Mitigating Deanonymization and DoS Attacks in Decentralized Networks

paper

Anonymous communication networks like Tor face critical security vulnerabilities from Sybil and Cellflood attacks. 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.

Improving Document Summarization Through Advanced Language Techniques: Fine tuning LLMs, and Retrieval-Augmented Generation

paper

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, while a RAG pipeline utilizing Neo4j knowledge graphs and FAISS vector stores provides dynamic factual grounding. The integrated system successfully mitigates hallucination while maintaining contextual fidelity.

Vision-Language Integration in LLMs: A Survey of Architectures, Training Paradigms, and Applications

paper

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 architectural design patterns, training paradigms, and evaluation methodologies, identifying critical design choices and performance tradeoffs.

Dynamic Resource Aware Task Scheduling for Mobile Edge Cloud Computing

paper

Mobile edge computing introduces an intermediate computational tier between mobile devices and cloud infrastructure. This work extends existing frameworks with a three-tier architecture incorporating dynamic resource models for battery state, workload-dependent power consumption, and time-varying network conditions. The two-phase scheduling approach applies HEFT-based algorithms followed by energy optimization via heuristic task migration or Q-learning.