Large Language Models (LLMs) have demonstrated remarkable progress in instruction following and general-purpose reasoning. However, achieving high-quality alignment with human intent and safety norms without human annotations remains a fundamental challenge. In this work, we propose an Uncertainty-Driven Adaptive Self-Alignment (UDASA) framework designed to improve LLM alignment in a fully automated manner. UDASA first generates multiple responses for each input and quantifies output uncertainty across three dimensions: semantics, factuality, and value alignment. Based on these uncertainty scores, the framework constructs preference pairs and categorizes training samples into three stages, conservative, moderate, and exploratory, according to their uncertainty difference. The model is then optimized progressively across these stages. In addition, we conduct a series of preliminary studies to validate the core design assumptions and provide strong empirical motivation for the proposed framework. Experimental results show that UDASA outperforms existing alignment methods across multiple tasks, including harmlessness, helpfulness, truthfulness, and controlled sentiment generation, significantly improving model performance.
Deliberative Alignment is Deep, but Uncertainty Remains: Inference time safety improvement in reasoning via attribution of unsafe behavior to base model
Phenomenological AI Safety Research Institute（现象学人工智能安全研究所）
The AGI alignment problem has a bimodal distribution of outcomes with most outcomes clustering around the poles of total success and existential, catastrophic failure. Consequently, attempts to solve AGI alignment should, all else equal, prefer false negatives (ignoring research programs that would have been successful) to false positives (pursuing research programs that will unexpectedly fail). Thus, we propose adopting a policy of responding to points of philosophical and practical uncertainty associated with the alignment problem by limiting and choosing necessary assumptions to reduce the …
University of California, Los Angeles（加利福尼亚大学洛杉矶分校）；Amazon.com Inc（亚马逊公司）
Predicting missing facts in a knowledge graph (KG) is crucial as modern KGs are far from complete. Due to labor-intensive human labeling, this phenomenon deteriorates when handling knowledge represented in various languages. In this paper, we explore multilingual KG completion, which leverages limited seed alignment as a bridge, to embrace the collective knowledge from multiple languages. However, language alignment used in prior works is still not fully exploited: (1) alignment pairs are treated equally to maximally push parallel entities to be close, which ignores KG capacity inconsistency; …
CLUE: Neural Networks Calibration via Learning Uncertainty-Error alignment
Software and Societal Systems Department, Carnegie Mellon University（卡内基梅隆大学软件与社会系统部门）；INESC-ID and Instituto Superior Técnico, Universidade de Lisboa（里斯本大学INESC-ID和理工学院）
School of Engineering and Built Environment, Electrical and Electronic Engineering, Griffith University（工程与建筑环境学院，电气与电子工程学院，格里菲斯大学）；School of Computer Science and Engineering, University of New South Wales（计算机科学与工程学院，新南威尔士大学）
National University of Defense Technology（国防科技大学）；The University of New South Wales（新南威尔士大学）；University of Amsterdam（阿姆斯特丹大学）
Entity alignment (EA) is the task of identifying the entities that refer to the same real-world object but are located in different knowledge graphs (KGs). For entities to be aligned, existing EA solutions treat them separately and generate alignment results as ranked lists of entities on the other side. Nevertheless, this decision-making paradigm fails to take into account the interdependence among entities. Although some recent efforts mitigate this issue by imposing the 1-to-1 constraint on the alignment process, they still cannot adequately model the underlying interdependence and the resu…
Prototype-Based Semantic Consistency Alignment for Domain Adaptive Retrieval
School of Computer Science and Technology, Guangdong University of Technology（广东工业大学计算机科学与技术学院）；School of Automation, Guangdong University of Technology（广东工业大学自动化学院）；School of Computer Science, Guangdong Polytechnic Normal University（广东 polytechnic 正规大学计算机科学学院）；School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen（哈尔滨工业大学深圳校区计算机科学与技术学院）；School of Artificial Intelligence, Guangzhou University（广州大学人工智能学院）
Bridge 2D-3D: Uncertainty-aware Hierarchical Registration Network with Domain Alignment
University of Science and Technology of China（中国科学技术大学）；School of Information Science and Technology（信息科学技术学院）；Deep Space Exploration Laboratory（深空探测实验室）
Planning under model uncertainty is a fundamental problem across many applications of decision making and learning. In this paper, we propose the Robust Adaptive Monte Carlo Planning (RAMCP) algorithm, which allows computation of risk-sensitive Bayes-adaptive policies that optimally trade off exploration, exploitation, and robustness. RAMCP formulates the risk-sensitive planning problem as a two-player zero-sum game, in which an adversary perturbs the agent's belief over the models. We introduce two versions of the RAMCP algorithm. The first, RAMCP-F, converges to an optimal risk-sensitive pol…
Universal Domain Adaptive Object Detection via Dual Probabilistic Alignment
CAIR, HKISI-CAS（中国科学院香港创新研究院人工智能与机器人创新研究所）；MAIS, Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所模式识别与智能系统研究部）；The Chinese University of Hong Kong（香港中文大学）
National University of Defense Technology（中国人民解放军国防科技大学）；The University of New South Wales（新南威尔士大学）；Collaborative Innovation Center of Geospatial Technology（地理空间技术协同创新中心）
Entity alignment (EA) identifies entities that refer to the same real-world object but locate in different knowledge graphs (KGs), and has been harnessed for KG construction and integration. When generating EA results, current solutions treat entities independently and fail to take into account the interdependence between entities. To fill this gap, we propose a collective EA framework. We first employ three representative features, i.e., structural, semantic and string signals, which are adapted to capture different aspects of the similarity between entities in heterogeneous KGs. In order to …
In order for reinforcement learning techniques to be useful in real-world decision making processes, they must be able to produce robust performance from limited data. Deep policy optimization methods have achieved impressive results on complex tasks, but their real-world adoption remains limited because they often require significant amounts of data to succeed. When combined with small sample sizes, these methods can result in unstable learning due to their reliance on high-dimensional sample-based estimates. In this work, we develop techniques to control the uncertainty introduced by these e…
