阿摩線上測驗 登入

最新科目

最新試卷

最新試題

最新申論題

最新課程

最新主題筆記

最新討論

62 “A corporation offering and selling its securities to the public has to file a registration statement with the competent authority. The registration statement automatically becomes effective 20 days after it is filed with the competent authority, at which point the issuer is free to sell the registered securities to the public. However, the competent authority has certain powers to delay or suspend the effectiveness of the registration statement if it appears that the statement is on its face incomplete or inaccurate in any material respect.” Based on the above description, which of the following is not a part of the securities registration process? (A)Discuss the terms of the offering with the competent authority. (B)Prepare the registration statement by the issuer. (C)File the registration statement with the competent authority. (D)Wait for a certain period of time before the registration statement becomes effective.

68 In Palko v. Connecticut, the petitioner argued that the Fourteenth Amendment__________ , as against the States, the Fifth Amendment requirement that no person “be subject for the same offence to be twice put in jeopardy of life or limb.” The Court disagreed, holding that federal double jeopardy standards were not applicable against the States. (A) inserted (B) incorporated (C) declared (D) enumerated

63 Which of the following statement about cross examination is INCORRECT under Taiwan Criminal Procedure Code? (A) Leading questions are not permitted on cross examination. (B)The scope of cross examination shall be limited to the matters or its related matter revealed in direct examination. (C) Leading questions may be asked in cross examination if necessary. (D) Matters in supporting of new allegation by the cross-examiner may be brought out in cross examination with the court’s permission.

33. 在工業物聯網架構中,進行設備預測性維護(Predictive Maintenance) 時,若面對異常事件發生頻率極低、樣本高度不平衡的時間序列資料, 下列哪一種方法最能兼顧模型穩定性與異常偵測效能? (A)將每筆異常事件資料複製多次以提升模型對異常的辨識敏感度,搭 配全序列訓練模型(如 LSTM); (B)對時間序列進行差分與標準化後,使用傳統監督式學習模型(如 SVM)進行分類訓練; (C)使用經過時間序列特化的 SMOTE 技術生成異常樣本,以平衡異常與 正常資料比例; (D) 採用基於重建誤差的自編碼器模型( Sequence-to-Sequence Autoencoder)進行異常偵測,並僅使用正常資料進行訓練

46. 某電商公司想預測用戶是否會購買特定商品,資料中包含多種用戶屬性與行為特徵。分析師希望選出對購買結果最有預測價值的特徵,以提 升模型效能。下列哪一種描述最符合監督式特徵選擇(Supervised Feature Selection)的概念? (A)根據特徵的整體分布、變異度或資訊量進行篩選,而不直接參考目 標變數; (B)評估每個特徵與目標變數之間的相關性,選擇對預測結果貢獻最大 的特徵; (C)使用模型評估特徵對預測結果的重要性,並保留對目標變數影響較 大的欄位; (D)將特徵透過降維方法(如 PCA)轉換為新特徵,再用於模型訓練

50. 某雲端服務公司計畫將大型語言模型部署於線上系統,並以批次推論 (Batch Inference)方式處理每日上百萬筆用戶請求。專案團隊在評 估可能遇到的挑戰時,下列哪一項通常不會被視為批次推論階段的主要難題? (A)如何確保訓練語料的涵蓋性與標註品質,以避免模型偏差影響輸出; (B)當批次規模增大時,如何降低推論延遲並保持即時回應能力; (C)在推論過程中,有效管理與分配龐大的輸入資料量以避免資源壅塞; (D)在叢集環境中精確安排推論任務,以提升 GPU/TPU 等硬體資源的利用率