Entity grounding, a crucial task in constructing multimodal knowledge graphs, aims to align entities from knowledge graphs with their corresponding images.Unlike conventional visual grounding tasks that use referring expressions (REs) as inputs, entity grounding relies solely on entity names and types, presenting a significant challenge.To address this, we introduce a novel Iterative Self-Refinement (ISR) scheme to enhance the multimodal large language model's capability to generate high quality REs for the given entities as explicit contextual clues.This training scheme, inspired by human learning dynamics and human annotation processes, enables the MLLM to iteratively generate and refine REs by learning from successes and failures, guided by outcome rewards from a visual grounding model.This iterative cycle of self-refinement avoids overfitting to fixed annotations and fosters continued improvement in referring expression generation.Extensive experiments demonstrate that our methods surpasses other methods in entity grounding, highlighting its effectiveness, robustness and potential for broader applications 1 .

Abstract With the increasing uncertainty and complexity of industrial production, dynamic events are inevitably happened during the production process. This paper deals with the transportation time constrained dynamic flexible job shop scheduling problem with machine breakdown, new jobs arrival, jobs cancellation, and processing time change of operations. Considering the actual production situation, a makespan rescheduling model based on transportation time is established. A Monte‐Carlo tree search based algorithm is proposed to solve the dynamic flexible job shop scheduling problem such that the makespan is minimized. In order to improve the accuracy of the developed algorithm and the quality of the obtained scheduling schemes, subtree keeping policy, multi‐branching simulation, reinforcement learning, greedy strategy, and expectation‐best evaluation method are employed. Moreover, a subtree pruning policy is embedded to improve the search efficiency of the developed algorithm. Simulation experiments are conducted on a series of instances of different sizes to compare the developed approach with the GA‐based scheduling approach and the Monte‐Carlo tree search based approach. The simulation results indicate that the developed approach is superior to the existing scheduling approaches in terms of the qualities of the obtained solutions.