AI Research

Researchers introduced a completeness estimation approach for multimodal sentiment analysis that quantifies how much sentiment-relevant information survives in incomplete data and uses that estimate to guide reconstructi…

A new arXiv paper (2609.10996v1) reports that verbalized confidence has become the more robust soft-scoring mechanism for LLM-as-a-Judge on post-2025 proprietary models, reversing the standard advice to prefer log-probab…

A study of five realistic noise conditions applied to 3,822 stereotype-related responses found that surface noise in text — typos, informal spelling, and broken punctuation — makes large language model judges far more li…

Researchers introduced CityPlanner, a sandbox-agent framework for executable urban planning, detailed in arXiv paper 2609.09578v1. CityPlanner uses UrbanSandbox, a unified file-based environment where agents inspect task…

Researchers released OpenDiscoveryTrace, a public dataset of 558 complete AI scientific agent trajectories across 124 scientific tasks in drug discovery, materials science, genomics, and literature analysis, spanning sev…

A new arXiv paper (2609.09233v1) reports that executing reusable agent skill packages as subagents outperforms loading skill instructions directly into a main agent's context window on long-horizon tasks. The authors fin…

Researchers introduced Valerant, a training-free framework that converts a pretrained action-conditioned world model into a World Action Model (WAM) for exploring and constructing persistent 3D game maps from a single im…

A new arXiv paper (2609.09448v1) introduces two methods, Latent Trajectory Dynamics (LTD) and the Action Representation Probe (ARP), that use a model's internal representations to predict task success in multi-turn agent…

A paper posted to arXiv (2609.09306v1) introduces "Gradland," an idealized world of neural networks whose physics are known and functions mostly differentiable, to test whether the first-order structure of physical inter…

Researchers introduced a human-in-the-loop, multi-agent GeoAI system for Arctic eco-navigation, detailed in arXiv paper 2609.09374v1, that integrates operational, physical, ecological, and community criteria into a unifi…

A new arXiv paper (2609.09589v1) develops a statistical-mechanical description of deep neural network learning directly in function space, treating parameter configurations as microscopic realizations and functions with …

A new arXiv paper (2609.09627v1) identifies seven non-exclusive sources of Physical AI capability formation: Recorded-Experience, Predictive-Modeling, Evaluative-Interaction, Surrogate-Environment, Mechanism-Grounded, Em…

A new arXiv paper (2609.09647v1) presents a black-box red teaming framework that found 56.25% average governance risk, 65% privacy risk in multi-agent configurations, and agent behavior vulnerabilities reaching 85% acros…

A paper posted to arXiv (2609.10589v1) introduces Halo, a modification that reuses an existing deep forecaster's architecture to add a second output estimating the scale of its implied distribution and trains it under th…

A new arXiv paper (arXiv:2609.10776v1) presents preliminary work showing that a Bellman optimality equation exists for optimizing plasticity within Markov decision processes, building on Abel et al. (2025), which defined…

A systematic literature mapping published on arXiv reviewed 96 articles from 2015 onward on AI and deep learning for lung cancer detection in radiology, finding convolutional neural networks with transfer learning and da…

Researchers introduced GeoSteer, an optimization-based method for norm-preserving activation steering in large language models that formulates steering as a Riemannian optimization problem and updates activations through…

Researchers proposed the Adaptive Margin Ordinal Loss (AMOL), a multiplicative per-class weighting scheme that suppresses "center-class hedging" in ordinal classification, according to an arXiv paper (arXiv:2609.10752v1)…

Researchers designed and implemented DR-LabStack, a React-Flask web system that integrates four externally developed pretrained diabetic retinopathy prediction models: RuleFit, Pruned RuleFit, Elaborative XGBoost, and Tw…

Researchers proposed counterfactual (CF) marginalisation, a test-time evaluation procedure for assessing how robust classification models are to nuisance variables such as age or sex, according to a paper published as ar…