While Vision-Language Models (VLMs) have excelled in 2D understanding and shown promise in 3D domains, their spatial reasoning is often hindered by incomplete visual observations. Since active physical exploration to gather missing information is usually impractical, leveraging visual generation to imagine the unobserved views provides a more feasible and efficient solution. We propose AIVE, an efficient active imagined-view exploration framework for spatial Visual Question Answering (VQA) that learns to directly select informative exploration actions, thereby amortizing costly test-time search and scoring. Departing from prior approaches that suffer from substantial inference latency, AIVE employs two key components: a VLM Planner generates targeted exploration strategies, while a generative World Model Dreamer synthesizes 3D-consistent future views accordingly. To facilitate training and evaluation, we introduce SpaThor, a benchmark comprising expert trajectories and action-conditioned visual transitions. Extensive experiments demonstrate that AIVE achieves state-of-the-art performance on challenging spatial reasoning tasks, outperforming prior baselines by an average margin of 5.2%. Notably, it delivers up to a 4.8× inference speedup over imagination-based methods, highlighting its potential for embodied spatial reasoning.

Large language models (LLMs) face persistent challenges when handling long-context tasks, most notably the “lost in the middle” issue, where information located in the middle of a long input tends to be underutilized. Some existing methods that reduce input have the risk of discarding key information, while others that extend context windows often lead to attention dispersion. To address these limitations, we propose Tree of Agents (TOA), a multi-agent reasoning framework that segments the input into chunks processed by independent agents. Each agent generates its local cognition, then agents dynamically exchange information for collaborative reasoning along tree-structured paths. TOA enables agents to probe different reasoning orders for multi-perspective understanding, effectively mitigating position bias and reducing hallucinations. To improve processing efficiency, we incorporate prefix-hash caching and adaptive pruning strategies, achieving significant performance improvements with comparable API overhead. Experiments show that TOA, powered by compact LLaMA3.1-8B, significantly outperforms multiple baselines and demonstrates comparable performance to the latest and much larger commercial models, such as Gemini1.5-pro, on various long-context tasks. Code is available at https://github.com/Aireduce952/Tree-of-Agents.

It is generally believed that a translation memory (TM) should be beneficial for machine translation tasks. Unfortunately, existing wisdom demonstrates the superiority of TM-based neural machine translation (NMT) only on the TM-specialized translation tasks rather than general tasks, with a non-negligible computational overhead. In this paper, we propose a fast and accurate approach to TM-based NMT within the Transformer framework: the model architecture is simple and employs a single bilingual sentence as its TM, leading to efficient training and inference; and its parameters are effectively optimized through a novel training criterion. Extensive experiments on six TM-specialized tasks show that the proposed approach substantially surpasses several strong baselines that use multiple TMs, in terms of BLEU and running time. In particular, the proposed approach also advances the strong baselines on two general tasks (WMT news Zh->En and En->De).