Hao Li, Yizheng Sun, Viktor Schlegel, Kailai Yang, Riza Batista-Navarro, Goran Nenadic. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Fiber-optic distributed acoustic sensor (DAS) offers great potential for railway event monitoring due to its high sensitivity and robustness in complex environments. However, accurate recognition of acoustic events remains challenging, under real-time constraints where only short-duration signal segments with limited discriminative information are available. To overcome this, an efficient recognition framework was proposed by integrating multi-encoding image fusion and a lightweight transformer-convolutional network (CNN). Specifically, 0.032-second DAS signal segments were converted into complementary image representations, which were then fused into RGB images to enhance feature diversity. These multimodal images were processed by a compact backbone that combined vision transformer (ViT) modules with convolutional components, enabling effective extraction of multi-scale local and global features. The hybrid framework effectively mitigated the lack of temporal information in short segments, achieving 98.19% accuracy with only 1.33 million parameters and 3.3 ms latency per sample, substantially faster than existing methods. Compared to baseline models, the proposed architecture achieved superior accuracy and compactness, making it highly suitable for real-time and edge deployment in DAS-based monitoring systems.

Vision-Language Models (VLMs) are powerful yet computationally intensive for widespread practical deployments.To address such challenge without costly re-training, post-training acceleration techniques like quantization and token reduction are extensively explored.However, current acceleration evaluations primarily target minimal overall performance degradation, overlooking a crucial question: does the accelerated model still give the same answers to the same questions as it did before acceleration?This is vital for stability-centered industrial applications where consistently correct answers for specific, known situations are paramount, such as in AI-based disease diagnosis.We systematically investigate this for accelerated VLMs, testing four leading models (LLaVA-1.5,LLaVA-Next, Qwen2-VL, Qwen2.5-VL) with eight acceleration methods on ten multimodal benchmarks.Our findings are stark: despite minimal aggregate performance drops, accelerated models changed original answers up to 20% of the time.Critically, up to 6.5% of these changes converted correct answers to incorrect.Input perturbations magnified these inconsistencies, and the trend is confirmed by case studies with the medical VLM LLaVA-Med.This research reveals a significant oversight in VLM acceleration, stressing an urgent need for instance-level stability checks to ensure trustworthy real-world deployment.

A unified foundation model for medical time series—pretrained on open access and ethically reviewed medical corpora—offers the potential to reduce annotation burdens, minimize model customization, and enable robust transfer across clinical institutions, modalities, and tasks, particularly in data-scarce or privacy-constrained environments. However, existing time series foundation models struggle to handle medical time series data due to its inherent challenges, including irregular intervals, heterogeneous sampling rates, and frequent missingness. To address these challenges, we introduce MIRA, a unified foundation model specifically designed for medical time series forecasting. MIRA incorporates a Continuous-Time Rotary Positional Encoding that enables fine-grained modeling of variable time intervals, a frequency-specific mixture-of-experts layer that routes computation across latent frequency regimes to further promote temporal specialization, and a Continuous Dynamics Extrapolation Block based on Neural ODE that models the continuous trajectory of latent states, enabling accurate forecasting at arbitrary target timestamps. Pretrained on a large-scale and diverse medical corpus comprising over 454 billion time points collect from publicly available datasets, MIRA achieving reductions in forecasting errors by an average of 8% and 6% in out-of-distribution and in-distribution scenarios, respectively. We also introduce a comprehensive benchmark spanning multiple downstream clinical tasks, establishing a foundation for future research in medical time series modeling.