Parameter-Efficient Retrievers for Polish and European Languages
本文提出了一种三阶段训练流程,通过跨语言对齐、关系知识蒸馏和对比微调来开发紧凑高效的检索器,以解决大规模索引、频繁语料更新及低延迟服务成本高的问题。
本文提出了一种三阶段训练流程,通过跨语言对齐、关系知识蒸馏和对比微调来开发紧凑高效的检索器,以解决大规模索引、频繁语料更新及低延迟服务成本高的问题。
研究通过引入波兰ModernBERT模型,解决了波兰语理解任务中的长文本处理问题,使用改进的预训练方法和基准测试,在多个任务上取得了最佳性能。
本文提出PUMA,一个包含900个任务的基准测试集,旨在评估多模态模型在波兰文化和语言背景下的表现,涵盖文本、图像、音频和文档处理。
This work addresses the limitation of Polish BERT-style encoders in processing long documents due to their short context windows. We propose a two-stage training strategy: first extending positional embeddings to support a context length of 8,192 tokens, followed by full-parameter continued pretraining; we further distill this model into a lightweight variant via knowledge distillation. To our knowledge, this is the first high-performance long-context encoder for Polish, accompanied by FinBench—a newly introduced benchmark comprising long financial documents. Evaluated across 25 tasks, including KLEJ and FinBench, our model outperforms existing Polish and multilingual baselines on average, demonstrating substantial gains on long-context tasks while preserving strong performance on short-text understanding.
This study addresses the limited accessibility of standard interfaces (e.g., voice, touch) for older adults and people with disabilities in smart home environments. We propose a non-invasive, multimodal biosignal interaction paradigm integrating surface electromyography (EMG), electrooculography (EOG), and speech signals. A lightweight AI model enables real-time intent recognition, while participatory design iteratively refines human–machine collaboration logic, implemented in the Sagacity prototype system. Our key contribution is the first integration of low-cost EMG/EOG sensing with edge AI to deliver a contactless, low-cognitive-load control solution tailored to vulnerable users. Experimental evaluation in realistic domestic settings confirms feasibility and robustness, identifies critical technical bottlenecks, and elicits authentic user requirements. The work provides a reproducible design framework and empirical evidence for accessible intelligent interaction.
本文提出了一种三阶段训练流程,通过跨语言对齐、关系知识蒸馏和对比微调来开发紧凑高效的检索器,以解决大规模索引、频繁语料更新及低延迟服务成本高的问题。
研究通过引入波兰ModernBERT模型,解决了波兰语理解任务中的长文本处理问题,使用改进的预训练方法和基准测试,在多个任务上取得了最佳性能。
本文提出PUMA,一个包含900个任务的基准测试集,旨在评估多模态模型在波兰文化和语言背景下的表现,涵盖文本、图像、音频和文档处理。
This work addresses the limitation of Polish BERT-style encoders in processing long documents due to their short context windows. We propose a two-stage training strategy: first extending positional embeddings to support a context length of 8,192 tokens, followed by full-parameter continued pretraining; we further distill this model into a lightweight variant via knowledge distillation. To our knowledge, this is the first high-performance long-context encoder for Polish, accompanied by FinBench—a newly introduced benchmark comprising long financial documents. Evaluated across 25 tasks, including KLEJ and FinBench, our model outperforms existing Polish and multilingual baselines on average, demonstrating substantial gains on long-context tasks while preserving strong performance on short-text understanding.
This study addresses the limited accessibility of standard interfaces (e.g., voice, touch) for older adults and people with disabilities in smart home environments. We propose a non-invasive, multimodal biosignal interaction paradigm integrating surface electromyography (EMG), electrooculography (EOG), and speech signals. A lightweight AI model enables real-time intent recognition, while participatory design iteratively refines human–machine collaboration logic, implemented in the Sagacity prototype system. Our key contribution is the first integration of low-cost EMG/EOG sensing with edge AI to deliver a contactless, low-cognitive-load control solution tailored to vulnerable users. Experimental evaluation in realistic domestic settings confirms feasibility and robustness, identifies critical technical bottlenecks, and elicits authentic user requirements. The work provides a reproducible design framework and empirical evidence for accessible intelligent interaction.