Spelling suggestions: "subject:"redundancy inn neural networks"" "subject:"redundancy iin neural networks""
1 |
DESENVOLVIMENTO DE UM SISTEMA BASEADO EM REDUNDÂNCIA ANALÍTICA E REDES NEURONAIS ARTIFICIAIS PARA RECUPERAÇÃO DE FALHAS NA INSTRUMENTAÇÃO DE SUBESTAÇÕES DE ENERGIA ELÉTRICA. / DEVELOPMENT OF A SYSTEM BASED ON REDUNDANCY ANALYTICAL AND ARTIFICIAL NEURONAL NETWORKS FOR RECOVERY OF ELECTRICITY SUBSTATION INSTRUMENTATION FAILURES.LOUREIRO, Ronnie Santiago 31 August 2012 (has links)
Submitted by Maria Aparecida (cidazen@gmail.com) on 2017-08-24T15:00:02Z
No. of bitstreams: 1
Ronnie.pdf: 3320281 bytes, checksum: 56be4f928c1366ece428d2ae6caf9627 (MD5) / Made available in DSpace on 2017-08-24T15:00:02Z (GMT). No. of bitstreams: 1
Ronnie.pdf: 3320281 bytes, checksum: 56be4f928c1366ece428d2ae6caf9627 (MD5)
Previous issue date: 2012-08-31 / This work aims to monitor and analyze the data from the instrumentation system of a
substation as a way to identify false alarms, which can result in a decision by the
mistaken maintenance and operation. This project was conceived because of the
need for a research and development project which is called Maintenance
Management Center (MMC) whose overall objective is to assist in the maintenance
of their equipment operational intervention. Data is extracted from the automation
system that has digital relay protection function and measurement of the electric grid,
passing through a sequence of data processing to achieve the results that will serve
for the detection and diagnosis of faults. We applied methods based on quantitative
model by transforming the data system of continuous variables (SVC) and qualitative
data by transforming the system of discrete event (SDE) applying analytical
redundancy techniques and neural networks respectively, thus aiming a simplified
model for detection and diagnosis fault (DDF). The model has been designed taking
into account the characteristics DDF due to its stages, thereby providing a good
system failure recovery. Know filter if certain event is real or a false alarm is not an
easy task, but this system will have to meet this purpose. Technological resources
are used fairly consolidated in the industrial process for the integration of the
solution, because the time factor and information processing are critical in the results
generated by the system recovery. Another key point of this trial was to have
developed a system based on experiential knowledge, because it has higher
robustness in results. / Este trabalho tem como objetivo monitorar e analisar os dados provenientes do
sistema de instrumentação de uma subestação como forma de identificar falsos
alarmes, que pode acarretar em uma tomada de decisão equivocada por parte da
manutenção e operação. Este projeto foi concebido devido à necessidade de um
projeto de pesquisa e desenvolvimento que se intitula Centro de Gestão da
Manutenção (CGM) cujo objetivo global é auxiliar a manutenção na intervenção
operacional de seus equipamentos. Os dados são extraídos do sistema de
automação provenientes dos reles digitais que tem função de proteção e medição da
rede elétrica, passando por um sequencia de transformação dos dados até chegar
aos resultados, que servirá para detecção e diagnostico de falhas. Foram aplicados
métodos baseados no modelo quantitativo através da transformação dos dados do
sistema de variáveis contínuas (SVC) e qualitativo através da transformação dos
dados do sistema de eventos discretos (SED) aplicando técnicas de redundância
analítica e redes neurais respectivamente, objetivando assim um modelo
simplificado para detecção e diagnóstico da falha (DDF). O modelo foi concebido
levando em consideração as características DDF decorrente de suas etapas,
propiciando assim um bom sistema de recuperação de falha. Saber filtrar se
determinado evento é real ou um falso alarme não é uma tarefa fácil, porém este
sistema terá que atender este propósito. Foram utilizados recursos tecnológicos
bastante consolidados no processo industrial para garantir a integração da solução,
pois o fator tempo e o processamento da informação são decisivos nos resultados
gerados pelo sistema de recuperação. Outro ponto fundamental neste trabalho foi ter
desenvolvido um sistema baseado no conhecimento experimental, pois se tem maior
robustez nos resultados.
|
2 |
Distillation or loss of information? : The effects of distillation on model redundancySventickaite, Eva Elzbieta January 2022 (has links)
The necessity for billions of parameters in large language models has lately been questioned as there are still unanswered questions regarding how information is captured in the networks. It could be argued that without this knowledge, there may be a tendency to overparametarize the models. In turn, the investigation of model redundancy and the methods which minimize it is important both to the academic and commercial entities. As such, the two main goals of this project were to, firstly, discover whether one of such methods, namely, distillation, reduces the redundancy of the language models without losing linguistic capabilities and, secondly, to determine whether the model architecture or multilingualism has a bigger effect on said reduction. To do so, ten models, both monolingual, multilingual, and their distilled counterparts, were evaluated layer and neuron-wise. In terms of layers, we have evaluated the layer correlation of all models by visualising heatmaps and calculating the average per layer similarity. For establishing the neuron-level redundancy, a classifier probe was applied on the model neurons, both the whole model and reduced by applying a clustering algorithm, and its performance was assessed for two tasks, Part-of-Speech (POS) and Dependency (DEP) tagging. To determine the distillation effects on the multilingualism of the models, we have investigated cross-lingual transfer for the same tasks and compared the results of the classifier as applied on multilingual models and one distilled variant in ten languages, nine Indo-European and one non-Indo-European. The results show that distillation reduces the number of redundant neurons at the cost of losing some of the linguistic knowledge. In addition, the redundancy in the distilled models is mainly attributed to the architecture on which it is based, with the multilingualism aspect having only a mild impact. Finally, the cross-lingual transfer experiments have shown that after distillation the model loses the ability to capture some languages more than others. In turn, the outcome of the project suggests that distillation could be applied to reduce the size of billion parameter models and is a promising method in terms of reducing the redundancy in current language models.
|
Page generated in 0.147 seconds