Real world data often lie on low-dimensional Riemannian manifolds embedded in high-dimensional spaces. This motivates learning degenerate normalizing flows that map between the ambient space and a low-dimensional latent space. However, if the manifold has a non-trivial topology, it can never be correctly learned using a single flow. Instead multiple flows must be `glued together'. In this paper, we first propose the general training scheme for learning such a collection of flows, and secondly we develop the first numerical algorithms for computing geodesics on such manifolds. Empirically, we demonstrate that this leads to highly significant improvements in topology estimation.

Endowing dialogue agents with persona information has proven to significantly improve the consistency and diversity of their generations. While much focus has been placed on aligning dialogues with provided personas, the adaptation to the interlocutor's profile remains largely underexplored. In this work, we investigate three key aspects: (1) a model's ability to align responses with both the provided persona and the interlocutor's; (2) its robustness when dealing with familiar versus unfamiliar interlocutors and topics, and (3) the impact of additional fine-tuning on specific persona-based dialogues. We evaluate dialogues generated with diverse speaker pairings and topics, framing the evaluation as an author identification task and employing both LLM-as-a-judge and human evaluations. By systematically masking or disclosing information about the interlocutor, we assess its impact on dialogue generation. Results show that access to the interlocutor's persona improves the recognition of the target speaker, while masking it does the opposite. Although models generalise well across topics, they struggle with unfamiliar interlocutors. Finally, we found that in zero-shot settings, LLMs often copy biographical details, facilitating identification but trivialising the task.

LLMs often excel on standard benchmarks but falter on real-world tasks. We introduce DeepQuestion, a scalable automated framework that augments existing datasets based on Bloom's taxonomy and creates novel questions that trace original solution paths to probe evaluative and creative skills. Extensive experiments across ten open-source and proprietary models, covering both general-purpose and reasoning LLMs, reveal substantial performance drops (even up to 70% accuracy loss) on higher-order tasks, underscoring persistent gaps in deep reasoning. Our work highlights the need for cognitively diverse benchmarks to advance LLM progress. DeepQuestion and related datasets will be released upon acceptance of the paper.

In this paper, we investigate code-integrated reasoning, where models generate code when necessary and integrate feedback by executing it through a code interpreter. To acquire this capability, models must learn when and how to use external code tools effectively, which is supported by tool-augmented reinforcement learning (RL) through interactive learning. Despite its benefits, tool-augmented RL can still suffer from potential instability in the learning dynamics. In light of this challenge, we present a systematic approach to improving the training effectiveness and stability of tool-augmented RL for code-integrated reasoning. Specifically, we develop enhanced training strategies that balance exploration and stability, progressively building tool-use capabilities while improving reasoning performance. Through extensive experiments on five mainstream mathematical reasoning benchmarks, our model demonstrates significant performance improvements over multiple competitive baselines. Furthermore, we conduct an in-depth analysis of the mechanism and effect of code-integrated reasoning, revealing several key insights, such as the extension of model's capability boundaries and the simultaneous improvement of reasoning efficiency through code integration. All data and code for reproducing this work are available at: https://github.com/RUCAIBox/CIR.

Artificial intelligence (AI) is poised to transform education, but the research community lacks a robust, general benchmark to evaluate AI models for learning. To assess state-of-the-art support for educational use cases, we ran an "arena for learning" where educators and pedagogy experts conduct blind, head-to-head, multi-turn comparisons of leading AI models. In particular, $N = 189$ educators drew from their experience to role-play realistic learning use cases, interacting with two models sequentially, after which $N = 206$ experts judged which model better supported the user's learning goals. The arena evaluated a slate of state-of-the-art models: Gemini 2.5 Pro, Claude 3.7 Sonnet, GPT-4o, and OpenAI o3. Excluding ties, experts preferred Gemini 2.5 Pro in 73.2% of these match-ups -- ranking it first overall in the arena. Gemini 2.5 Pro also demonstrated markedly higher performance across key principles of good pedagogy. Altogether, these results position Gemini 2.5 Pro as a leading model for learning.

Although multi-agent systems based on large language models show strong capabilities on multiple tasks, they are still limited by high computational overhead, information loss, and robustness. Inspired by ResNet's residual learning, we propose Residual Mixture-of-Agents (RMoA), integrating residual connections to optimize efficiency and reliability. To maximize information utilization from model responses while minimizing computational costs, we innovatively design an embedding-based diversity selection mechanism that greedily selects responses via vector similarity. Furthermore, to mitigate iterative information degradation, we introduce a Residual Extraction Agent to preserve cross-layer incremental information by capturing inter-layer response differences, coupled with a Residual Aggregation Agent for hierarchical information integration. Additionally, we propose an adaptive termination mechanism that dynamically halts processing based on residual convergence, further improving inference efficiency. RMoA achieves state-of-the-art performance on the benchmarks of across alignment, mathematical reasoning, code generation, and multitasking understanding, while significantly reducing computational overhead. Code is available at https://github.com/mindhunter01/RMoA.

Accurate modeling of subjective phenomena such as emotion expression requires data annotated with authors' intentions. Commonly such data is collected by asking study participants to donate and label genuine content produced in the real world, or create content fitting particular labels during the study. Asking participants to create content is often simpler to implement and presents fewer risks to participant privacy than data donation. However, it is unclear if and how study-created content may differ from genuine content, and how differences may impact models. We collect study-created and genuine multimodal social media posts labeled for emotion and compare them on several dimensions, including model performance. We find that compared to genuine posts, study-created posts are longer, rely more on their text and less on their images for emotion expression, and focus more on emotion-prototypical events. The samples of participants willing to donate versus create posts are demographically different. Study-created data is valuable to train models that generalize well to genuine data, but realistic effectiveness estimates require genuine data.

Over the last thirty years, considerable progress has been made with the development of systems that can drive cars, play games, predict protein folding and generate natural language. These systems are described as intelligent and there has been a great deal of talk about the rapid increase in artificial intelligence and its potential dangers. However, our theoretical understanding of intelligence and ability to measure it lag far behind our capacity for building systems that mimic intelligent human behaviour. There is no commonly agreed definition of the intelligence that AI systems are said to possess. No-one has developed a practical measure that would enable us to compare the intelligence of humans, animals and AIs on a single ratio scale. This paper sets out a new universal measure of intelligence that is based on the hypothesis that prediction is the most important component of intelligence. As an agent interacts with its normal environment, the accuracy of its predictions is summed up and the complexity of its predictions and perceived environment is accounted for using Kolmogorov complexity. Two experiments were carried out to evaluate the practical feasibility of the algorithm. These demonstrated that it could measure the intelligence of an agent embodied in a virtual maze and an agent that makes predictions about time-series data. This universal measure could be the starting point for a new comparative science of intelligence that ranks humans, animals and AIs on a single ratio scale.

Sign Language Translation (SLT) aims to convert sign language (SL) videos into spoken language text, thereby bridging the communication gap between the sign and the spoken community. While most existing works focus on translating a single sign language into a single spoken language (one-to-one SLT), leveraging multilingual resources could mitigate low-resource issues and enhance accessibility. However, multilingual SLT (MLSLT) remains unexplored due to language conflicts and alignment difficulties across SLs and spoken languages. To address these challenges, we propose a multilingual gloss-free model with dual CTC objectives for token-level SL identification and spoken text generation. Our model supports 10 SLs and handles one-to-one, many-to-one, and many-to-many SLT tasks, achieving competitive performance compared to state-of-the-art methods on three widely adopted benchmarks: multilingual SP-10, PHOENIX14T, and CSL-Daily.

Data fusion techniques integrate information from heterogeneous data sources to improve learning, generalization, and decision making across data sciences. In causal inference, these methods leverage rich observational data to improve causal effect estimation, while maintaining the trustworthiness of randomized controlled trials. Existing approaches often relax the strong no unobserved confounding assumption by instead assuming exchangeability of counterfactual outcomes across data sources. However, when both assumptions simultaneously fail - a common scenario in practice - current methods cannot identify or estimate causal effects. We address this limitation by proposing a novel partial identification framework that enables researchers to answer key questions such as: Is the causal effect positive or negative? and How severe must assumption violations be to overturn this conclusion? Our approach introduces interpretable sensitivity parameters that quantify assumption violations and derives corresponding causal effect bounds. We develop doubly robust estimators for these bounds and operationalize breakdown frontier analysis to understand how causal conclusions change as assumption violations increase. We apply our framework to the Project STAR study, which investigates the effect of classroom size on students' third-grade standardized test performance. Our analysis reveals that the Project STAR results are robust to simultaneous violations of key assumptions, both on average and across various subgroups of interest. This strengthens confidence in the study's conclusions despite potential unmeasured biases in the data.