Reasoning is the fundamental capability of large language models (LLMs). Due to the rapid progress of LLMs, there are two main issues of current benchmarks: i) these benchmarks can be crushed in a short time (less than 1 year), and ii) these benchmarks may be easily hacked. To handle these issues, we propose the ever-scalingness for building the benchmarks which are uncrushable, unhackable, auto-verifiable and general. This paper presents Nondeterministic Polynomial-time Problem Challenge (NPPC), an ever-scaling reasoning benchmark for LLMs. Specifically, the NPPC has three main modules: i) npgym, which provides a unified interface of 25 well-known NP-complete problems and can generate any number of instances with any levels of complexities, ii) npsolver: which provides a unified interface to evaluate the problem instances with both online and offline models via APIs and local deployments, respectively, and iii) npeval: which provides the comprehensive and ready-to-use tools to analyze the performances of LLMs over different problems, the number of tokens, the aha moments, the reasoning errors and the solution errors. Extensive experiments over widely-used LLMs demonstrate: i) NPPC can successfully decrease the performances of advanced LLMs' performances to below 10%, demonstrating that NPPC is uncrushable, ii) DeepSeek-R1, Claude-3.7-Sonnet, and o1/o3-mini are the most powerful LLMs, where DeepSeek-R1 outperforms Claude-3.7-Sonnet and o1/o3-mini in most NP-complete problems considered, and iii) the numbers of tokens, aha moments in the advanced LLMs, e.g., Claude-3.7-Sonnet and DeepSeek-R1, are observed first to increase and then decrease when the problem instances become more and more difficult. We believe that NPPC is the first ever-scaling reasoning benchmark, serving as the uncrushable and unhackable testbed for LLMs toward artificial general intelligence (AGI).
Recent advances in Large Language Models have demonstrated their capabilities across a variety of tasks. However, automatically extracting implicit knowledge from natural language remains a significant challenge, as machines lack active experience with the physical world. Given this scenario, semantic knowledge graphs can serve as conceptual spaces that guide the automated text generation reasoning process to achieve more efficient and explainable results. In this paper, we apply a logic-augmented generation (LAG) framework that leverages the explicit representation of a text through a semantic knowledge graph and applies it in combination with prompt heuristics to elicit implicit analogical connections. This method generates extended knowledge graph triples representing implicit meaning, enabling systems to reason on unlabeled multimodal data regardless of the domain. We validate our work through three metaphor detection and understanding tasks across four datasets, as they require deep analogical reasoning capabilities. The results show that this integrated approach surpasses current baselines, performs better than humans in understanding visual metaphors, and enables more explainable reasoning processes, though still has inherent limitations in metaphor understanding, especially for domain-specific metaphors. Furthermore, we propose a thorough error analysis, discussing issues with metaphorical annotations and current evaluation methods.
Infants often exhibit goal-directed behaviors, such as reaching for a sensory stimulus, even when no external reward criterion is provided. These intrinsically motivated behaviors facilitate spontaneous exploration and learning of the body and environment during early developmental stages. Although computational modeling can offer insight into the mechanisms underlying such behaviors, many existing studies on intrinsic motivation focus primarily on how exploration contributes to acquiring external rewards. In this paper, we propose a novel density model for an agent's own multimodal sensory experiences, called the "self-prior," and investigate whether it can autonomously induce goal-directed behavior. Integrated within an active inference framework based on the free energy principle, the self-prior generates behavioral references purely from an intrinsic process that minimizes mismatches between average past sensory experiences and current observations. This mechanism is also analogous to the acquisition and utilization of a body schema through continuous interaction with the environment. We examine this approach in a simulated environment and confirm that the agent spontaneously reaches toward a tactile stimulus. Our study implements intrinsically motivated behavior shaped by the agent's own sensory experiences, demonstrating the spontaneous emergence of intentional behavior during early development.
Unsupervised reinforcement learning (RL) aims at pre-training agents that can solve a wide range of downstream tasks in complex environments. Despite recent advancements, existing approaches suffer from several limitations: they may require running an RL process on each downstream task to achieve a satisfactory performance, they may need access to datasets with good coverage or well-curated task-specific samples, or they may pre-train policies with unsupervised losses that are poorly correlated with the downstream tasks of interest. In this paper, we introduce a novel algorithm regularizing unsupervised RL towards imitating trajectories from unlabeled behavior datasets. The key technical novelty of our method, called Forward-Backward Representations with Conditional-Policy Regularization, is to train forward-backward representations to embed the unlabeled trajectories to the same latent space used to represent states, rewards, and policies, and use a latent-conditional discriminator to encourage policies to ``cover'' the states in the unlabeled behavior dataset. As a result, we can learn policies that are well aligned with the behaviors in the dataset, while retaining zero-shot generalization capabilities for reward-based and imitation tasks. We demonstrate the effectiveness of this new approach in a challenging humanoid control problem: leveraging observation-only motion capture datasets, we train Meta Motivo, the first humanoid behavioral foundation model that can be prompted to solve a variety of whole-body tasks, including motion tracking, goal reaching, and reward optimization. The resulting model is capable of expressing human-like behaviors and it achieves competitive performance with task-specific methods while outperforming state-of-the-art unsupervised RL and model-based baselines.
Spiking Neural Networks (SNNs) offer promising energy efficiency advantages, particularly when processing sparse spike trains. However, their incompatibility with traditional datasets, which consist of batches of input vectors rather than spike trains, necessitates the development of efficient encoding methods. This paper introduces a novel, open-source PyTorch-compatible Python framework for spike encoding, designed for neuromorphic applications in machine learning and reinforcement learning. The framework supports a range of encoding algorithms, including Leaky Integrate-and-Fire (LIF), Step Forward (SF), Pulse Width Modulation (PWM), and Ben's Spiker Algorithm (BSA), as well as specialized encoding strategies covering population coding and reinforcement learning scenarios. Furthermore, we investigate the performance trade-offs of each method on embedded hardware using C/C++ implementations, considering energy consumption, computation time, spike sparsity, and reconstruction accuracy. Our findings indicate that SF typically achieves the lowest reconstruction error and offers the highest energy efficiency and fastest encoding speed, achieving the second-best spike sparsity. At the same time, other methods demonstrate particular strengths depending on the signal characteristics. This framework and the accompanying empirical analysis provide valuable resources for selecting optimal encoding strategies for energy-efficient SNN applications.
The internet contains large amounts of low-quality content, yet users expect web search engines to deliver high-quality, relevant results. The abundant presence of low-quality pages can negatively impact retrieval and crawling processes by wasting resources on these documents. Therefore, search engines can greatly benefit from techniques that leverage efficient quality estimation methods to mitigate these negative impacts. Quality scoring methods for web pages are useful for many processes typical for web search systems, including static index pruning, index tiering, and crawling. Building on work by Chang et al.~\cite{chang2024neural}, who proposed using neural estimators of semantic quality for static index pruning, we extend their approach and apply their neural quality scorers to assess the semantic quality of web pages in crawling prioritisation tasks. In our experimental analysis, we found that prioritising semantically high-quality pages over low-quality ones can improve downstream search effectiveness. Our software contribution consists of a Docker container that computes an effective quality score for a given web page, allowing the quality scorer to be easily included and used in other components of web search systems.
Link prediction on graphs has applications spanning from recommender systems to drug discovery. Temporal link prediction (TLP) refers to predicting future links in a temporally evolving graph and adds additional complexity related to the dynamic nature of graphs. State-of-the-art TLP models incorporate memory modules alongside graph neural networks to learn both the temporal mechanisms of incoming nodes and the evolving graph topology. However, memory modules only store information about nodes seen at train time, and hence such models cannot be directly transferred to entirely new graphs at test time and deployment. In this work, we study a new transfer learning task for temporal link prediction, and develop transfer-effective methods for memory-laden models. Specifically, motivated by work showing the informativeness of structural signals for the TLP task, we augment a structural mapping module to the existing TLP model architectures, which learns a mapping from graph structural (topological) features to memory embeddings. Our work paves the way for a memory-free foundation model for TLP.
Conversational Recommender Systems (CRSs) aim to provide personalized recommendations by interacting with users through conversations. Most existing studies of CRS focus on extracting user preferences from conversational contexts. However, due to the short and sparse nature of conversational contexts, it is difficult to fully capture user preferences by conversational contexts only. We argue that multi-modal semantic information can enrich user preference expressions from diverse dimensions (e.g., a user preference for a certain movie may stem from its magnificent visual effects and compelling storyline). In this paper, we propose a multi-modal semantic graph prompt learning framework for CRS, named MSCRS. First, we extract textual and image features of items mentioned in the conversational contexts. Second, we capture higher-order semantic associations within different semantic modalities (collaborative, textual, and image) by constructing modality-specific graph structures. Finally, we propose an innovative integration of multi-modal semantic graphs with prompt learning, harnessing the power of large language models to comprehensively explore high-dimensional semantic relationships. Experimental results demonstrate that our proposed method significantly improves accuracy in item recommendation, as well as generates more natural and contextually relevant content in response generation.
Task arithmetic is a straightforward yet highly effective strategy for model merging, enabling the resultant model to exhibit multi-task capabilities. Recent research indicates that models demonstrating linearity enhance the performance of task arithmetic. In contrast to existing methods that rely on the global linearization of the model, we argue that this linearity already exists within the model's submodules. In particular, we present a statistical analysis and show that submodules (e.g., layers, self-attentions, and MLPs) exhibit significantly higher linearity than the overall model. Based on these findings, we propose an innovative model merging strategy that independently merges these submodules. Especially, we derive a closed-form solution for optimal merging weights grounded in the linear properties of these submodules. Experimental results demonstrate that our method consistently outperforms the standard task arithmetic approach and other established baselines across different model scales and various tasks. This result highlights the benefits of leveraging the linearity of submodules and provides a new perspective for exploring solutions for effective and practical multi-task model merging.
Federated Learning (FL) has emerged as an effective solution for multi-institutional collaborations without sharing patient data, offering a range of methods tailored for diverse applications. However, real-world medical datasets are often multimodal, and computational resources are limited, posing significant challenges for existing FL approaches. Recognizing these limitations, we developed the Federated Healthcare Benchmark(FHBench), a benchmark specifically designed from datasets derived from real-world healthcare applications. FHBench encompasses critical diagnostic tasks across domains such as the nervous, cardiovascular, and respiratory systems and general pathology, providing comprehensive support for multimodal healthcare evaluations and filling a significant gap in existing benchmarks. Building on FHBench, we introduced Efficient Personalized Federated Learning with Adaptive LoRA(EPFL), a personalized FL framework that demonstrates superior efficiency and effectiveness across various healthcare modalities. Our results highlight the robustness of FHBench as a benchmarking tool and the potential of EPFL as an innovative approach to advancing healthcare-focused FL, addressing key limitations of existing methods.
As the post-training of large language models (LLMs) advances from instruction-following to complex reasoning tasks, understanding how different data affect finetuning dynamics remains largely unexplored. In this paper, we present a spectral analysis of layer-wise gradients induced by low/high-quality instruction and reasoning data for LLM post-training. Our analysis reveals that widely-studied metrics for data evaluation, e.g., IFD, InsTag, Difficulty, and Reward, can be explained and unified by spectral properties computed from gradients' singular value decomposition (SVD). Specifically, higher-quality data are usually associated with lower nuclear norms and higher effective ranks. Notably, effective rank exhibits better robustness and resolution than nuclear norm in capturing subtle quality differences. For example, reasoning data achieves substantially higher effective ranks than instruction data, implying richer gradient structures on more complex tasks. Our experiments also highlight that models within the same family share similar gradient patterns regardless of their sizes, whereas different model families diverge significantly. Providing a unified view on the effects of data quality across instruction and reasoning data, this work illuminates the interplay between data quality and training stability, shedding novel insights into developing better data exploration strategies for post-training.
Cross-Domain Few-Shot Object Detection (CD-FSOD) poses significant challenges to existing object detection and few-shot detection models when applied across domains. In conjunction with NTIRE 2025, we organized the 1st CD-FSOD Challenge, aiming to advance the performance of current object detectors on entirely novel target domains with only limited labeled data. The challenge attracted 152 registered participants, received submissions from 42 teams, and concluded with 13 teams making valid final submissions. Participants approached the task from diverse perspectives, proposing novel models that achieved new state-of-the-art (SOTA) results under both open-source and closed-source settings. In this report, we present an overview of the 1st NTIRE 2025 CD-FSOD Challenge, highlighting the proposed solutions and summarizing the results submitted by the participants.
Process Reward Models (PRMs) provide step-level supervision to large language models (LLMs), but scaling up training data annotation remains challenging for both humans and LLMs. To address this limitation, we propose an active learning approach, ActPRM, which proactively selects the most uncertain samples for training, substantially reducing labeling costs. During training, we use the PRM to estimate uncertainty after the forward pass, retaining only highly uncertain data. A capable yet costly reasoning model then labels this data. Then we compute the loss with respect to the labels and update the PRM's weights. We compare ActPRM vs. vanilla fine-tuning, on a pool-based active learning setting, demonstrating that ActPRM reduces 50% annotation, but achieving the comparable or even better performance. Beyond annotation efficiency, we further advance the actively trained PRM by filtering over 1M+ math reasoning trajectories with ActPRM, retaining 60% of the data. A subsequent training on this selected dataset yields a new state-of-the-art (SOTA) PRM on ProcessBench (75.0%) and PRMBench (65.5%) compared with same sized models.
Recent advances in recommender systems have highlighted the complementary strengths of generative modeling and pretrained language models. We propose a hybrid framework that augments the Hierarchical Sequential Transduction Unit (HSTU) generative recommender with BLaIR -- a contrastive text embedding model. This integration enriches item representations with semantic signals from textual metadata while preserving HSTU's powerful sequence modeling capabilities. We evaluate our method on two domains from the Amazon Reviews 2023 dataset, comparing it against the original HSTU and a variant that incorporates embeddings from OpenAI's state-of-the-art text-embedding-3-large model. While the OpenAI embedding model is likely trained on a substantially larger corpus with significantly more parameters, our lightweight BLaIR-enhanced approach -- pretrained on domain-specific data -- consistently achieves better performance, highlighting the effectiveness of contrastive text embeddings in compute-efficient settings.
Color plays an important role in human perception and usually provides critical clues in visual reasoning. However, it is unclear whether and how vision-language models (VLMs) can perceive, understand, and leverage color as humans. This paper introduces ColorBench, an innovative benchmark meticulously crafted to assess the capabilities of VLMs in color understanding, including color perception, reasoning, and robustness. By curating a suite of diverse test scenarios, with grounding in real applications, ColorBench evaluates how these models perceive colors, infer meanings from color-based cues, and maintain consistent performance under varying color transformations. Through an extensive evaluation of 32 VLMs with varying language models and vision encoders, our paper reveals some undiscovered findings: (i) The scaling law (larger models are better) still holds on ColorBench, while the language model plays a more important role than the vision encoder. (ii) However, the performance gaps across models are relatively small, indicating that color understanding has been largely neglected by existing VLMs. (iii) CoT reasoning improves color understanding accuracies and robustness, though they are vision-centric tasks. (iv) Color clues are indeed leveraged by VLMs on ColorBench but they can also mislead models in some tasks. These findings highlight the critical limitations of current VLMs and underscore the need to enhance color comprehension. Our ColorBenchcan serve as a foundational tool for advancing the study of human-level color understanding of multimodal AI.
Large language models (LLMs) have demonstrated remarkable effectiveness in completing various tasks such as reasoning, translation, and question answering. However the issue of factual incorrect content in LLM-generated responses remains a persistent challenge. In this study, we evaluate both proprietary and open-source LLMs on their ability to respond with relevant research papers and accurate links to articles hosted on the arXiv platform, based on high level prompts. To facilitate this evaluation, we introduce arXivBench, a benchmark specifically designed to assess LLM performance across eight major subject categories on arXiv and five subfields within computer science, one of the most popular categories among them. Our findings reveal a concerning accuracy of LLM-generated responses depending on the subject, with some subjects experiencing significantly lower accuracy than others. Notably, Claude-3.5-Sonnet exhibits a substantial advantage in generating both relevant and accurate responses. And interestingly, most LLMs achieve a much higher accuracy in the Artificial Intelligence sub-field than other sub-fields. This benchmark provides a standardized tool for evaluating the reliability of LLM-generated scientific responses, promoting more dependable use of LLMs in academic and research environments. Our code is open-sourced at https://github.com/arxivBenchLLM/arXivBench and our dataset is available on huggingface at https://huggingface.co/datasets/arXivBenchLLM/arXivBench.
In this paper, we present Roamify, an Artificial Intelligence powered travel assistant that aims to ease the process of travel planning. We have tested and used multiple Large Language Models like Llama and T5 to generate personalised itineraries per user preferences. Results from user surveys highlight the preference for AI powered mediums over existing methods to help in travel planning across all user age groups. These results firmly validate the potential need of such a travel assistant. We highlight the two primary design considerations for travel assistance: D1) incorporating a web-scraping method to gather up-to-date news articles about destinations from various blog sources, which significantly improves our itinerary suggestions, and D2) utilising user preferences to create customised travel experiences along with a recommendation system which changes the itinerary according to the user needs. Our findings suggest that Roamify has the potential to improve and simplify how users across multiple age groups plan their travel experiences.
Checkboxes are critical in real-world document processing where the presence or absence of ticks directly informs data extraction and decision-making processes. Yet, despite the strong performance of Large Vision and Language Models across a wide range of tasks, they struggle with interpreting checkable content. This challenge becomes particularly pressing in industries where a single overlooked checkbox may lead to costly regulatory or contractual oversights. To address this gap, we introduce the CheckboxQA dataset, a targeted resource designed to evaluate and improve model performance on checkbox-related tasks. It reveals the limitations of current models and serves as a valuable tool for advancing document comprehension systems, with significant implications for applications in sectors such as legal tech and finance. The dataset is publicly available at: https://github.com/Snowflake-Labs/CheckboxQA
We present MultiLoKo, a new benchmark for evaluating multilinguality in LLMs covering 31 languages. MultiLoKo consists of three partitions: a main partition consisting of 500 questions per language, separately sourced to be locally relevant to the specific language, and two translated partitions, containing human-authored translations from 30 non-English languages to English and vice versa. For comparison, we also release corresponding machine-authored translations. The data is equally distributed over two splits: a dev split and a blind, out-of-distribution test split. MultiLoKo can be used to study a variety of questions regarding the multilinguality of LLMs as well as meta-questions about multilingual benchmark creation. We compute MultiLoKo scores for 11 base and chat models marketed to be multilingual and study their average performance, their performance parity across languages, how much their ability to answer questions depends on the question language, and which languages are most difficult. None of the models we studied performs well on MultiLoKo, as indicated by low average scores as well as large differences between the best and worst scoring languages. Furthermore, we find a substantial effect of the question language, indicating sub-optimal knowledge transfer between languages. Lastly, we find that using local vs English-translated data can result in differences more than 20 points for the best performing models, drastically change the estimated difficulty of some languages. For using machines instead of human translations, we find a weaker effect on ordering of language difficulty, a larger difference in model rankings, and a substantial drop in estimated performance for all models.
Graphical User Interface (GUI) agents offer cross-platform solutions for automating complex digital tasks, with significant potential to transform productivity workflows. However, their performance is often constrained by the scarcity of high-quality trajectory data. To address this limitation, we propose training Vision Language Models (VLMs) on data-rich, reasoning-intensive tasks during a dedicated mid-training stage, and then examine how incorporating these tasks facilitates generalization to GUI planning scenarios. Specifically, we explore a range of tasks with readily available instruction-tuning data, including GUI perception, multimodal reasoning, and textual reasoning. Through extensive experiments across 11 mid-training tasks, we demonstrate that: (1) Task generalization proves highly effective, yielding substantial improvements across most settings. For instance, multimodal mathematical reasoning enhances performance on AndroidWorld by an absolute 6.3%. Remarkably, text-only mathematical data significantly boosts GUI web agent performance, achieving a 5.6% improvement on WebArena and 5.4% improvement on AndroidWorld, underscoring notable cross-modal generalization from text-based to visual domains; (2) Contrary to prior assumptions, GUI perception data - previously considered closely aligned with GUI agent tasks and widely utilized for training - has a comparatively limited impact on final performance; (3) Building on these insights, we identify the most effective mid-training tasks and curate optimized mixture datasets, resulting in absolute performance gains of 8.0% on WebArena and 12.2% on AndroidWorld. Our work provides valuable insights into cross-domain knowledge transfer for GUI agents and offers a practical approach to addressing data scarcity challenges in this emerging field. The code, data and models will be available at https://github.com/hkust-nlp/GUIMid.