Technology – Medium Level – GK Questions

XAI methods (SHAP, LIME, attention visualization) provide interpretable explanations for AI decisions. Critical for regulated sectors (finance, healthcare) and user acceptance.

Scatter plots display individual data points for two variables, revealing correlations, clusters, and outliers. Foundation for regression analysis and exploratory data science.

F1 Score = 2 * (Precision * Recall) / (Precision + Recall). Harmonic mean balancing false positives and false negatives. Critical for evaluating models on imbalanced datasets.

Data splitting strategies: holdout (simple split), cross-validation (multiple folds), bootstrap (resampling). Critical for reliable model evaluation and generalization assessment.

Serverless (Lambda, Cloud Functions) uses pay-per-execution pricing, auto-scaling from zero. Trade-offs: cold starts, execution limits, vendor lock-in. Critical for cost-optimized cloud architectures.

Multi-AZ deployment distributes resources across physically separate data centers within a region, protecting against zone failures. Critical for business continuity and SLA compliance.

CaaS (Container as a Service) provides managed Kubernetes (EKS, AKS, GKE) for container deployment, scaling, and management. Abstracts infrastructure while retaining control. Critical for cloud-native development.

Deception deploys fake assets (servers, credentials) to lure attackers, enabling early detection and threat intelligence. Critical for proactive defense against advanced adversaries.

MITRE ATT&CK matrix catalogs adversary TTPs across enterprise, cloud, mobile. Enables threat-informed defense, detection engineering, and red teaming. Critical for mature security operations.

Anomaly detection establishes baselines of normal behavior and flags deviations, enabling zero-day threat detection. Critical for advanced threat protection in modern SOCs.

Feature stores (Feast, Tecton) manage feature engineering, versioning, and serving, preventing training-serving skew. Critical for reliable ML system performance in production.

Responsible MLOps: model cards document limitations, fairness audits detect bias across groups, A/B testing validates real-world impact. Critical for ethical AI deployment.

Prompt injection exploits LLM instruction-following to bypass safety guards, extract training data, or execute unauthorized actions. Critical vulnerability for LLM applications requiring robust input sanitization.

Reinforcement Learning from Human Feedback (RLHF) trains reward models from human rankings, then optimizes LLM to maximize rewards. Critical for aligning AI with human values and safety.

GPAI (founded 2020) brings together experts for responsible AI: research, policy, best practices. India is founding member. Critical for understanding global AI governance landscape.

Digital diplomacy: cyber norms, data governance, tech partnerships, capacity building. India actively engages in UN, GPAI, OECD on AI ethics, cybersecurity, digital trade. Critical for foreign policy questions.

Telecommunications Standards Development Society, India (TSDSI) develops indigenous telecom standards, represents India in 3GPP, ITU. Critical for technology sovereignty and 5G/6G leadership.

India Stack layers: identity (Aadhaar), payments (UPI), data sharing (AA), documents (DigiLocker). Enables innovation on public digital infrastructure. Critical for digital governance questions.

Responsible data science: fairness audits for bias, privacy-preserving techniques, impact assessments. Critical for ethical AI deployment and regulatory compliance (DPDP Act).

Principal Component Analysis (PCA) transforms features into orthogonal components ordered by variance explained. Enables visualization, noise reduction, and faster model training.