Alabama Pastor Convicted of 1999 Double Homicide After 20 Years
Advancements in DNA genealogy and the use of public DNA databases have enabled law enforcement to solve cold cases, bringing justice to victims and their families, even after decades.
評估基於最先進機器學習的線上網路入侵偵測系統的可解釋性
儘管基於機器學習的網路入侵偵測系統 (NIDS) 具有很高的偵測效能，但由於其黑盒子特性，安全專家在將其部署到實際環境中時仍猶豫不決。本研究使用可解釋人工智慧 (XAI) 技術分析了最先進的基於機器學習的線上 NIDS 模型，發現某些模型比其他模型更容易解釋，不同 XAI 方法對同一模型的解釋存在衝突，並且某些模型比其他模型更容易受到歸納偏差的影響。
최첨단 머신러닝 기반 온라인 네트워크 침입 탐지 시스템의 설명 가능성 평가
최첨단 머신러닝 기반 온라인 네트워크 침입 탐지 시스템은 높은 탐지 성능을 보이지만, 블랙박스 모델의 특성으로 인해 실제 환경 배포에 어려움을 겪고 있으며, 설명 가능한 AI(xAI) 기법을 활용하여 이러한 시스템의 의사 결정 과정을 분석하고 설명 가능성을 평가해야 한다.
最先端の機械学習ベースオンラインネットワーク侵入検知システムの説明可能性の評価
機械学習ベースのネットワーク侵入検知システム（NIDS）は高い検知性能と精度を実現する一方で、そのブラックボックスな性質のため、実際の運用環境への導入には躊躇がある。本稿では、説明可能なAI（xAI）技術を用いて、最先端の機械学習ベースのオンラインNIDSモデルを分析し、説明可能性、誘導バイアスへの脆弱性、xAI手法間の説明の一致性について評価する。
FRACTURED-SORRY-Bench: A Framework for Exposing Vulnerabilities in Large Language Model Safety Measures Using Multi-Turn Conversational Attacks
FRACTURED-SORRY-Bench reveals that current safety measures in Large Language Models (LLMs) are vulnerable to multi-turn conversational attacks, highlighting the need for more robust defenses.
Evaluating the Explainability and Inductive Bias of Machine Learning-Based Online Network Intrusion Detection Systems
While Machine Learning (ML) based Network Intrusion Detection Systems (NIDS) offer improved performance, their black-box nature hinders trust and deployment; this research leverages Explainable AI (xAI) techniques to analyze and compare the explainability and vulnerability to inductive bias of state-of-the-art ML-based online NIDS.
基於相關輸入擾動的差分隱私範圍查詢
本文提出了一種針對線性查詢（特別是範圍查詢）的差分隱私機制，該機制利用相關輸入擾動來同時實現無偏性、一致性、統計透明度，並根據以特定查詢邊界或分層數據庫結構所暗示的準確性目標來控制效用需求。
상관 입력 섭동을 사용한 차등 개인정보 보호 범위 쿼리
본 논문에서는 상관 관계가 있는 가우시안 노이즈를 활용한 입력 섭동 메커니즘을 통해 일관성, 통계적 투명성, 효율성을 유지하면서 개인정보 보호 범위 쿼리에서 높은 정확도를 달성하는 방법을 제시합니다.
相関入力摂動を用いた差分プライベートレンジクエリ
本稿では、ノイズの相関構造を活用することで、従来の入力摂動法の欠点を克服し、高い精度とデータ有用性を実現する、差分プライベートなレンジクエリのための新しい入力摂動メカニズムを提案する。
Differentially Private Range Queries with Correlated Input Perturbation: Achieving Unbiasedness, Consistency, and Utility Control
This paper introduces a novel differentially private mechanism for linear queries, particularly range queries, that leverages correlated input perturbation to achieve unbiasedness, consistency, statistical transparency, and fine-grained utility control, outperforming traditional methods in accuracy and efficiency.
© 2024 by Linnk AI