Spoken Language Models (SLMs) aim to learn linguistic competence directly from speech using discrete units, widening access to Natural Language Processing (NLP) technologies for languages with limited written resources. However, progress has been largely English-centric due to scarce spoken evaluation benchmarks and training data, making cross-lingual learning difficult. We present a cross-lingual interleaving method that mixes speech tokens across languages without textual supervision. We also release an EN-FR training dataset, TinyStories (~42k hours), together with EN-FR spoken StoryCloze and TopicCloze benchmarks for cross-lingual semantic evaluation, both synthetically generated using GPT-4. On 360M and 1B SLMs under matched training-token budgets, interleaving improves monolingual semantic accuracy, enables robust cross-lingual continuation, and strengthens cross-lingual hidden-state alignment. Taken together, these results indicate that cross-lingual interleaving is a simple, scalable route to building multilingual SLMs that understand and converse across languages. All resources will be made open-source to support reproducibility.

Point cloud registration aligns multiple unposed point clouds into a common frame, and is a core step for 3D reconstruction and robot localization. In this work, we cast registration as conditional generation: a learned continuous, point-wise velocity field transports noisy points to a registered scene, from which the pose of each view is recovered. Unlike previous methods that conduct correspondence matching to estimate the transformation between a pair of point clouds and then optimize the pairwise transformations to realize multi-view registration, our model directly generates the registered point cloud. With a lightweight local feature extractor and test-time rigidity enforcement, our approach achieves state-of-the-art results on pairwise and multi-view registration benchmarks, particularly with low overlap, and generalizes across scales and sensor modalities. It further supports downstream tasks including relocal-ization, multi-robot SLAM, and multi-session map merging.

Large reasoning models (LRMs) extend large language models by generating explicit chain-of-thought (CoT) reasoning, significantly improving mathematical and logical problem solving. However, this explicit reasoning process also introduces new safety risks, as unsafe behaviors often emerge within intermediate reasoning trajectories, even when final answers appear harmless. Existing safety alignment approaches primarily rely on supervised fine-tuning (SFT) over safety-oriented long CoT datasets. While intuitive, we find that SFT produces inconsistent safety improvements, degrades reasoning ability, and generalizes poorly across model families. These limitations suggest that purely supervised approaches are insufficient for robust safety alignment in LRMs. To address this, we investigate reinforcement learning (RL) as a complementary optimization framework for LRM safety training. Unlike SFT, RL directly optimizes model policies with reward feedback, enabling more adaptive and stable alignment. Extensive experiments across multiple model families and benchmarks show that RL achieves stronger and more consistent safety gains while maintaining reasoning competence. Further analysis of reflection dynamics and token-level entropy reveals that RL suppresses unsafe exploratory reasoning while preserving reflective depth, leading to safer and more reliable reasoning processes.

Audio-based depression detection models have demonstrated promising performance but often suffer from gender bias due to imbalanced training data. Epidemiological statistics show a higher prevalence of depression in females, leading models to learn spurious correlations between gender and depression. Consequently, models tend to over-diagnose female patients while underperforming on male patients, raising significant fairness concerns. To address this, we propose a novel Counterfactual Debiasing Framework grounded in causal inference. We construct a causal graph to model the decision-making process and identify gender bias as the direct causal effect of gender on the prediction. During inference, we employ counterfactual inference to estimate and subtract this direct effect, ensuring the model relies primarily on authentic acoustic pathological features. Extensive experiments on the DAIC-WOZ dataset using two advanced acoustic backbones demonstrate that our framework not only significantly reduces gender bias but also improves overall detection performance compared to existing debiasing strategies.

Deep neural networks suffer from catastrophic forgetting, where performance on previous tasks degrades after training on a new task. This issue arises due to the model's tendency to overwrite previously acquired knowledge with new information. We present a novel approach to address this challenge, focusing on the intersection of memory-based methods and regularization approaches. We formulate a regularization strategy, termed Information Maximization (IM) regularizer, for memory-based continual learning methods, which is based exclusively on the expected label distribution, thus making it class-agnostic. As a consequence, IM regularizer can be directly integrated into various rehearsal-based continual learning methods, reducing forgetting and favoring faster convergence. Our empirical validation shows that, across datasets and regardless of the number of tasks, our proposed regularization strategy consistently improves baseline performance at the expense of a minimal computational overhead. The lightweight nature of IM ensures that it remains a practical and scalable solution, making it applicable to real-world continual learning scenarios where efficiency is paramount. Finally, we demonstrate the data-agnostic nature of our regularizer by applying it to video data, which presents additional challenges due to its temporal structure and higher memory requirements. Despite the significant domain gap, our experiments show that IM regularizer also improves the performance of video continual learning methods.

Current multimodal models aim to transcend the limitations of single-modality representations by unifying understanding and generation, often using text-to-image (T2I) tasks to calibrate semantic consistency. However, their reliance on static, single-image generation in training and evaluation leads to overfitting to static pattern matching and semantic fusion, while fundamentally hindering their ability to model dynamic processes that unfold over time. To address these constraints, we propose Envision-a causal event progression benchmark for chained text-to-multi-image generation. Grounded in world knowledge and structured by spatiotemporal causality, it reorganizes existing evaluation dimensions and includes 1,000 four-stage prompts spanning six scientific and humanities domains. To transition evaluation from single images to sequential frames and assess whether models truly internalize world knowledge while adhering to causal-temporal constraints, we introduce Envision-Score, a holistic metric integrating multi-dimensional consistency, physicality, and aesthetics. Comprehensive evaluation of 15 models (10 specialized T2I models, 5 unified models) uncovers: specialized T2I models demonstrate proficiency in aesthetic rendering yet lack intrinsic world knowledge. Unified multimodal models bridge this gap, consistently outperforming specialized counterparts in causal narrative coherence. However, even these unified architectures remain subordinate to closed-source models and struggle to overcome the core challenge of spatiotemporal consistency. This demonstrates that a focus on causally-isolated single images impedes multi-frame reasoning and generation, promoting static pattern matching over dynamic world modeling-ultimately limiting world knowledge internalization, generation.

As Large Language Models (LLMs) become integrated into high-stakes domains, there is a growing need for evaluation methods that are both scalable for real-time deployment and reliable for critical decision-making. While human evaluation is reliable, it is slow and costly. Single LLM judges are biased, and static juries lack adaptability. To overcome these limitations, we propose LLM Jury-on-Demand - a dynamic, learning-based framework for scalable and context-aware evaluation. Our method trains a set of reliability predictors to assess when LLM judges will agree with human experts, leveraging token distributions, embeddings, and structural input features. This enables a fully adaptive evaluation where, for each data point, an optimal jury of the most reliable judges is dynamically selected, and their scores are aggregated using their reliability as weights. Experiments on summarization and RAG benchmarks show that our dynamic jury system achieves significantly higher correlation with human judgment than both single-judge and static-jury baselines. These results highlight the promise of adaptive, learning-based juries for building scalable, more reliable and trustworthy evaluation systems for modern LLMs in high-stakes domains.

While reinforcement learning (RL) successfully enhances reasoning in large language models, its role in fostering compositional generalization (the ability to synthesize novel skills from known components) is often conflated with mere length generalization. To this end, we study what RL post-training teaches about skill composition and how the structure of the composition affects the skill transfer. We focus on the Countdown task (given n numbers and a target, form an expression that evaluates to the target) and analyze model solutions as expression trees, where each subtree corresponds to a reusable subtask and thus can be viewed as a ``skill.'' Tracking tree shapes and their success rates over training, we find: (i) out-of-distribution (OOD) generalization to larger n and to unseen tree shapes, indicating compositional reuse of subtasks; (ii) a structure-dependent hierarchy of learnability -- models master shallow balanced trees (workload is balanced between subtasks) before deep unbalanced ones, with persistent fragility on right-heavy structures (even when the composition depth is the same as some left-heavy structures). Our diagnostic reveals what is learned, in what order, and where generalization fails, clarifying how RL-only post-training induces OOD generalization beyond what standard metrics such as pass@k reveal.

Last-layer retraining (LLR) methods -- wherein the last layer of a neural network is reinitialized and retrained on a held-out set following ERM training -- have garnered interest as an efficient approach to rectify dependence on spurious correlations and improve performance on minority groups. Surprisingly, LLR has been found to improve worst-group accuracy even when the held-out set is an imbalanced subset of the training set. We initially hypothesize that this ``unreasonable effectiveness'' of LLR is explained by its ability to mitigate neural collapse through the held-out set, resulting in the implicit bias of gradient descent benefiting robustness. Our empirical investigation does not support this hypothesis. Instead, we present strong evidence for an alternative hypothesis: that the success of LLR is primarily due to better group balance in the held-out set. We conclude by showing how the recent algorithms CB-LLR and AFR perform implicit group-balancing to elicit a robustness improvement.

Integrated sensing and communication (ISAC) is a key enabler for low-altitude wireless networks (LAWNs), providing simultaneous environmental perception and data transmission in complex aerial scenarios. By combining heterogeneous sensing modalities such as visual, radar, lidar, and positional information, multimodal ISAC can improve both situational awareness and robustness of LAWNs. However, most existing multimodal fusion approaches use static fusion strategies that treat all modalities equally and cannot adapt to channel heterogeneity or time-varying modality reliability in dynamic low-altitude environments. To address this fundamental limitation, we propose a mixture-of-experts (MoE) framework for multimodal ISAC in LAWNs. Each modality is processed by a dedicated expert network, and a lightweight gating module adaptively assigns fusion weights according to the instantaneous informativeness and reliability of each modality. To improve scalability under the stringent energy constraints of aerial platforms, we further develop a sparse MoE variant that selectively activates only a subset of experts, thereby reducing computation overhead while preserving the benefits of adaptive fusion. Comprehensive simulations on three typical ISAC tasks in LAWNs demonstrate that the proposed frameworks consistently outperform conventional multimodal fusion baselines in terms of learning performance and training sample efficiency.