20 articles tagged with "Machine Learning"
A newly developed language model with approximately 9 million parameters aims to clarify the workings of larger models. Built using a vanilla transformer architecture, it utilizes synthetic conversation data.
Recent findings highlight the variability in clinical predictions from large language models, suggesting a need for adaptive deliberation strategies in healthcare applications.
A recent publication discusses the implementation of multi-step interaction loops in agentic applications, focusing on planning, action execution, and environment feedback.
The PAR$^2$-RAG framework introduces a method to enhance multi-hop question answering by addressing the limitations of large language models through improved retrieval and reasoning techniques.
Google has released a new time-series foundation model featuring 200 million parameters and a context length of 16,000, aimed at improving data analysis capabilities.
A new method, Process-Aware Policy Optimization (PAPO), enhances stability in AI training by integrating process-level evaluations into Group Relative Policy Optimization (GRPO).
A novel framework aims to enhance reasoning capabilities in Large Language Models (LLMs) while addressing computational efficiency and exploration depth trade-offs.
Advanced technologies like Natural Language Processing and Machine Learning are being employed to address food security challenges in regions with limited data availability.
A novel approach utilizing LLM-enhanced energy contrastive learning is introduced for detecting out-of-distribution instances in text-attributed graphs, relevant for network modeling.
A new open-source project demonstrates the feasibility of executing a 397 billion parameter AI model on Mac systems equipped with 48GB RAM, highlighting infrastructure implications.
A new machine learning model streamlines the identification of molecules by accurately predicting electric dipole moments, offering significant implications for research efficiency.
The latest research introduces Bayesian evolution and warm rotation strategies to enhance AI model performance, focusing on geometric and neuromorphic methodologies.
Modern language model-based AI systems exhibit significant capabilities, yet their potential is constrained by human design. Key limitations include model weights and training data.
Recent advancements in AI agents are facilitating automated scientific discovery, driven by improvements in synthetic task scaling that enhance research efficiency.
The InfoDensity approach targets the verbosity of reasoning in Large Language Models, aiming to optimize computational efficiency and reasoning capabilities.
A new method aims to improve the reasoning capabilities of diffusion large language models (dLLMs) by addressing coordination issues in multi-step tasks.
The integration of machine learning in marine diesel engine maintenance could significantly enhance early detection of catastrophic failures, preventing severe operational disruptions.
This paper discusses a novel approach to model editing for Large Language Models, focusing on semantic routing for efficient updates and improved modular isolation.
The HEAL framework introduces a novel approach to distilling reasoning capabilities from larger models into smaller, more efficient ones, addressing traditional sampling limitations.
A recent study examines the trade-offs between accuracy and cost in Agentic Retrieval-Augmented Generation systems, emphasizing the need for effective budget management.