Autonomous agents, tool use, and agentic systems
![[2602.00185] QUASAR: A Universal Autonomous System for Atomistic Simulation and a Benchmark of Its Capabilities](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
Abstract page for arXiv paper 2602.00185: QUASAR: A Universal Autonomous System for Atomistic Simulation and a Benchmark of Its Capabilities
Abstract page for arXiv paper 2506.22653: URSA: The Universal Research and Scientific Agent
Abstract page for arXiv paper 2505.00472: UserCentrix: An Agentic Memory-augmented AI Framework for Smart Spaces
The paper introduces StarEmbed, a benchmark for evaluating time series foundation models on astronomical observations of variable stars, ...
This paper explores multilingual routing in Mixture-of-Experts (MoE) architectures, revealing how these models handle multilingual data a...
The paper presents adaptive exploration strategies for latent-state bandits, addressing challenges in reward estimation and action select...
COGITAO introduces a novel framework for studying compositionality and generalization in visual reasoning, offering extensive task genera...
The paper introduces MedReasoner, a framework that utilizes reinforcement learning for precise medical reasoning and pixel-level groundin...
This paper explores Mixture-of-Experts (MoE) architectures through a geometric lens, analyzing their impact on function representation an...
This paper explores the expressive power of graph transformers, comparing their capabilities under different logical frameworks, particul...
The paper presents FreqPolicy, a novel flow-based visuomotor policy that enhances efficiency in robotic manipulation by imposing frequenc...
This paper explores how high entropy regularization in Dec-POMDPs leads to symmetry equivariant policies, ensuring convergence to a consi...
This article explores modeling human behavior in strategic network games, focusing on the Junior High Game (JHG) and comparing various be...
This article presents a novel approach to reducing variance in reinforcement learning through shrinkage baselines, enhancing training sta...
The paper presents FindAnything, a framework for open-vocabulary and object-centric mapping that enhances robot exploration in unknown en...
The paper presents Cluster-PFN, a Transformer-based model for unsupervised Bayesian clustering, demonstrating improved accuracy and speed...
The paper explores how Transformers can learn algorithmic solutions for graph connectivity, demonstrating that success depends on the tra...
This article presents a novel approach combining Chain-of-Thought (CoT) and Retrieval Augmented Generation (RAG) to improve rare disease ...
This survey reviews advancements in spatiotemporal consistency in video generation, addressing challenges and methodologies in creating c...
The paper introduces $ ext{Pinet}$, a novel output layer for neural networks that optimizes hard constraints using orthogonal projection ...
The paper presents Cocoa, a system designed to enhance human-agent collaboration in AI tasks by allowing flexible co-planning and co-exec...
This paper introduces a novel Positional Recovery Training (Port) framework for improving temporal grounding in animal behavior analysis,...
This paper introduces a novel method for evaluating language model agency through negotiation games, addressing limitations of existing b...
Get the latest news, tools, and insights delivered to your inbox.
Daily or weekly digest • Unsubscribe anytime