Hasil Pencarian  ::  Simpan CSV :: Kembali

Hasil Pencarian

Ditemukan 2 dokumen yang sesuai dengan query
cover
Kezia Sulami
"Machine Learning (ML) sebagai bagian dari Artificial Intelligence (AI) telah membuat komputer mampu melakukan hal-hal yang membutuhkan kecerdasan manusia secara otomatis. Binarized Neural Network (BNN) merupakan arsitektur ML modern yang memiliki keunggulan yakni penggunaan memori yang efisien dan performa yang baik. Namun, seperti neural network pada umumnya, BNN juga merupakan black-box model yang memiliki kesulitan dalam menjelaskan prediksi yang dihasilkan. Penelitian ini menggunakan teknik abduction untuk memperoleh minimal explanations, dalam bentuk himpunan pasangan fitur dan nilainya, dari hasil prediksi BNN. BNN dimodelkan sebagai model Mixed-Integer Linear Programming (MILP) dan selanjutnya disederhanakan menjadi model Integer Linear Programming (ILP) yang merupakan bentuk formal agar dapat dilakukan teknik abduction. Hasil penelitian menunjukkan bahwa teknik abduction dapat digunakan untuk menjelaskan hasil prediksi BNN. Penelitian ini juga menerapkan teknik abduction untuk menghasilkan penjelasan subset-minimal pada hasil prediksi BNN untuk beberapa dataset.

Machine Learning (ML) as part of Artificial Intelligence (AI) has enabled computers to do things that require human intelligence automatically. Binarized Neural Network (BNN) is a modern ML architecture that has some advantages: efficient use of memory and good performance. However, like other neural networks in general, BNN is also a black-box model that has difficulties in explaining the resulting predictions. This research employs the abduction technique to obtain minimal explanations, that is a set of pairs of features and its values, from a BNN prediction. BNN is modeled as a Mixed-Integer Linear Programming (MILP) model and then further simplified into an Integer Linear Programming (ILP) model which is a suitable formalism for finding explanations using abduction. This research shows that the abduction technique can be used to explain BNN predictions. Furthermore, this research applies the abduction technique to produce subset-minimal explanations on BNN predictions for several datasets."
Depok: Fakultas Ilmu Komputer Universitas Indonesia, 2022
S-pdf
UI - Skripsi Membership  Universitas Indonesia Library
cover
Dwi Susilowati
"The urban population in Indonesia will soon be 50% of its total population. In the past, nutrition improvement programs were addressed to the rural community stressing on underweight. It is presumed that in the urban area both under and overweight are found. However, data on the subject are insufficient. A cross sectional study was done in East Jakarta. The study consisted of 679 households and 3442 peoples were chosen through a multistage random sampling. The analyses of the findings were based on households as well as on individuals.
The households were categorized based on the anthropometry of the individual family members. The indicator used, was BMI with a cut off 18.5 and 25 for those aged 17 years and more, and weight-for-age based on NCHS median + 2SD for those aged less then 17-year-old. Four households' categorizations were identified, Normal (all members had normal anthropometric status) 22.2%, Normal-Under (a combination of normal and underweight) 31.6%, Normal-Over (a combination of normal and overweight) 27.1% and Normal-Under-Over (a combination of normal, under and overweight) 17.2%. It covered 98.3% of all households.
Within households classified as having any overweight members, there were 37.9% overweight adults vs 2.3% overweight children. Within households classified as having any underweight members, there were 26.9% underweight adults vs 35.8% underweight children. Poorer households and households with children faced the risk for having underweight members, while richer households and households with adults faced the risk for having overweight members. Underweight is not only caused by poor socioeconomic background. There were differences in the food distribution patterns among the household members, and households with over weight members consumed more variety of foods.
Among children aged less than 17 years old, 21.4% were stunted (HAZ<-2SD), and 21.2% were underweight (WAZ<-2SD). Children aged less than 11 years 11.4% were wasted (WHZ<-2SD)_ Children aged less than 17 years whose HAZ, WAZ and WHZ were 2SD were less than 3%. Children aged 11 to 17 years old whose BMI was below 5%-tiles were 5.2%, while those whose BMI was above 85%-tiles were 15.2%. Adults aged 17 to 55 years old with BMI <18.5 were 15.6% and those with BMI >25 was 19.6%_ People aged more than 55 years old with BMI <18.5 was 18.7%, while those with BMI of X25 were 20.9%. Thus, underweight was prevalent in all age groups, but overweight was mainly found in adults and elderly. Among the adolescents the older age group showed a tendency of being overweight.
It is recommended that, household groupings according to the anthropometric status of the nuclear family members can be used as a screening method which serves the purpose of distinguishing the target population, in administrative, and project design and management terms, and to do an anthropometry assessment at regular intervals for all age groups. Special attention to detect overweight among adolescents should be given. Prevention of overweight in the community requires a multi sectoral approach rather than purely nutrition oriented programs. To continue the existing programs in nutrition and health, to prevent underweight in the community, especially for under fives, school children and adults. Within the poorer households there is positive deviance (as shown in Normal groups), this gives room for intervention."
Depok: Universitas Indonesia, 1997
D384
UI - Disertasi Membership  Universitas Indonesia Library