Memory-based approaches have shown strong performance in Continual Relation Extraction (CRE). However, storing examples from previous tasks increases memory usage and raises privacy concerns. Recently, prompt-based methods have emerged as a promising alternative, as they do not rely on storing past samples. Despite this progress, current prompt-based techniques face several core challenges in CRE, particularly in accurately identifying task identities and mitigating catastrophic forgetting. Existing prompt selection strategies often suffer from inaccuracies, lack robust mechanisms to prevent forgetting in shared parameters, and struggle to handle both cross-task and within-task variations. In this paper, we propose WAVE++, a novel approach inspired by the connection between prefix-tuning and mixture of experts. Specifically, we introduce task-specific prompt pools that enhance flexibility and adaptability across diverse tasks while avoiding boundary-spanning risks; this design more effectively captures variations within each task and across tasks. To further refine relation classification, we incorporate label descriptions that provide richer, more global context, enabling the model to better distinguish among different relations. We also propose a training-free mechanism to improve task prediction during inference. Moreover, we integrate a generative model to consolidate prior knowledge within the shared parameters, thereby removing the need for explicit data storage. Extensive experiments demonstrate that WAVE++ outperforms state-of-the-art prompt-based and rehearsal-based methods, offering a more robust solution for continual relation extraction. Our code is publicly available at https://github.com/PiDinosauR2804/WAVE-CRE-PLUS-PLUS.
With the rapid advancement of large language models like Gemini, GPT, and others, bridging the gap between the human brain and language processing has become an important area of focus. To address this challenge, researchers have developed various models to decode EEG signals into text. However, these models still face significant performance limitations. To overcome these shortcomings, we propose a new model, R1 Translator, which aims to improve the performance of EEG-to-text decoding. The R1 Translator model combines a bidirectional LSTM encoder with a pretrained transformer-based decoder, utilizing EEG features to produce high-quality text outputs. The model processes EEG embeddings through the LSTM to capture sequential dependencies, which are then fed into the transformer decoder for effective text generation. The R1 Translator excels in ROUGE metrics, outperforming both T5 (previous research) and Brain Translator. Specifically, R1 achieves a ROUGE-1 score of 38.00% (P), which is up to 9% higher than T5 (34.89%) and 3% better than Brain (35.69%). It also leads in ROUGE-L, with a F1 score of 32.51%, outperforming T5 by 3% (29.67%) and Brain by 2% (30.38%). In terms of CER, R1 achieves a CER of 0.5795, which is 2% lower than T5 (0.5917) and 4% lower than Brain (0.6001). Additionally, R1 performs better in WER with a score of 0.7280, outperforming T5 by 4.3% (0.7610) and Brain by 3.6% (0.7553). Code is available at https://github.com/Mmurrad/EEG-To-text.
Current theories of language propose an innate (Baker 2001; Chomsky 1981) or a functional (Greenberg 1963; Dryer 2007; Hawkins 2014) origin for the surface structures (i.e. word order) that we observe in languages of the world, while evolutionary modeling (Dunn et al. 2011) suggests that descent is the primary factor influencing such patterns. Although there are hypotheses for word order change from both innate and usage-based perspectives for specific languages and families, there are key disagreements between the two major proposals for mechanisms that drive the evolution of language more broadly (Wasow 2002; Levy 2008). This paper proposes a universal underlying mechanism for word order change based on a large tagged parallel dataset of over 1,500 languages representing 133 language families and 111 isolates. Results indicate that word class length is significantly correlated with word order crosslinguistically, but not in a straightforward manner, partially supporting opposing theories of processing, while at the same time predicting historical word order change in two different phylogenetic lines and explaining more variance than descent or language area in regression models. Such findings suggest an integrated "Min-Max" theory of language evolution driven by competing pressures of processing and information structure, aligning with recent efficiency-oriented (Levshina 2023) and information-theoretic proposals (Zaslavsky 2020; Tucker et al. 2025).
Large Language Models (LLMs) have recently achieved remarkable progress in mathematical reasoning. To enable such capabilities, many existing works distill strong reasoning models into long chains of thought or design algorithms to construct high-quality math QA data for training. However, these efforts primarily focus on generating correct reasoning paths and answers, while largely overlooking the validity of the questions themselves. In this work, we propose Math Question Verification (MathQ-Verify), a novel five-stage pipeline designed to rigorously filter ill-posed or under-specified math problems. MathQ-Verify first performs format-level validation to remove redundant instructions and ensure that each question is syntactically well-formed. It then formalizes each question, decomposes it into atomic conditions, and verifies them against mathematical definitions. Next, it detects logical contradictions among these conditions, followed by a goal-oriented completeness check to ensure the question provides sufficient information for solving. To evaluate this task, we use existing benchmarks along with an additional dataset we construct, containing 2,147 math questions with diverse error types, each manually double-validated. Experiments show that MathQ-Verify achieves state-of-the-art performance across multiple benchmarks, improving the F1 score by up to 25 percentage points over the direct verification baseline. It further attains approximately 90% precision and 63% recall through a lightweight model voting scheme. MathQ-Verify offers a scalable and accurate solution for curating reliable mathematical datasets, reducing label noise and avoiding unnecessary computation on invalid questions. Our code and data are available at https://github.com/scuuy/MathQ-Verify.
Recent advances in large language models (LLMs) have intensified efforts to fuse heterogeneous open-source models into a unified system that inherits their complementary strengths. Existing logit-based fusion methods maintain inference efficiency but treat vocabulary dimensions independently, overlooking semantic dependencies encoded by cross-dimension interactions. These dependencies reflect how token types interact under a model's internal reasoning and are essential for aligning models with diverse generation behaviors. To explicitly model these dependencies, we propose \textbf{InfiGFusion}, the first structure-aware fusion framework with a novel \textit{Graph-on-Logits Distillation} (GLD) loss. Specifically, we retain the top-$k$ logits per output and aggregate their outer products across sequence positions to form a global co-activation graph, where nodes represent vocabulary channels and edges quantify their joint activations. To ensure scalability and efficiency, we design a sorting-based closed-form approximation that reduces the original $O(n^4)$ cost of Gromov-Wasserstein distance to $O(n \log n)$, with provable approximation guarantees. Experiments across multiple fusion settings show that GLD consistently improves fusion quality and stability. InfiGFusion outperforms SOTA models and fusion baselines across 11 benchmarks spanning reasoning, coding, and mathematics. It shows particular strength in complex reasoning tasks, with +35.6 improvement on Multistep Arithmetic and +37.06 on Causal Judgement over SFT, demonstrating superior multi-step and relational inference.
Visual-language Chain-of-Thought (CoT) data resources are relatively scarce compared to text-only counterparts, limiting the improvement of reasoning capabilities in Vision Language Models (VLMs). However, high-quality vision-language reasoning data is expensive and labor-intensive to annotate. To address this issue, we leverage a promising resource: game code, which naturally contains logical structures and state transition processes. Therefore, we propose Code2Logic, a novel game-code-driven approach for multimodal reasoning data synthesis. Our approach leverages Large Language Models (LLMs) to adapt game code, enabling automatic acquisition of reasoning processes and results through code execution. Using the Code2Logic approach, we developed the GameQA dataset to train and evaluate VLMs. GameQA is cost-effective and scalable to produce, challenging for state-of-the-art models, and diverse with 30 games and 158 tasks. Surprisingly, despite training solely on game data, VLMs demonstrated out of domain generalization, specifically Qwen2.5-VL-7B improving performance by 2.33\% across 7 diverse vision-language benchmarks. Our code and dataset are available at https://github.com/tongjingqi/Code2Logic.
Recent reasoning-focused language models achieve high accuracy by generating lengthy intermediate reasoning paths before producing final answers. While this approach is effective in solving problems that require logical thinking, long reasoning paths significantly increase memory usage and throughput of token generation, limiting the practical deployment of such models. We propose Reasoning Path Compression (RPC), a training-free method that accelerates inference by leveraging the semantic sparsity of reasoning paths. RPC periodically compresses the KV cache by retaining KV cache that receive high importance score, which are computed using a selector window composed of recently generated queries. Experiments show that RPC improves generation throughput of QwQ-32B by up to 1.60$\times$ compared to the inference with full KV cache, with an accuracy drop of 1.2% on the AIME 2024 benchmark. Our findings demonstrate that semantic sparsity in reasoning traces can be effectively exploited for compression, offering a practical path toward efficient deployment of reasoning LLMs. Our code is available at https://github.com/jiwonsong-dev/ReasoningPathCompression.
Associative memory engages in the integration of relevant information for comprehension in the human cognition system. In this work, we seek to improve alignment between language models and human brain while processing speech information by integrating associative memory. After verifying the alignment between language model and brain by mapping language model activations to brain activity, the original text stimuli expanded with simulated associative memory are regarded as input to computational language models. We find the alignment between language model and brain is improved in brain regions closely related to associative memory processing. We also demonstrate large language models after specific supervised fine-tuning better align with brain response, by building the \textit{Association} dataset containing 1000 samples of stories, with instructions encouraging associative memory as input and associated content as output.
The Kolmogorov-Arnold Network (KAN) has been gaining popularity as an alternative to the multi-layer perceptron (MLP) with its increased expressiveness and interpretability. However, the KAN can be orders of magnitude slower due to its increased computational cost and training instability, limiting its applicability to larger-scale tasks. Recently, the Kolmogorov-Arnold Transformer (KAT) has been proposed, which can achieve FLOPs similar to the traditional Transformer with MLPs by leveraging Group-Rational KAN (GR-KAN). Unfortunately, despite the comparable FLOPs, our characterizations reveal that the KAT is still 123x slower in training speeds, indicating that there are other performance bottlenecks beyond FLOPs. In this paper, we conduct a series of experiments to understand the root cause of the slowdown in KAT. We uncover that the slowdown can be isolated to memory stalls and, more specifically, in the backward pass of GR-KAN caused by inefficient gradient accumulation. To address this memory bottleneck, we propose FlashKAT, which builds on our restructured kernel that minimizes gradient accumulation with atomic adds and accesses to slow memory. Evaluations demonstrate that FlashKAT can achieve a training speedup of 86.5x compared with the state-of-the-art KAT, while reducing rounding errors in the coefficient gradients. Our code is available at https://github.com/OSU-STARLAB/FlashKAT.
We introduce a novel resampling criterion using lift scores, for improving compositional generation in diffusion models. By leveraging the lift scores, we evaluate whether generated samples align with each single condition and then compose the results to determine whether the composed prompt is satisfied. Our key insight is that lift scores can be efficiently approximated using only the original diffusion model, requiring no additional training or external modules. We develop an optimized variant that achieves relatively lower computational overhead during inference while maintaining effectiveness. Through extensive experiments, we demonstrate that lift scores significantly improved the condition alignment for compositional generation across 2D synthetic data, CLEVR position tasks, and text-to-image synthesis. Our code is available at http://github.com/rainorangelemon/complift.
Multilingual alignment of sentence representations has mostly required bitexts to bridge the gap between languages. We investigate whether visual information can bridge this gap instead. Image caption datasets are very easy to create without requiring multilingual expertise, so this offers a more efficient alternative for low-resource languages. We find that multilingual image-caption alignment can implicitly align the text representations between languages, languages unseen by the encoder in pretraining can be incorporated into this alignment post-hoc, and these aligned representations are usable for cross-lingual Natural Language Understanding (NLU) and bitext retrieval.
Just-in-time Information Recommendation (JIR) is a service designed to deliver the most relevant information precisely when users need it, , addressing their knowledge gaps with minimal effort and boosting decision-making and efficiency in daily life. Advances in device-efficient deployment of foundation models and the growing use of intelligent wearable devices have made always-on JIR assistants feasible. However, there has been no systematic effort to formally define JIR tasks or establish evaluation frameworks. To bridge this gap, we present the first mathematical definition of JIR tasks and associated evaluation metrics. Additionally, we introduce JIR-Arena, a multimodal benchmark dataset featuring diverse, information-request-intensive scenarios to evaluate JIR systems across critical dimensions: i) accurately inferring user information needs, ii) delivering timely and relevant recommendations, and iii) avoiding irrelevant content that may distract users. Developing a JIR benchmark dataset poses challenges due to subjectivity in estimating user information needs and uncontrollable system variables affecting reproducibility. To address these, JIR-Arena: i) combines input from multiple humans and large AI models to approximate information need distributions; ii) assesses JIR quality through information retrieval outcomes using static knowledge base snapshots; and iii) employs a multi-turn, multi-entity validation framework to improve objectivity and generality. Furthermore, we implement a baseline JIR system capable of processing real-time information streams aligned with user inputs. Our evaluation of this baseline system on JIR-Arena indicates that while foundation model-based JIR systems simulate user needs with reasonable precision, they face challenges in recall and effective content retrieval. To support future research in this new area, we fully release our code and data.
While the capabilities of large language models (LLMs) have progressed significantly, their use in high-stakes applications have been limited due to risks of hallucination. One key approach in reducing hallucination is retrieval-augmented generation (RAG), but even in such setups, LLMs may still hallucinate when presented with questions outside of the knowledge base. Such behavior is unacceptable in high-stake applications where LLMs are expected to abstain from answering queries it does not have sufficient context on. In this work, we present a novel methodology for systematically evaluating out-of-knowledge base (OOKB) robustness of LLMs (whether LLMs know or do not know) in the RAG setting, without the need for manual annotation of gold standard answers. We implement our methodology in knowornot, an open-source library that enables users to develop their own customized evaluation data and pipelines for OOKB robustness. knowornot comprises four main features. Firstly, it provides a unified, high-level API that streamlines the process of setting up and running robustness benchmarks. Secondly, its modular architecture emphasizes extensibility and flexibility, allowing users to easily integrate their own LLM clients and RAG settings. Thirdly, its rigorous data modeling design ensures experiment reproducibility, reliability and traceability. Lastly, it implements a comprehensive suite of tools for users to customize their pipelines. We demonstrate the utility of knowornot by developing a challenging benchmark, PolicyBench, which spans four Question-Answer (QA) chatbots on government policies, and analyze its OOKB robustness. The source code of knowornot is available https://github.com/govtech-responsibleai/KnowOrNot.
Large Language Models (LLMs) demonstrate impressive capabilities but lack robust temporal intelligence, struggling to integrate reasoning about the past with predictions and plausible generations of the future. Meanwhile, existing methods typically target isolated temporal skills, such as question answering about past events or basic forecasting, and exhibit poor generalization, particularly when dealing with events beyond their knowledge cutoff or requiring creative foresight. To address these limitations, we introduce \textit{Time-R1}, the first framework to endow a moderate-sized (3B-parameter) LLM with comprehensive temporal abilities: understanding, prediction, and creative generation. Our approach features a novel three-stage development path; the first two constitute a \textit{reinforcement learning (RL) curriculum} driven by a meticulously designed dynamic rule-based reward system. This framework progressively builds (1) foundational temporal understanding and logical event-time mappings from historical data, (2) future event prediction skills for events beyond its knowledge cutoff, and finally (3) enables remarkable generalization to creative future scenario generation without any fine-tuning. Strikingly, experiments demonstrate that Time-R1 outperforms models over 200 times larger, including the state-of-the-art 671B DeepSeek-R1, on highly challenging future event prediction and creative scenario generation benchmarks. This work provides strong evidence that thoughtfully engineered, progressive RL fine-tuning allows smaller, efficient models to achieve superior temporal performance, offering a practical and scalable path towards truly time-aware AI. To foster further research, we also release \textit{Time-Bench}, a large-scale multi-task temporal reasoning dataset derived from 10 years of news data, and our series of \textit{Time-R1} checkpoints.
Recent advances in Large Language Models (LLMs) have demonstrated promising knowledge and reasoning abilities, yet their performance in multilingual and low-resource settings remains underexplored. Existing benchmarks often exhibit cultural bias, restrict evaluation to text-only, rely on multiple-choice formats, and, more importantly, are limited for extremely low-resource languages. To address these gaps, we introduce IRLBench, presented in parallel English and Irish, which is considered definitely endangered by UNESCO. Our benchmark consists of 12 representative subjects developed from the 2024 Irish Leaving Certificate exams, enabling fine-grained analysis of model capabilities across domains. By framing the task as long-form generation and leveraging the official marking scheme, it does not only support a comprehensive evaluation of correctness but also language fidelity. Our extensive experiments of leading closed-source and open-source LLMs reveal a persistent performance gap between English and Irish, in which models produce valid Irish responses less than 80\% of the time, and answer correctly 55.8\% of the time compared to 76.2\% in English for the best-performing model. We release IRLBench (https://huggingface.co/datasets/ReliableAI/IRLBench) and an accompanying evaluation codebase (https://github.com/ReML-AI/IRLBench) to enable future research on robust, culturally aware multilingual AI development.
Cognitive diagnosis (CD) plays a crucial role in intelligent education, evaluating students' comprehension of knowledge concepts based on their test histories. However, current CD methods often model students, exercises, and knowledge concepts solely on their ID relationships, neglecting the abundant semantic relationships present within educational data space. Furthermore, contemporary intelligent tutoring systems (ITS) frequently involve the addition of new students and exercises, a situation that ID-based methods find challenging to manage effectively. The advent of large language models (LLMs) offers the potential for overcoming this challenge with open-world knowledge. In this paper, we propose LLM4CD, which Leverages Large Language Models for Open-World Knowledge Augmented Cognitive Diagnosis. Our method utilizes the open-world knowledge of LLMs to construct cognitively expressive textual representations, which are then encoded to introduce rich semantic information into the CD task. Additionally, we propose an innovative bi-level encoder framework that models students' test histories through two levels of encoders: a macro-level cognitive text encoder and a micro-level knowledge state encoder. This approach substitutes traditional ID embeddings with semantic representations, enabling the model to accommodate new students and exercises with open-world knowledge and address the cold-start problem. Extensive experimental results demonstrate that our proposed method consistently outperforms previous CD models on multiple real-world datasets, validating the effectiveness of leveraging LLMs to introduce rich semantic information into the CD task.
Metaphors play a pivotal role in expressing emotions, making them crucial for emotional intelligence. The advent of multimodal data and widespread communication has led to a proliferation of multimodal metaphors, amplifying the complexity of emotion classification compared to single-mode scenarios. However, the scarcity of research on constructing multimodal metaphorical fine-grained emotion datasets hampers progress in this domain. Moreover, existing studies predominantly focus on English, overlooking potential variations in emotional nuances across languages. To address these gaps, we introduce a multimodal dataset in Chinese comprising 5,000 text-image pairs of metaphorical advertisements. Each entry is meticulously annotated for metaphor occurrence, domain relations and fine-grained emotion classification encompassing joy, love, trust, fear, sadness, disgust, anger, surprise, anticipation, and neutral. Our dataset is publicly accessible (https://github.com/DUTIR-YSQ/EmoMeta), facilitating further advancements in this burgeoning field.
Time series clustering promises to uncover hidden structural patterns in data with applications across healthcare, finance, industrial systems, and other critical domains. However, without validated ground truth information, researchers cannot objectively assess clustering quality or determine whether poor results stem from absent structures in the data, algorithmic limitations, or inappropriate validation methods, raising the question whether clustering is "more art than science" (Guyon et al., 2009). To address these challenges, we introduce CSTS (Correlation Structures in Time Series), a synthetic benchmark for evaluating the discovery of correlation structures in multivariate time series data. CSTS provides a clean benchmark that enables researchers to isolate and identify specific causes of clustering failures by differentiating between correlation structure deterioration and limitations of clustering algorithms and validation methods. Our contributions are: (1) a comprehensive benchmark for correlation structure discovery with distinct correlation structures, systematically varied data conditions, established performance thresholds, and recommended evaluation protocols; (2) empirical validation of correlation structure preservation showing moderate distortion from downsampling and minimal effects from distribution shifts and sparsification; and (3) an extensible data generation framework enabling structure-first clustering evaluation. A case study demonstrates CSTS's practical utility by identifying an algorithm's previously undocumented sensitivity to non-normal distributions, illustrating how the benchmark enables precise diagnosis of methodological limitations. CSTS advances rigorous evaluation standards for correlation-based time series clustering.
Large Vision-Language Models (VLMs) have shown strong capabilities in multimodal understanding and reasoning, yet they are primarily constrained by text-based reasoning processes. However, achieving seamless integration of visual and textual reasoning which mirrors human cognitive processes remains a significant challenge. In particular, effectively incorporating advanced visual input processing into reasoning mechanisms is still an open question. Thus, in this paper, we explore the interleaved multimodal reasoning paradigm and introduce DeepEyes, a model with "thinking with images" capabilities incentivized through end-to-end reinforcement learning without the need for cold-start SFT. Notably, this ability emerges natively within the model itself, leveraging its inherent grounding ability as a tool instead of depending on separate specialized models. Specifically, we propose a tool-use-oriented data selection mechanism and a reward strategy to encourage successful tool-assisted reasoning trajectories. DeepEyes achieves significant performance gains on fine-grained perception and reasoning benchmarks and also demonstrates improvement in grounding, hallucination, and mathematical reasoning tasks. Interestingly, we observe the distinct evolution of tool-calling behavior from initial exploration to efficient and accurate exploitation, and diverse thinking patterns that closely mirror human visual reasoning processes. Code is available at https://github.com/Visual-Agent/DeepEyes.
This paper presents a novel inertial localization framework named Egocentric Action-aware Inertial Localization (EAIL), which leverages egocentric action cues from head-mounted IMU signals to localize the target individual within a 3D point cloud. Human inertial localization is challenging due to IMU sensor noise that causes trajectory drift over time. The diversity of human actions further complicates IMU signal processing by introducing various motion patterns. Nevertheless, we observe that some actions observed through the head-mounted IMU correlate with spatial environmental structures (e.g., bending down to look inside an oven, washing dishes next to a sink), thereby serving as spatial anchors to compensate for the localization drift. The proposed EAIL framework learns such correlations via hierarchical multi-modal alignment. By assuming that the 3D point cloud of the environment is available, it contrastively learns modality encoders that align short-term egocentric action cues in IMU signals with local environmental features in the point cloud. These encoders are then used in reasoning the IMU data and the point cloud over time and space to perform inertial localization. Interestingly, these encoders can further be utilized to recognize the corresponding sequence of actions as a by-product. Extensive experiments demonstrate the effectiveness of the proposed framework over state-of-the-art inertial localization and inertial action recognition baselines.