Large Language Models (LLMs) have exhibited remarkable reasoning capabilities and become the foundation of language technologies. Inspired by the great success of code data in training LLMs, we naturally wonder at which training stage introducing code data can really help LLMs reasoning. To this end, this paper systematically explores the impact of code data on LLMs at different stages. Concretely, we introduce the code data at the pre-training stage, instruction-tuning stage, and both of them, respectively. Then, the reasoning capability of LLMs is comprehensively and fairly evaluated via six reasoning tasks in five domains. We critically analyze the experimental results and provide conclusions with insights. First, pre-training LLMs with the mixture of code and text can significantly enhance LLMs' general reasoning capability almost without negative transfer on other tasks. Besides, at the instruction-tuning stage, code data endows LLMs the task-specific reasoning capability. Moreover, the dynamic mixing strategy of code and text data assists LLMs to learn reasoning capability step-by-step during training. These insights deepen the understanding of LLMs regarding reasoning ability for their application, such as scientific question answering, legal support, etc. The source code and model parameters are released at the link:~\url{https://github.com/yingweima2022/CodeLLM}.
Multimodal sentiment analysis in videos is a key task in many real-world applications, which usually requires integrating multimodal streams including visual, verbal and acoustic behaviors. To improve the robustness of multimodal fusion, some of the existing methods let different modalities communicate with each other and modal the crossmodal interaction via transformers. However, these methods only use the single-scale representations during the interaction but forget to exploit multi-scale representations that contain different levels of semantic information. As a result, the representations learned by transformers could be biased especially for unaligned multimodal data. In this paper, we propose a multi-scale cooperative multimodal transformer (MCMulT) architecture for multimodal sentiment analysis. On the whole, the "multi-scale" mechanism is capable of exploiting the different levels of semantic information of each modality which are used for fine-grained crossmodal interactions. Meanwhile, each modality learns its feature hierarchies via integrating the crossmodal interactions from multiple level features of its source modality. In this way, each pair of modalities progressively builds feature hierarchies respectively in a cooperative manner. The empirical results illustrate that our MCMulT model not only outperforms existing approaches on unaligned multimodal sequences but also has strong performance on aligned multimodal sequences.
Demand estimation plays an important role in dynamic pricing where the optimal price can be obtained via maximizing the revenue based on the demand curve. In online hotel booking platform, the demand or occupancy of rooms varies across room-types and changes over time, and thus it is challenging to get an accurate occupancy estimate. In this paper, we propose a novel hotel demand function that explicitly models the price elasticity of demand for occupancy prediction, and design a price elasticity prediction model to learn the dynamic price elasticity coefficient from a variety of affecting factors. Our model is composed of carefully designed elasticity learning modules to alleviate the endogeneity problem, and trained in a multi-task framework to tackle the data sparseness. We conduct comprehensive experiments on real-world datasets and validate the superiority of our method over the state-of-the-art baselines for both occupancy prediction and dynamic pricing.
High-fidelity quantum dynamics emulators can be used to predict the time evolution of complex physical systems. Here, we introduce an efficient training framework for constructing machine learning-based emulators. Our approach is based on the idea of knowledge distillation and uses elements of curriculum learning. It works by constructing a set of simple, but rich-in-physics training examples (a curriculum). These examples are used by the emulator to learn the general rules describing the time evolution of a quantum system (knowledge distillation). The goal is not only to obtain high-quality predictions, but also to examine the process of how the emulator learns the physics of the underlying problem. This allows us to discover new facts about the physical system, detect symmetries, and measure relative importance of the contributing physical processes. We illustrate this approach by training an artificial neural network to predict the time evolution of quantum wave packages propagating through a potential landscape. We focus on the question of how the emulator learns the rules of quantum dynamics from the curriculum of simple training examples and to which extent it can generalize the acquired knowledge to solve more challenging cases.
Recent progress in large language models (LLMs) has led to impressive performance on a range of tasks, yet advanced instruction following (IF)-especially for complex, multi-turn, and system-prompted instructions-remains a significant challenge. Rigorous evaluation and effective training for such capabilities are hindered by the lack of high-quality, human-annotated benchmarks and reliable, interpretable reward signals. In this work, we introduce AdvancedIF (we will release this benchmark soon), a comprehensive benchmark featuring over 1,600 prompts and expert-curated rubrics that assess LLMs ability to follow complex, multi-turn, and system-level instructions. We further propose RIFL (Rubric-based Instruction-Following Learning), a novel post-training pipeline that leverages rubric generation, a finetuned rubric verifier, and reward shaping to enable effective reinforcement learning for instruction following. Extensive experiments demonstrate that RIFL substantially improves the instruction-following abilities of LLMs, achieving a 6.7% absolute gain on AdvancedIF and strong results on public benchmarks. Our ablation studies confirm the effectiveness of each component in RIFL. This work establishes rubrics as a powerful tool for both training and evaluating advanced IF in LLMs, paving the way for more capable and reliable AI systems.
We introduce MMBench-GUI, a hierarchical benchmark for evaluating GUI automation agents across Windows, macOS, Linux, iOS, Android, and Web platforms. It comprises four levels: GUI Content Understanding, Element Grounding, Task Automation, and Task Collaboration, covering essential skills for GUI agents. In addition, we propose a novel Efficiency-Quality Area (EQA) metric to assess GUI agent execution efficiency in online automation scenarios. Through MMBench-GUI, we identify accurate visual grounding as a critical determinant of overall task success, emphasizing the substantial benefits of modular frameworks that integrate specialized grounding modules. Furthermore, to achieve reliable GUI automation, an agent requires strong task planning and cross-platform generalization abilities, with long-context memory, a broad action space, and long-term reasoning playing a critical role. More important, task efficiency remains a critically underexplored dimension, and all models suffer from substantial inefficiencies, with excessive redundant steps even when tasks are ultimately completed. The integration of precise localization, effective planning, and early stopping strategies is indispensable to enable truly efficient and scalable GUI automation. Our benchmark code, evaluation data, and running environment will be publicly available at https://github.com/open-compass/MMBench-GUI.
Context: Pull-based development model is widely used in open source, leading the trends in distributed software development. One aspect which has garnered significant attention is studies on pull request decision - identifying factors for explanation. Objective: This study builds on a decade long research on pull request decision to explain it. We empirically investigate how factors influence pull request decision and scenarios that change the influence of factors. Method: We identify factors influencing pull request decision on GitHub through a systematic literature review and infer it by mining archival data. We collect a total of 3,347,937 pull requests with 95 features from 11,230 diverse projects on GitHub. Using this data, we explore the relations of the factors to each other and build mixed-effect logistic regression models to empirically explain pull request decision. Results: Our study shows that a small number of factors explain pull request decision with the integrator same or different from the submitter as the most important factor. We also noted that some factors are important only in special cases e.g., the percentage of failed builds is important for pull request decision when continuous integration is used.
Numerical reasoning is required when solving most problems in our life, but it has been neglected in previous artificial intelligence researches. FinQA challenge has been organized to strengthen the study on numerical reasoning where the participants are asked to predict the numerical reasoning program to solve financial question. The result of FinQA will be evaluated by both execution accuracy and program accuracy. In this paper, we present our approach to tackle the task objective by developing models with different specialized capabilities and fusing their strength. Overall, our approach achieves the 1st place in FinQA challenge, with 71.93% execution accuracy and 67.03% program accuracy.
As single-center computing approaches power constraints, decentralized training becomes essential. However, traditional Reinforcement Learning (RL) methods, crucial for enhancing large model post-training, cannot adapt to decentralized distributed training due to the tight coupling between parameter learning and rollout sampling. For this, we propose HeteroRL, a heterogeneous RL architecture that decouples these processes, enabling stable training across geographically distributed nodes connected via the Internet. The core component is Group Expectation Policy Optimization (GEPO), an asynchronous RL algorithm robust to latency caused by network delays or heterogeneity in computational resources. Our study reveals that high latency significantly increases KL divergence, leading to higher variance of importance weights and training instability. GEPO mitigates this issue by using group expectation weighting to exponentially reduce the variance of importance weights, with theoretical guarantees. Experiments show GEPO achieves superior stability - only a 3% performance drop from online to 1800s latency-and reduces the best-to-last gap by 85% versus GSPO (1.8 vs. 12.0) while attaining the highest scores, highlighting its effectiveness in decentralized, resource-heterogeneous environments.
Medical image segmentation plays a crucial role in clinical workflows, but domain shift often leads to performance degradation when models are applied to unseen clinical domains. This challenge arises due to variations in imaging conditions, scanner types, and acquisition protocols, limiting the practical deployment of segmentation models. Unlike natural images, medical images typically exhibit consistent anatomical structures across patients, with domain-specific variations mainly caused by imaging conditions. This unique characteristic makes medical image segmentation particularly challenging. To address this challenge, we propose a domain generalization framework tailored for medical image segmentation. Our approach improves robustness to domain-specific variations by introducing implicit feature perturbations guided by domain statistics. Specifically, we employ a learnable semantic direction selector and a covariance-based semantic intensity sampler to modulate domain-variant features while preserving task-relevant anatomical consistency. Furthermore, we design an adaptive consistency constraint that is selectively applied only when feature adjustment leads to degraded segmentation performance. This constraint encourages the adjusted features to align with the original predictions, thereby stabilizing feature selection and improving the reliability of the segmentation. Extensive experiments on two public multi-center benchmarks show that our framework consistently outperforms existing domain generalization approaches, achieving robust and generalizable segmentation performance across diverse clinical domains.
Recently, Multimodal Large Language Models (MLLMs) have been used as agents to control keyboard and mouse inputs by directly perceiving the Graphical User Interface (GUI) and generating corresponding commands. However, current agents primarily demonstrate strong understanding capabilities in static environments and are mainly applied to relatively simple domains, such as Web or mobile interfaces. We argue that a robust GUI agent should be capable of perceiving temporal information on the GUI, including dynamic Web content and multi-step tasks. Additionally, it should possess a comprehensive understanding of various GUI scenarios, including desktop software and multi-window interactions. To this end, this paper introduces a new dataset, termed GUI-World, which features meticulously crafted Human-MLLM annotations, extensively covering six GUI scenarios and eight types of GUI-oriented questions in three formats. We evaluate the capabilities of current state-of-the-art MLLMs, including Image LLMs and Video LLMs, in understanding various types of GUI content, especially dynamic and sequential content. Our findings reveal that current models struggle with dynamic GUI content without manually annotated keyframes or operation history. On the other hand, Video LLMs fall short in all GUI-oriented tasks given the sparse GUI video dataset. Therefore, we take the initial step of leveraging a fine-tuned Video LLM, GUI-Vid, as a GUI-oriented assistant, demonstrating an improved understanding of various GUI tasks. However, due to the limitations in the performance of base LLMs, we conclude that using video LLMs as GUI agents remains a significant challenge. We believe our work provides valuable insights for future research in dynamic GUI content understanding. All the dataset and code are publicly available at: https://gui-world.github.io.
Algorithms which minimize the averaged loss have been widely designed for dealing with noisy labels. Intuitively, when there is a finite training sample, penalizing the variance of losses will improve the stability and generalization of the algorithms. Interestingly, we found that the variance should be increased for the problem of learning with noisy labels. Specifically, increasing the variance will boost the memorization effects and reduce the harmfulness of incorrect labels. By exploiting the label noise transition matrix, regularizers can be easily designed to reduce the variance of losses and be plugged in many existing algorithms. Empirically, the proposed method by increasing the variance of losses significantly improves the generalization ability of baselines on both synthetic and real-world datasets.
Given only positive (P) and unlabeled (U) data, PU learning can train a binary classifier without any negative data. It has two building blocks: PU class-prior estimation (CPE) and PU classification; the latter has been well studied while the former has received less attention. Hitherto, the distributional-assumption-free CPE methods rely on a critical assumption that the support of the positive data distribution cannot be contained in the support of the negative data distribution. If this is violated, those CPE methods will systematically overestimate the class prior; it is even worse that we cannot verify the assumption based on the data. In this paper, we rethink CPE for PU learning-can we remove the assumption to make CPE always valid? We show an affirmative answer by proposing Regrouping CPE (ReCPE) that builds an auxiliary probability distribution such that the support of the positive data distribution is never contained in the support of the negative data distribution. ReCPE can work with any CPE method by treating it as the base method. Theoretically, ReCPE does not affect its base if the assumption already holds for the original probability distribution; otherwise, it reduces the positive bias of its base. Empirically, ReCPE improves all state-of-the-art CPE methods on various datasets, implying that the assumption has indeed been violated here.