Aug 2026· Frontiers in Artificial Intelligence· 0 citations· 53 references
TL;DR
An adaptive threshold selection policy that chooses thresholds on validation data using pseudo-label precision and sample count is introduced and is combined with confidence-aware verifier training to support confidence-based selection of pseudo-labeled subsets.
Abstract
This paper studies how to improve the reasoning ability of large language models (LLMs) with minimal supervision. Recent gains in LLM reasoning largely come from learning intermediate reasoning traces, and many methods reduce supervision cost by using traces whose final answers are correct. In realistic settings, however, obtaining even those answer labels can be costly, motivating methods that extend reasoning from a very small labeled set with abundant unlabeled questions. Verifier-based semi-supervised learning is a promising approach: a verifier trained on a small labeled set can score reasoning traces on unlabeled questions and identify candidates for pseudo-labeling. However, even with a verifier, it remains unclear how pseudo-labeled samples should be selected to support downstream training. In particular, pseudo-label selection must balance quality and quantity. To address this, we introduce an adaptive threshold selection policy that chooses thresholds on validation data using pseudo-label precision and sample count. We further combine this policy with confidence-aware verifier training to support confidence-based selection. Experiments on verifiable math reasoning benchmarks show that, under our training setup, this combination improves downstream reasoning accuracy over the tested baselines and selects pseudo-labeled subsets with a more favorable reliability–coverage trade-off. These results suggest a practical design direction for verifier-guided pseudo-label selection in answer-verifiable, minimal-label reasoning settings.
FLARE is proposed, a novel framework that endows VLAs with robust error recovery capabilities through a ``Retry" and ``Reset" Paradigm, and significantly improves task success and robustness.
Ganlong Zhao, Zijia Tang, Xingping Chen et al.· 3 citations
LifeSciBench is introduced, a benchmark of 750 expert-authored tasks designed to evaluate whether language models can handle realistic life science research work, with each constituent task paired with a human expert-written rubric.
Amelia Liu, Andrew Ho, Anne Marie Droste et al.· bioRxiv· 2 citations
General-purpose language models can reason and synthesize knowledge, but complex work also requires sustained interaction with files, information sources, and executable code, together with state maintenance, failure recovery, and verifiable delivery. We call this \emph{working capability}: sustained, verifiable progress toward a real-world objective. Apodex 1.1 develops this capability along two complementary dimensions. \emph{Environment Scaling} expands the diversity and verifiability of executable file, search, and code environments, while \emph{Agentic Coordination Scaling} trains agents to decompose long-horizon tasks, delegate parallel work, integrate asynchronous results, and replan. A shared execution harness and AgentOS maintain task state and provenance across tools and agents, and training turns environment trajectories and coordination traces into reliable behavior. Across complex professional work, finance, scientific research, mathematics, coding, and search, Apodex 1.1 reaches the leading performance band despite using a substantially smaller model than many frontier systems. The 35B-parameter Apodex 1.1 Mini further retains strong working capability in a locally deployable form. These results ground agentic intelligence in useful, verifiable work completed over time and advance our goal of building a \emph{Heavy-Duty Solver} for ambitious, long-running tasks.
Apodex Team B. An, B. Li, B. Wang et al.· 1 citation
The complex multi-energy coupling characteristics inherent to integrated energy system (IES) present unprecedented challenges for the implementation of low-carbon scheduling. Existing optimization methods often exhibit limitations in system scalability, algorithm adaptivity, and carbon reduction efficacy for complex IES. This paper proposes a Large Language Model (LLM)-Embedded Multi-Agent Reinforcement Learning (LEMARL) to address the aforementioned issues. The proposed method integrates the global perception capability of LLMs with the dynamic optimization capability of MARL. Specifically, the LLM-Embedded module generates high-quality reward functions and policy frameworks from a global perspective, while the MARL module leverages these LLM-generated strategies for distributed interactive iterations—greatly enhancing computation efficiency and scalability. Simulation results demonstrate that LEMARL reduces carbon emissions by 7.76% and simultaneously decreases operating costs by 4.49% in a small-scale IES. Furthermore, LEMARL also exhibits superior applicability and scalability in large-scale IES of the IEEE 141-bus power grid integrated with 51-node thermal system.
Chen Xia, Tong Gou, Yinliang Xu et al.· IEEE Transactions on Smart G...· 1 citation
Experimental results show that CritICL consistently outperforms standard in-context learning and achieves performance competitive with or superior to test-time scaling methods, while requiring significantly fewer generations and lower token cost.
Yu-Fan Wu, Yinghui He, Zhengyi Hu et al.· 1 citation
TestifAI, a deep learning testing framework for efficient and accurate estimation of robustness against combinations of perturbations, is proposed and partial model tomography is introduced, a novel approach to reconstructing model behaviour in a multi-perturbation space from tests that apply only a small number of perturbations.
Arooj Arif, T. Hartung, E. Botoeva et al.· 1 citation
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
