Security
Network Security
Distributed Systems
Anonymous communication networks like Tor face critical security vulnerabilities from Sybil and Cellflood attacks. Sybil attacks exploit decentralized trust models by deploying multiple malicious nodes to gain disproportionate network influence, while Cellflood attacks leverage the computational asymmetry between encryption and decryption to overwhelm relay nodes with circuit creation requests. A multi-layered defense framework is proposed incorporating enhanced Directory Authority monitoring, IP-based admission control, and client-side cryptographic puzzles. This approach achieves significant resistance to both attack vectors while preserving the network's fundamental anonymity guarantees and decentralized architecture. Experimental analysis demonstrates that these countermeasures maintain acceptable performance overhead while substantially increasing attack costs.
NLP
Machine Learning
LLMs
RAG
Document summarization systems face significant challenges from data scarcity, computational constraints, and hallucination in generated outputs. This work presents a hybrid architecture combining parameter-efficient fine-tuning of large language models with retrieval-augmented generation for scholarly document summarization. The approach applies LoRA and PEFT to Mistral 7B, achieving effective adaptation with minimal training data, while a RAG pipeline utilizing Neo4j knowledge graphs and FAISS vector stores provides dynamic factual grounding. Comparative evaluation against bidirectional LSTM baselines demonstrates superior performance in coherence, factual accuracy, and cross-domain generalization. The integrated system successfully mitigates hallucination while maintaining contextual fidelity, demonstrating the viability of modern LLM-based approaches in resource-constrained academic settings.
Computer Vision
NLP
Multimodal AI
Survey
This work presents a comprehensive survey of vision-language integration in large language models, tracing the architectural evolution from early dual-encoder systems to contemporary unified frameworks. The survey systematically analyzes key developments across three dimensions: architectural design patterns for cross-modal alignment, training paradigms including contrastive learning and generative pre-training, and evaluation methodologies for vision-language tasks. Representative models spanning the progression from CLIP and ALIGN to recent instruction-tuned multimodal LLMs are examined, identifying critical design choices and performance tradeoffs. The survey synthesizes current understanding of effective vision-language integration and highlights open challenges in scaling, grounding, and compositional reasoning.
Bayesian Inference
Probabilistic Modeling
Safety Systems
Bird strikes pose significant risks to aviation safety, yet existing assessment frameworks fail to adequately quantify uncertainty in risk predictions. This work presents a Bayesian simulation framework that propagates uncertainty from raw observations through to trajectory-level risk assessments. The system combines Kalman filtering with seasonal behavioral priors for bird tracking, Beta-Binomial conjugate models with Monte Carlo sampling for parameter estimation, and Gaussian Process regression with composite kernels for spatial risk modeling. The framework generates posterior predictive distributions enabling exceedance probability calculations along flight corridors. Simulation results demonstrate effective risk differentiation across trajectories with quantified uncertainty bounds that decrease as observational evidence accumulates. This probabilistic approach provides rigorous foundations for uncertainty-aware decision support in aviation safety applications.
Distributed Systems
Mobile Computing
Optimization
Mobile edge computing introduces an intermediate computational tier between mobile devices and cloud infrastructure, enabling new tradeoffs in task scheduling for energy-constrained applications. This work extends existing mobile cloud computing frameworks with a three-tier architecture incorporating dynamic resource models for battery state, workload-dependent power consumption, and time-varying network conditions. Applications are represented as DAGs with tasks characterized by computational and communication requirements. The two-phase scheduling approach applies HEFT-based algorithms for makespan minimization, followed by energy optimization via heuristic task migration or Q-learning. Experimental evaluation demonstrates that the three-tier architecture achieves significant energy reduction compared to traditional mobile-cloud systems while satisfying deadline constraints across diverse device states and network conditions.