Kuis Perkiraan Permintaan AI

Jawaban dan penjelasan muncul saat Anda mengikuti kuis.

1. Why can one global forecasting model help when an organization has many related series?
   • It can learn shared patterns across series while retaining series-specific inputs
   • It guarantees the same forecast error for every series
   • It replaces chronological validation with a random split
   • It means one model is trained separately for each series
2. Mengapa perkiraan probabilistik (kuantil) berharga untuk perencanaan inventaris?
   • Mereka memberikan satu nomor jaminan
   • Indikator-indikator tersebut mengungkapkan jangkauan dan keyakinan sehingga para perencana dapat menetapkan persediaan pengaman terhadap risiko
   • Mereka menghilangkan kebutuhan akan data historis apa pun
   • Mereka membuat perkiraan berjalan lebih cepat
3. Manakah metode perkiraan statistik klasik yang mendahului pendekatan ML modern?
   • Transformator Fusi Temporal
   • DeepAR
   • ARIMA
   • WaktuFM
4. Kendala penting apa yang harus dihindari saat melatih peramal permintaan?
   • Menggunakan fitur kalender
   • Termasuk informasi promosi
   • Menghasilkan interval kepercayaan
   • Membocorkan data masa depan ke dalam pelatihan prediksi masa lalu
5. Sinyal eksternal manakah yang paling masuk akal akan meningkatkan perkiraan penjualan es krim?
   • Suhu dan cuaca harian
   • Log waktu aktif server
   • Ulang tahun CEO
   • Pilihan font di situs web
6. What did the cited decoder-only foundation-model paper report about its model’s zero-shot forecasts?
   • It guarantees better forecasts for every business series
   • On several public datasets, out-of-the-box accuracy came close to supervised models trained for each dataset
   • It eliminates the need for statistical baselines
   • It forecasts only with fine-tuning on each target series
7. What architecture does the DeepAR paper use to produce forecasts across related time series?
   • A single ARIMA model fitted separately to each series
   • A static lookup table of future values
   • An autoregressive recurrent network trained on many related series
   • A diffusion image model that reads plotted charts
8. Why evaluate a demand model against statistical baselines?
   • To ensure a neural model always wins
   • To make data leakage unavoidable
   • To remove uncertainty from demand
   • To test whether added model complexity improves the target forecast task