利用人工智能解锁网络洞察力.pdf

1、 2025 Expel,Inc.Unlocking cyber insights with AI:Using ML and LLMs for next-gen analysisXenia Mountrouidou,PhDPrincipal Cyber Data Scientist|ExpelSANS AI Cyber SummitMarch 31-April 1,2025 2025 Expel,Inc.AI for cybersecurityHow can we help security practitioners with AI?EDA with LLMsHow can we apply

2、AI to explore cybersecurity data?Classification with LLMsDoes AI help in predicting malicious behavior in cybersecurity data?Agenda 2025 Expel,Inc.AI for cybersecurityEDA with LLMsClassification with LLMsAgenda 2025 Expel,Inc.Predict or Generate?AI in CybersecurityData&Applications 2025 Expel,Inc.Ty

3、pesDataSources 2025 Expel,Inc.Predict the futureMake decisionsAutomateTestApplicationsRecognize patterns 2025 Expel,Inc.Predict or Generate?Predictive models learn decision Predictive models learn decision boundaries.boundaries.Traditional ML models,pre LLMsCannot generate new dataSpecifically meant

4、 for classification,predictionPredictive models do not possess generative propertiesGenerative models learn the input Generative models learn the input distributiondistributionLLMs and Gen AI multi-modal modelsTypically not used for classification,prediction Can generate new dataGenerative models po

5、ssess discriminative properties 2025 Expel,Inc.Exploring where no man has gone beforeEDA with AI 2025 Expel,Inc.Exploratory Data Analysis(EDA)VisualizationsVisualizationsStatisticsStatistics 2025 Expel,Inc.Why EDA?Data is messyData is messyit may be a complete it may be a complete black box.black bo

6、x.Know your data better:Know your data better:Feature engineeringModelingQuestions about data:Questions about data:Is it accessible?Is it usable?EDAIs it understandable?EDAIs it sizeable?EDAIs it reliable?2025 Expel,Inc.Why EDA for cybersecurity data?Cybersecurity data is messyCybersecurity data is