The paper proposes a continuous-depth formulation for Deep Gaussian Processes (DGPs), treating depth as a latent continuous evolution rather than a discrete stack. It introduces two instantiations—CDGP, which applies a GP prior directly to the state–depth drift, and FlowDGP, which uses a flow-evolved prior to capture richer depth dependence at lower training cost. Experiments on synthetic, benchmark, and large-scale regression tasks demonstrate that residual DGP improves over plain DGP, while CDGP and FlowDGP offer stronger performance across the studied regimes.
Lean Finder V2 is a semantic search framework for Mathlib that tackles the problem of short, ambiguous user queries by using adversarial training and query rewriting to disambiguate them. It achieves a new state‑of‑the‑art performance on the MathlibQR benchmark, improving retrieval by at least 11% over previous engines, and further boosts proof success when integrated with Claude Code.
TRIAD is a new benchmark designed to test omni‑modal large language models’ ability to resolve ambiguity across text, vision, and audio. Each item consists of a text–image–audio triplet, a question, and answer options, crafted so that all three modalities together yield a unique answer while any single modality leaves the answer ambiguous. The benchmark contains 327 hand‑written queries in 86 scene groups, covering 18 ambiguity categories, and shows that current models lag humans significantly in scene‑level disambiguation, especially when answer options are removed.
The paper introduces TESSA, a governance-oriented Custom GPT designed to aid small and medium-sized enterprises (SMEs) in complying with the European Union Artificial Intelligence Act (EU AI Act). Using a design science research approach, TESSA was evaluated against baseline ChatGPT across 100 benchmark scenarios, showing higher governance alignment rates (90% vs. 83%) and consistent performance in repeated executions. The study presents a human-validated regulatory update mechanism and a multidimensional evaluation framework for governance-oriented AI artifacts.
Training large language models for complex reasoning tasks relies on two complementary signals: sparse, outcome-grounded reinforcement learning (RL) for correctness, and dense, token-level distillatio...
Trace2World is a framework that reconstructs app worlds from real interaction traces, separating global behavior reconstruction from implementation. It reconciles traces into a shared, source‑linked app model and compiles an executable environment, allowing independent replay of source workflows to localize mismatches for bounded implementation repair. Using 1,570 traces, the system builds 50 environments and achieves 91.1% end‑to‑end reproduction on 200 workflows across ten apps, outperforming a strong baseline and improving benchmark success and external GUI benchmark scores.
PhyEvolve is a multi-agent framework that improves physics-aware video generation by enhancing natural-language prompts through structured state-evolution planning. It employs three agents—a Director, a Physical Evolver, and a Composer—to ground scenes, identify physical mechanisms, and construct causally linked state transitions that guide video generation. The method works with existing generators via prompt replacement and achieves state‑of‑the‑art performance, outperforming the prior agentic baseline by an average of 7.9 percentage points on four major benchmarks.
ULME introduces a post‑training framework that unifies text embedding and generation within a single causal decoder‑only LLM by inserting learnable slots for semantic and reasoning embeddings. It trains three objectives—contrastive, language‑modeling, and stop‑gradient predictive—so that generation actively shapes the embedding space. Across Qwen2.5 and Qwen3 models (0.5B–8B), ULME improves MTEB scores by 3.3–4.1 points, tightens BRIGHT and AdvBench‑IR, and matches generation‑only fine‑tunes on standard accuracy benchmarks.
The paper investigates whether scaling data, diversity, and model capacity benefits embodied agents in virtual worlds. Using 120 training setups across four model sizes, six data budgets, and five data mixtures, the authors find that more data improves in‑corpus performance, diversity aids transfer to related worlds, and larger models boost data efficiency but may generalize less to distant worlds. They also identify distinct training phases and fit a scaling law, concluding that continued pre‑training alone is insufficient for fully generalist agents.
By Kuan Zhang, Yukun Chen, Zhihao Yang, Yue Su, Xiangnan Wu, Zirong Chen, Run Luo, Jinkun Hou, Tao Tan, Yinhe Zheng, Yiming Li
OpenEarth-OVS44 is a new multimodal remote sensing dataset featuring 290,587 expert‑verified segments from 2,832 RGB images and 1,341 RGB‑SAR pairs at sub‑meter resolution, covering 40 cities in France, Japan, and the United States. It adopts a unified 44‑class taxonomy organized into nine base categories and offers three evaluation tracks: an RGB‑only fine‑grained benchmark, a multimodal track with synthetic cloudy conditions for weather‑robust evaluation, and a held‑out cross‑region track. Experiments show that training on OpenEarth-OVS44 improves fine‑grained segmentation performance, achieving up to 7.1 mIoU points on unseen categories and 27.6 mIoU points on seen categories compared to the strongest training‑free baseline.
By Yimin Wei, Hongruixuan Chen, Hao Zheng, Hao Liu, Jian Song, Yuting Zhu, Junshi Xia, Naoto Yokoya
Doob-Based Generative Policy Optimization (DoobPO) introduces a likelihood‑free framework for fine‑tuning generative policies such as diffusion and flow models. It first learns a lightweight ratio model to represent the desired policy update and then distills this ratio‑tilted target policy into the generative policy via conditional drift matching. The method supports standard policy improvement rules like PPO and GRPO, proves gradient equivalence to an ideal Doob drift matching objective, and demonstrates strong performance on MuJoCo continuous‑control benchmarks.
By Maojiang Su, Qijie Zhu, Yu Xiao, Shuyang Yu, Minshuo Chen, Zhaoran Wang, Han Liu
The paper introduces Probabilistic Latent Reasoning Models (PLRM), a parallelizable variational framework that moves reasoning from token‑by‑token language generation into a continuous latent space. PLRM uses a JEPA‑style objective to infer latent distributions in parallel and a mixture‑based prior with a discrete strategy variable to preserve multiple distinct reasoning trajectories. Experiments show PLRM matches deterministic latent models on single‑trajectory tasks and outperforms stochastic baselines on multi‑trajectory benchmarks while enabling efficient training.
By Seokhyun Lee, Seongjun Lee, Changhee Lee
The paper introduces a new benchmark for personalized shopping that focuses on determining how far an autonomous agent should commit to purchasing an item. It defines a three-valued, evidence‑conditioned boundary—‘don’t buy’, ‘confirm first’, or ‘buy automatically’—based on product information sufficiency, reversibility, price norm, and urgency. Experiments with instruction‑tuned models show that these agents often overreach, committing even when fit is low or evidence is missing, and that a shallow learner trained on a person’s own labels can predict their auto‑purchase line more accurately than any prompt‑model combination.
By Ryotaro Shimizu, Sai HtaungKham, Mo Fan
The paper investigates the monitorability of chain-of-thought (CoT) reasoning in large language models, showing that non-zero mutual information between CoT and output is necessary but not sufficient for effective monitoring. It identifies two approximation errors—information gap and elicitation error—that can degrade monitor performance, and demonstrates that CoT obfuscation can occur even without optimization pressure. The authors propose a label‑free mutual‑information‑based training objective that mitigates CoT obfuscation and reward hacking across multiple environments, outperforming an oracle‑based method.
By Usman Anwar, Tim Bakker, Dana Kianfar, Cristina Pinneri, Christos Louizos
The study fine‑tuned DINOv2‑base on 2,253 weighed Nutrition5k dishes to estimate food mass from a single RGB photo and compared it with the zero‑shot vision‑language model Qwen3‑VL‑32B across Nutrition5k and three external datasets. While the fine‑tuned model achieved a mean absolute error of 24.8 g on its own domain—better than Qwen—it failed to outperform Qwen on external datasets and its gram scale varied with scene coverage and differed per food type. Adding a single labeled photo of the same food reduced external error significantly, indicating that single‑image mass estimators require labeled photographs from the target setting.
By Yongsheng Lian
AgentPolicyForge is a framework that leverages executable environments to build policy benchmarks and iteratively train policy guards for task agents. It introduces PolicyGym, a dataset covering ten policy families with 698 training scenarios and 254 human‑reviewed test scenarios, where tool calls are executed and outcomes recorded to assess compliance. The framework supports an attacker‑generated request loop that fine‑tunes guards on balanced, environment‑verified examples, leading to a 40‑44% reduction in attack success rates compared to policy prompting alone.
By Chao Shang, Wanyu Du, Yi Zhang, Sandesh Swamy
The paper introduces Role-Enhanced Zero-Shot Chain-of-Thought (RE‑ZS‑CoT), a prompting framework that incorporates domain‑specific expert roles to improve reasoning in large language models without task‑specific examples. Two variants are evaluated: a static version that tests eight predefined roles across arithmetic, commonsense, and reading comprehension tasks, and a dynamic version where the model selects the most suitable role for each benchmark. Across 20 LLMs and seven benchmarks, RE‑ZS‑CoT consistently outperforms zero‑shot and zero‑shot‑CoT baselines, achieving up to 44 percentage‑point gains and demonstrating strong alignment between static and dynamic approaches, while ablation studies confirm the importance of task‑role alignment.
By Ibrahim Khan, Thai Van Nguyen, Yi Xia, Pratch Suntichaikul, Mustafa Can Gursesli, Daniel Fernandez Galeote, Juho Hamari, Ruck Thawonmas
ChronoSynth is a new framework for generating natural‑language questions from temporal knowledge graphs while preserving essential constraints such as answer field, time condition, reference event, and temporal order. It introduces a Temporal Intent State to make question semantics explicit, retrieves compatible typed patterns from a memory, adapts them to the input graph, and validates the output, using feedback to revise any inconsistencies. Experiments on CRONQUESTIONS and MultiTQ show that ChronoSynth achieves state‑of‑the‑art performance, with ablation studies confirming the effectiveness of its components.
By Feng Liang, Weixin Zeng, Runhao Zhao, Mingjun Zhou, Shaoshi Yang, Xiang Zhao
The paper introduces CORE, an optimization method that controls output rankings in large language model (LLM)-based search engines by appending strategically designed optimization content—string-based, reasoning-based, and review-based—to the retrieved content. CORE targets the black-box nature of LLM interactions with search engines, steering the ranking of product recommendations. Experiments on four LLMs (GPT‑4o, Gemini‑2.5, Claude‑4, and Grok‑3) using the new ProductBench benchmark show that CORE achieves high promotion success rates (91.4% @Top‑5, 86.6% @Top‑3, 80.3% @Top‑1) across 15 product categories, outperforming existing ranking manipulation methods while maintaining content fluency.
By Haibo Jin, Huimin Zeng, Yifeng Luo, Haohan Wang
HealthMamba is an uncertainty‑aware spatiotemporal graph state space model designed to predict healthcare facility visits more accurately and reliably. It integrates a Unified Spatiotemporal Context Encoder, a novel GraphMamba module for hierarchical modeling, and a comprehensive uncertainty quantification component. Evaluated on large‑scale datasets from California, New York, Texas, and Florida, it improves prediction accuracy by about 6.0% and uncertainty quantification by 3.5% over state‑of‑the‑art baselines.
By Dahai Yu, Lin Jiang, Rongchao Xu, Guang Wang