Ask a question from expert
Optimal Parameter Settings Assignment
6 Pages1055 Words280 Views
Added on 2020-07-23
Optimal Parameter Settings Assignment
Added on 2020-07-23
BookmarkShareRelated Documents
Optimal Parameter Settings Using a GeneticAlgorithm
IntroductionRecent years have observed substantial improvements in personal computereyesight, as demonstrated simply by considerable progress of standardinformation sets in places such as encounter acceptance, object recognition,plus action recognition. Much of this particular improvement comes fromcombining methods within single strategies. Consequently, many moderneye-sight techniques contain image processing, device learning, and patternreputation techniques that work to solve a certain problem together.Unfortunately, fine tuning these multi-part algorithms can be difficult,particularly when changing the parameter in one part of a program mayhave unforeseen results on another.Training, Fine tuning, and Test DatasetsThe face area Recognition Vendor Test 06\ showed that face identificationtechnology could confirm the person’s identity with 99% accuracy in highquality pictures taken under controlled situations [13]. However, encounterrecognition in uncontrolled problems is much more difficult. The GBUchallenge problem [11] contains three dividers of face images associatedwith varying difficulty from uncontrollable environments. The Good partitionconsists of images that are easy to fit, while the Ugly partition is extremelydifficult, and the Poor partition is in between someplace. The purpose of GBUis to enhance performance on the Bad plus Ugly partitions without sacrificingfunctionality on the Good.Algorithm changes are represented as individuals within a simulatedindividuals, where healthier individuals are selected for achievement andbreeding. In this test, the population contains 100 at random generatedindividuals and each version uses these steps:oIndividuals of the habitants are chosen randomly.
oThose people are combined to produce a new person whereconfiguration factors are chosen randomly from the parents.oSmall perturbations are made to the newest individual to simulate.Genetic Algorithm and Configuration SpaceParameter Type Range Manual Value GA ValueRegion 0: Full Face Float 0.50 - 1.00 1.00 0.927Region 1: Left Eye Float 0.10 - 0.50 0.33 0.433Region 2: Right Eye Float 0.10 - 0.50 0.33 0.342Region 3: Far Left Brow Float 0.10 - 0.36 0.33 0.360Region 4: Center Left Brow Float 0.10 - 0.36 0.33 0.285Region 5: Center Right Brow Float 0.10 - 0.36 0.33 0.286Region 6: Far Right Brow Float 0.10 - 0.36 0.33 0.360Region 7: Nose Bridge Float 0.10 - 0.66 0.33 0.395Region 8: Nose Tip Float 0.10 - 0.66 0.33 0.100Region 9: Left Nose Float 0.10 - 0.70 0.33 0.211Region 10: Right Nose Float 0.10 - 0.70 0.33 0.259Region 11: Left Mouth Float 0.10 - 0.20 0.20 0.167Region 12: Center Mouth Float 0.10 - 0.20 0.20 0.200Region 13: Right Mouth Float 0.10 - 0.20 0.20 0.154SQI Blurring Radius Float 0.5 - 20.0 3.0 19.43PCA Min Dimension Int 0 - 20 2 19PCA Max Dimension Int 100 - 400 250 169PCA Whitening Enabled Bool True/False True TrueFinal Basis Dimensions Int 100 - 4000 3500 880The new individual is definitely examined using the fitnessfunctionality.If the new person scores higher than the formerly lowest rank individualwithin the population, that lowest positioned individual is replaced. Theparticular fitness and health function evaluates every individual bycompleting the full tests and training process. The particular algorithm wastrained utilizing the configuration in the genetic program code and then wasevaluated for the tuning-subset at the particular fake accept rate of zero.001. The GA had been run on the particular quad-core Intel i7 with 8employee processes which completed 40 evaluations per hour resulting in5004 total evaluations. Table one summarizes the 19 recommendations that
End of preview
Want to access all the pages? Upload your documents or become a member.