SpecTr-GBV: Multi-Draft Block Verification Accelerates Speculative Decoding
A new study proposes SpecTr-GBV, the first method to combine multi-draft strategies with block verification techniques, …
📂 Research
Latest articles in Research
Breaking Defenses Word by Word: ICD Jailbreak Strategy Exposes New LLM Security Vulnerabilities
Researchers propose Incremental Completion Decomposition (ICD), a jailbreak strategy that guides large language models t…
Evaluation Results Released for RAG-Powered Virtual Assistant in Undergraduate Project Supervision
A new study explores the effectiveness of a generative AI virtual assistant based on Retrieval-Augmented Generation (RAG…
Unsupervised Graph Neural Network Modeling: A New Framework for Accounting Account Anomaly Detection
A latest arXiv paper proposes an unsupervised discriminative framework based on graph neural networks that abstracts acc…
DORA System: Asynchronous Reinforcement Learning Breaks Through LLM Training Bottlenecks
A research team has proposed the DORA system, a scalable asynchronous reinforcement learning architecture that addresses…
New Research: Explainable Transformer Achieves Counterfactual Fairness
A latest arXiv paper proposes an efficient and explainable Transformer architecture designed to resolve the three-way te…
SWAN: Environment-Aware Adaptive Multimodal Networks
A research team has proposed the SWAN framework, enabling multimodal deep neural networks to dynamically adapt at runtim…
Entropy Centroids as Intrinsic Rewards: A New Paradigm for Test-Time Compute Scaling
A latest arXiv paper proposes the "Entropy Centroids" method, which scales LLM computation at test time without external…
New Causal Bandit Framework: Intelligent Decision-Making Under Budget Constraints
A latest arXiv paper proposes a Budget-Constrained Causal Bandit framework that bridges uplift modeling with sequential …
CNN, Transformer, and Mamba: Which Architecture Best Suits PPG-Based Emotion Recognition?
A new study systematically compares three deep learning architectures — CNN, Transformer, and Mamba — on photoplethysmog…
MomentumGNN: A New Graph Neural Network Architecture with Built-In Conservation of Momentum
Researchers propose MomentumGNN, a novel architecture that embeds momentum conservation constraints directly into graph …
AI Spatial Clustering Evaluates Heat Vulnerability in Rio's Favelas
Researchers propose a data-driven framework combining spatially constrained clustering with land surface temperature ana…