AboutTopCourses
← heapsort-ai

AI Research

146 items

RESEARCHarXiv CS.CL·12d ago

Bridging the Stability-Expressivity Gap: Synthetic Data Scaling and Preference Alignment for Low-Resource Spoken Language Models

This research addresses the Stability-Expressivity Gap in Spoken Language Models (SLMs) for low-resource languages, caused by the extensive use of synthetic data. While synthetic data improves phonetic accuracy, it degrades prosodic expressivity, a phenomenon termed Synthetic Erosion. The paper introduces self-alignment frameworks to recover expressivity.

synthetic dataspeech synthesisspoken language modelsLow-resource languages
27
RESEARCHarXiv CS.CL·4/6/2026

Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets

Este estudo argumenta, com base na Desigualdade de Processamento de Dados, que LLMs de agente único são mais eficientes em termos de informação do que sistemas multiagente sob orçamentos de token de raciocínio iguais. A pesquisa testa empiricamente esta previsão, que sugere que sistemas multiagente se tornam competitivos quando a utilização de contexto de um único agente é degradada ou mais poder computacional é despendido.

LLMsInformation EfficiencyComputational BudgetMulti-Hop Reasoning
27
RESEARCHarXiv CS.LG·5/6/2026

Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning

This survey provides an optimizer-agnostic view of rollout strategies for RL-based post-training of reasoning LLMs. It formalizes rollout pipelines with a unified notation and introduces the Generate-Filter-Control-Replay (GFCR) lifecycle taxonomy, decomposing pipelines into four modular stages.

Rollout Strategiesreinforcement learningmachine learningAI Research
27
RESEARCHarXiv CS.CL·4/7/2026

Evolutionary Search for Automated Design of Uncertainty Quantification Methods

Este artigo explora o uso de busca evolucionária impulsionada por LLMs para desenvolver automaticamente métodos de Quantificação de Incerteza (UQ) não supervisionados. Os métodos evoluídos superam baselines manuais em verificação de alegações, demonstrando generalização robusta e estratégias distintas entre diferentes modelos de LLM.

LLMsUncertainty QuantificationEvolutionary SearchAI Research
27
RESEARCHarXiv CS.CL·28d ago

jina-embeddings-v5-omni: Geometry-preserving Embeddings via Locked Aligned Towers

This work introduces GELATO, a novel approach to multimodal embedding models that extends VLM-style architectures. It results in the jina-embeddings-v5-omni suite, which efficiently encodes text, image, audio, and video into a single semantic embedding space by freezing backbone text models and training only connecting components.

embedding modelsmultimodal AIdeep learningmachine learning
27
RESEARCHarXiv CS.AI·19d ago

High Quality Embeddings for Horn Logic Reasoning

This paper introduces novel approaches for creating high-quality embeddings for logical statements, crucial for training neural networks to efficiently rank choices made by logical reasoners. These methods involve generating anchors with repeated terms, balancing easy, medium, and hard examples for triplet loss training, and periodically emphasizing the hardest examples.

neural networksLogic reasoningmachine learningembeddings
27
RESEARCHarXiv CS.AI·21d ago

Scalable Uncertainty Reasoning in Knowledge Graphs

This research proposes a modular framework to address scalable uncertainty reasoning in Knowledge Graphs, where real-world data often inherently contains uncertainty. It tackles three levels of uncertainty—imprecise attributes, probabilistic triple existence, and incomplete schema knowledge—through tailored techniques like probabilistic literals, probabilistic circuits, and geometric embeddings.

Uncertainty ReasoningKnowledge GraphsProbabilistic ModelsSemantic Web
27
RESEARCHarXiv CS.AI·19d ago

AgentCo-op: Retrieval-Based Synthesis of Interoperable Multi-Agent Workflows

AgentCo-op is a retrieval-based synthesis framework that composes interoperable multi-agent workflows from reusable skills, tools, and external agents. It applies bounded self-guided local repair to components upon execution failure and has been demonstrated in genomics case studies to coordinate specialized agents for collaborative discovery.

workflow automationGenomicsScientific Discoverymulti-agent systems
27
ARTICLEHugging Face (YouTube)·19d ago

On the slow death of Scaling (birth of Adaption Labs) | Sara Hooker | HF ML Club India EP2

This content explores the evolution of AI methodologies, discussing the decline of traditional scaling approaches and the emergence of new strategies, exemplified by the birth of Adaption Labs. Presented by Sara Hooker, the HF ML Club India episode delves into significant shifts within the field of artificial intelligence.

Adaption Labsmachine learningscalingAI Research
On the slow death of Scaling (birth of Adaption Labs) | Sara Hooker | HF ML Club India EP2
27