Active learning (AL) is a principled strategy to reduce annotation cost in data-hungry deep learning. However, existing AL algorithms focus almost exclusively on unimodal data, overlooking the substantial annotation burden in multimodal learning. We introduce the first framework for multimodal active learning with unaligned data, where the learner must actively acquire cross-modal alignments rather than labels on pre-aligned pairs. This setting captures the practical bottleneck in modern multimodal pipelines such as CLIP and SigLIP, where unimodal features are easy to obtain but high-quality alignment is costly. We develop a new algorithm that combines uncertainty and diversity principles in a modality-aware design, achieves linear-time acquisition, and applies seamlessly to both pool-based and streaming-based settings. Extensive experiments on benchmark datasets demonstrate that our approach consistently reduces multimodal annotation cost while preserving performance; for instance, on the ColorSwap dataset it cuts annotation requirements by up to $40\%$ without loss in accuracy.

Pruning is critical for scaling large language models (LLMs). Global pruning achieves strong performance but requires $\mathcal{O}(N)$ memory, which is infeasible for billion-parameter models. Local pruning reduces GPU memory usage to that of a single layer by pruning layers independently, but it neglects inter-layer dependencies and often leads to suboptimal performance in high-sparsity regimes. Unlike unstructured pruning, structured pruning produces regular sparsity patterns that align well with GPU kernels and library optimizations, making it more hardware-efficient. However, structured pruning typically relies on global pruning, since structured patterns are more prone to severe performance degradation under local optimization. To jointly achieve structured pruning and the memory efficiency of local pruning, we propose a divide-and-conquer strategy that decomposes the global pruning problem into coordinated subproblems across different modules, each of which fits within limited GPU memory. Building on this idea, we design \textbf{STRUPRUNE}, an ADMM-based framework that integrates structured sparsity into the pruning process, combining the memory efficiency of local pruning with the hardware compatibility of structured methods. We derive a closed-form analytical solution for structured pruning masks that provides an explicit rule for layer-wise sparsity allocation, and further develop an energy-based asymptotic framework yielding a softmax-form allocation scheme that simplifies optimization while adapting to heterogeneous layer importance. Experiments demonstrate that STRUPRUNE matches the perplexity of global structured pruning while reducing memory cost from $\mathcal{O}(N)$ to $\mathcal{O}(\sqrt{N})$, enabling practical deployment at the billion-parameter scale.

Ensuring the reliability of deep neural networks (DNNs) in the presence of real world noise and intentional perturbations remains a significant challenge. To address this, attribution methods have been proposed, though their efficacy remains suboptimal and necessitates further refinement. In this paper, we propose a novel category of transferable adversarial attacks, called transferable frequency-aware attacks, enabling frequency-aware exploration via both high-and low-frequency components. Based on this type of attacks, we also propose a novel attribution method, named Frequency-Aware Model Parameter Explorer (FAMPE), which improves the explainability for DNNs. Relative to the current state-of-the-art method AttEXplore, our FAMPE attains an average gain of 13.02% in Insertion Score, thereby outperforming existing approaches. Through detailed ablation studies, we also investigate the role of both high- and low-frequency components in explainability.

Foundation models (FMs) deployed in real-world tasks such as computer-use agents must integrate diverse modalities. How good are FMs at performing joint reasoning, simultaneously reasoning over multiple modalities, especially when the modalities interact and relate to each other to form cross-modal context? To better understand this problem, we study FMs on cross-modal conflicts: scenarios where conflicting evidence is presented across modalities. This allows us to examine whether FMs prioritize one modality over another or reason jointly to reconcile the conflict. Our experiments reveal that FMs can recognize conflicts in unimodal contexts, composed of a single modality, 90% of the time, but the ratio falls as low as 3% when evidence is split across modalities -- similar observations hold in cross-lingual contexts, composed of multiple languages. We trace this failure to cross-modal attention imbalance, showing that FMs exhibit extreme asymmetry in attention scores, disproportionately prioritizing certain modalities. We show that cross-modal attention imbalance does not go away by simply scaling up multimodal or multilingual datasets blindly, since they lack training examples that explicitly require cross-modal reasoning. We demonstrate that even a simple and scalable method of explicitly combining multiple modalities within each training instance significantly reduces attention imbalance. Reduced attention imbalance directly translates to improved downstream performance on several vision-language benchmarks. Our findings underscore the importance of systematically addressing cross-modal contexts to build reliable foundation models.

The impressive performance gains of modern language models currently rely on scaling parameters: larger models store more world knowledge and reason better. Yet compressing all world knowledge into parameters is unnecessary, as only a fraction is used per prompt, and impractical for edge devices with limited inference-time memory and compute. We address this shortcoming by a memory-augmented architecture and a pretraining strategy aligned with existing hardware paradigms. We introduce small language models that access large hierarchical parametric memory banks encoding world knowledge. During pretraining and inference, we fetch a small, context-dependent memory block and add it to the model. Our pretraining learns to store long-tail world knowledge in the memory parameters, while the small language model acts as an anchor capturing common knowledge and general reasoning abilities. Through trillion-token-scale experiments, we show significant gains: a 160M-parameters model augmented with an 18M-parameters memory fetched from a 4.6B memory bank obtains comparable performance to a regular model with more than 2x the parameters. Through extensive experiments, we study the optimal type and size of parametric memories in transformers, scaling them to over 21B parameters. We find that our proposed hierarchical feed-forward memories work robustly across transformer architectures, whether added during pretraining or post-hoc.

Information seeking is a fundamental requirement for humans. However, existing LLM agents rely heavily on open-web search, which exposes two fundamental weaknesses: online content is noisy and unreliable, and many real-world tasks require precise, domain-specific knowledge unavailable from the web. The emergence of the Model Context Protocol (MCP) now allows agents to interface with thousands of specialized tools, seemingly resolving this limitation. Yet it remains unclear whether agents can effectively leverage such tools -- and more importantly, whether they can integrate them with general-purpose search to solve complex tasks. Therefore, we introduce InfoMosaic-Bench, the first benchmark dedicated to multi-source information seeking in tool-augmented agents. Covering six representative domains (medicine, finance, maps, video, web, and multi-domain integration), InfoMosaic-Bench requires agents to combine general-purpose search with domain-specific tools. Tasks are synthesized with InfoMosaic-Flow, a scalable pipeline that grounds task conditions in verified tool outputs, enforces cross-source dependencies, and filters out shortcut cases solvable by trivial lookup. This design guarantees both reliability and non-triviality. Experiments with 14 state-of-the-art LLM agents reveal three findings: (i) web information alone is insufficient, with GPT-5 achieving only 38.2% accuracy and 67.5% pass rate; (ii) domain tools provide selective but inconsistent benefits, improving some domains while degrading others; and (iii) 22.4% of failures arise from incorrect tool usage or selection, highlighting that current LLMs still struggle with even basic tool handling.

Claim verification with large language models (LLMs) has recently attracted growing attention, due to their strong reasoning capabilities and transparent verification processes compared to traditional answer-only judgments. However, existing approaches to online claim verification, which requires iterative evidence retrieval and reasoning, still mainly rely on prompt engineering or pre-designed reasoning workflows, without unified training to improve necessary skills. Therefore, we introduce Veri-R1, an online reinforcement learning (RL) framework that enables an LLM to interact with a search engine and to receive reward signals that explicitly shape its planning, retrieval, and reasoning behaviors. This dynamic interaction of LLM with retrieval systems more accurately reflects real-world verification scenarios and fosters comprehensive verification skills. Empirical results show that Veri-R1 improves joint accuracy by up to 30% and doubles the evidence score, often surpassing its larger-scale model counterparts. Ablation studies further reveal the impact of reward components, and the link between output logits and label accuracy. Our results highlight the effectiveness of online RL for precise and faithful claim verification, providing an important foundation for future research. We release our code to support community progress in LLM empowered claim verification.

Model checking is a key technique for verifying safety-critical systems against formal specifications, where recent applications of deep learning have shown promise. However, while ubiquitous for vision and language domains, representation learning remains underexplored in formal verification. We introduce Contrastive Neural Model Checking (CNML), a novel method that leverages the model checking task as a guiding signal for learning aligned representations. CNML jointly embeds logical specifications and systems into a shared latent space through a self-supervised contrastive objective. On industry-inspired retrieval tasks, CNML considerably outperforms both algorithmic and neural baselines in cross-modal and intra-modal settings. We further show that the learned representations effectively transfer to downstream tasks and generalize to more complex formulas. These findings demonstrate that model checking can serve as an objective for learning representations for formal languages.

Breast magnetic resonance imaging is a critical tool for cancer detection and treatment planning, but its clinical utility is hindered by poor specificity, leading to high false-positive rates and unnecessary biopsies. This study introduces a transformer-based framework for automated classification of breast lesions in dynamic contrast-enhanced MRI, addressing the challenge of distinguishing benign from malignant findings. We implemented a SegFormer architecture that achieved an AUC of 0.92 for lesion-level classification, with 100% sensitivity and 67% specificity at the patient level - potentially eliminating one-third of unnecessary biopsies without missing malignancies. The model quantifies malignant pixel distribution via semantic segmentation, producing interpretable spatial predictions that support clinical decision-making. To establish reproducible benchmarks, we curated BreastDCEDL_AMBL by transforming The Cancer Imaging Archive's AMBL collection into a standardized deep learning dataset with 88 patients and 133 annotated lesions (89 benign, 44 malignant). This resource addresses a key infrastructure gap, as existing public datasets lack benign lesion annotations, limiting benign-malignant classification research. Training incorporated an expanded cohort of over 1,200 patients through integration with BreastDCEDL datasets, validating transfer learning approaches despite primary tumor-only annotations. Public release of the dataset, models, and evaluation protocols provides the first standardized benchmark for DCE-MRI lesion classification, enabling methodological advancement toward clinical deployment.

The Muon optimizer is consistently faster than Adam in training Large Language Models (LLMs), yet the mechanism underlying its success remains unclear. This paper demystifies this mechanism through the lens of associative memory. By ablating the transformer components optimized by Muon, we reveal that the associative memory parameters of LLMs, namely the Value and Output (VO) attention weights and Feed-Forward Networks (FFNs), are the primary contributors to Muon's superiority. Motivated by this associative memory view, we then explain Muon's superiority on real-world corpora, which are intrinsically heavy-tailed: a few classes (tail classes) appear far less frequently than others. The superiority is explained through two key properties: (i) its update rule consistently yields a more isotropic singular spectrum than Adam; and as a result, (ii) on heavy-tailed data, it optimizes tail classes more effectively than Adam. Beyond empirical evidence, we theoretically confirm these findings by analyzing a one-layer associative memory model under class-imbalanced data. We prove that Muon consistently achieves balanced learning across classes regardless of feature embeddings, whereas Adam can induce large disparities in learning errors depending on embedding properties. In summary, our empirical observations and theoretical analyses reveal Muon's core advantage: its update rule aligns with the outer-product structure of linear associative memories, enabling more balanced and effective learning of tail classes in heavy-tailed distributions than Adam.