Technology – Hard Level – GK Questions

Anomaly detection uses: Supervised Learning (trained on labeled attacks), Unsupervised Learning (clustering normal vs abnormal patterns), and Reinforcement Learning (adaptive response to threats). Combined approaches improve detection of zero-day attacks and reduce false positives in SIEM systems.

AES-256 (Advanced Encryption Standard with 256-bit key) is approved by NIST for top-secret information. Provides strong security against brute-force attacks. DES/3DES are deprecated; RC4 has known vulnerabilities. AES is efficient in both hardware and software implementations.

Multimodal models (CLIP, LLaVA) process text, images, audio jointly, enabling visual question answering, image captioning, and cross-modal retrieval.

Code LLMs (Codex, CodeLlama) trained on code repositories enable natural language to code generation. Combined with fine-tuning and prompt engineering for developer productivity tools.

Event-driven architecture uses message brokers for asynchronous, decoupled service communication. Enables scalability, resilience, real-time processing.

Smart contracts are self-executing programs stored on blockchain that automatically enforce terms when conditions are satisfied. Used in DeFi, supply chain, insurance.

OWASP Application Security Verification Standard (ASVS) and ISO 27034 provide frameworks for secure software development, testing, and vulnerability management.

Active-active runs full workload in multiple regions simultaneously, enabling instant failover with zero RPO/RTO. Highest cost but maximum resilience. Critical for mission-critical systems.

High-dimensional handling: dimensionality reduction (PCA, autoencoders) compresses features; sparse representations (L1 regularization) select relevant features. Critical for text, recommendation systems.

Code LLMs (Codex, CodeLlama) trained on code repositories enable natural language to code generation, debugging, and documentation. Critical for developer productivity tools.

Digital Public Goods Alliance curates open-source solutions; G20 DPI Framework promotes interoperable identity, payments, data sharing. India leads global DPI advocacy.

Model-based imputation (MICE, Bayesian) models relationships between variables to predict missing values, accounting for uncertainty. More robust than simple methods for complex missingness patterns.

Event-driven architecture uses message brokers (Kafka, RabbitMQ) for decoupled, asynchronous service communication. Enables scalability, resilience, and real-time processing.

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

Effective tech policy: proportionate to actual risks, technology-neutral (focus on outcomes), risk-based (higher scrutiny for higher-risk activities). Critical for balanced policy design.

Imbalance handling: oversampling (SMOTE) increases minority, undersampling reduces majority, class weighting adjusts loss function. Choice depends on data size and model type.

Stateless microservices store session data externally (Redis, database), enabling horizontal scaling by adding instances. Critical for elastic, resilient cloud applications.

Advanced decoding: beam search explores multiple hypotheses, nucleus (top-p) sampling truncates low-probability tokens. Balance diversity and coherence in text generation.

India endorses OECD AI Principles (2019) and UNESCO Recommendation on AI Ethics (2021). Both emphasize human rights, transparency, accountability, sustainability. Critical for AI governance.

Time series methods: ARIMA for stationary series, Prophet for seasonality/holidays, LSTM for complex patterns. Choice depends on data characteristics and forecast horizon.