the learning of certain aspects
of the target model can be aided by this information transfer. Previously,
work has been done for transfer learning in the area of probabilistic graphical
models; however, the incorporation of transfer learning to dynamic models is relatively
unexplored. In this thesis, a methodology for inducing a Temporal Nodes
Bayesian Network using transfer learning is proposed. The algorithm defines a
separate strategy for learning each component, where transfer learning is applied
to induce the dynamic aspects of the model, as well as the structure and the
parameters. Experimentation was carried out to evaluate the algorithm, and
the results were compared to those obtained without transfer and by applying
naive transfer. Results show that the proposed algorithm obtains models that
are significantly better than those obtained by learning only from scarce data sets
(without transfer). In addition, they show that the algorithm is able to retrieve
reliable models even when few records are available for the task of interest. | |
| dc.format | application/pdf | |
| dc.language | eng | |
| dc.publisher | Instituto Nacional de Astrofísica, Óptica y Electrónica | |
| dc.relation | citation:Fiedler-Cameras L. | |
| dc.rights | info:eu-repo/semantics/openAccess | |
| dc.rights | http://creativecommons.org/licenses/by-nc-nd/4.0 | |
| dc.subject | info:eu-repo/classification/Transferencia de aprendizaje/Transfer learning | |
| dc.subject | info:eu-repo/classification/Redes bayesianas de nodos temporales/Temporal nodes bayesiun networks | |
| dc.subject | info:eu-repo/classification/Modelos gráficos probabilísticos/Probabilistic graphical models | |
| dc.subject | info:eu-repo/classification/Razonamiento temporal/Temporal reasoning | |
| dc.subject | info:eu-repo/classification/cti/1 | |
| dc.subject | info:eu-repo/classification/cti/12 | |
| dc.subject | info:eu-repo/classification/cti/1203 | |
| dc.subject | info:eu-repo/classification/cti/1203 | |
| dc.title | Transfer learning for temporal nodes bayesian networks | |
| dc.type | info:eu-repo/semantics/masterThesis | |
| dc.type | info:eu-repo/semantics/acceptedVersion | |
| dc.audience | students | |
| dc.audience | researchers | |
| dc.audience | generalPublic | |
