This paper explores advancements in automated Question-Answer (QA) extraction using large language models (LLMs), addressing challenges in transforming unstructured text into high-quality, retrievable QA pairs. Traditional approaches, whether through segmented question and answer generation or end-to-end extraction, often struggle with efficiency, dataset limitations, and performance consistency. Leveraging recent progress in LLMs, we constructed a large-scale Chinese QA extraction dataset with 143,846 documents and evaluated multiple fine-tuned models on public and private datasets. Surprisingly, code-based English LLMs outperformed Chinese-specialized models on Chinese text with a lower hallucination rate. Building upon this finding, we enhanced the best-performing code-based model with an expanded Chinese vocabulary, creating Code Llama-M, which achieved better results. Integrating Code Llama-M into our internal assistant, Luo Ying, demonstrated notable user satisfaction gains, affirming its practical impact. Key contributions include: (i) creation of a robust Chinese QA extraction instruction dataset; (ii) evidence of cross-lingual efficacy of code-based LLMs for Chinese QA tasks, further enhanced through Code Llama-M's expanded Chinese vocabulary; and (iii) successful application of the fine-tuned LLM in a live assistant system, enhancing user experience.

With SSDs gradually replacing HDDs as the main-stream storage media in modern large-scale data centers, SSD failure analysis has become increasingly important. We conducted an in-depth data-driven analysis of the failure characteristics of an SSD-based data center in Alibaba based on the failure datasets in 2018 and 2019 and SMART logs on December 31, 2019. Our objectives focus on 3 “W”s, illustrated as follows. What factors influence the occurrence of failures? Why do these factors influence the occurrence of failures? How can companies reduce the occurrence of SSD failures? We expect that our findings and analysis can benefit future SSD-based storage system designs.

With the advancements of machine learning and AI technologies, robots have been more widely used in our everyday life and they have also been used in education. The present study introduces a 12-week child-robot theater afterschool program designed to promote science, technology, engineering, and mathematics (STEM) education with art elements (STEAM) for elementary students using social robots. Four modules were designed to introduce robot mechanisms as well as arts: Acting (anthropomorphism), Dance (robot movements), Music and Sounds (music composition), and Drawing (robot art). These modules provided children with basic knowledge about robotics and STEM and guided children to create a live robot theater play. A total of 16 students participated in the program, and 11 of them were involved in completing questionnaires and interviews regarding their perceptions towards robots, STEAM, and the afterschool program. Four afterschool program teachers participated in interviews, reflecting their perceptions of the program and observations of children’s experiences during the program. Our findings suggest that the present program effectively maintained children’s engagement and improved their interest in STEAM by connecting social robots and theater production. We conclude with design guidelines and recommendations for future research and programs.

As robots have become more pervasive in our everyday life, social aspects of robots have attracted researchers’ attention. Because emotions play a crucial role in social interactions, research has been conducted on conveying emotions via speech. Our study sought to investigate the synchronization of multimodal interaction in human-robot interaction (HRI). We conducted a within-subjects exploratory study with 40 participants to investigate the effects of non-speech sounds (natural voice, synthesized voice, musical sound, and no sound) and basic emotions (anger, fear, happiness, sadness, and surprise) on user perception with emotional body gestures of an anthropomorphic robot (Pepper). While listening to a fairytale with the participant, a humanoid robot responded to the story with recorded emotional non-speech sounds and gestures. Participants showed significantly higher emotion recognition accuracy from the natural voice than from other sounds. The confusion matrix showed that happiness and sadness had the highest emotion recognition accuracy, which is in line with previous research. The natural voice also induced higher trust, naturalness, and preference compared to other sounds. Interestingly, the musical sound mostly showed lower perception ratings, even compared to no sound. Results are discussed with design guidelines for emotional cues from social robots and future research directions.

The use of social robots has recently been investigated in various areas, including STEM (Science, Technology, Engineering, and Mathematics) education and artistic performances. To inform children of the seriousness of climate change and awareness that they can make change, we created the Robot Musical Theater performance. In this project, natural elements (wind, earth, fire, and water) were anthropomorphized and represented by humanoid robots (Pepper, Milo, and Nao). The robots were designed to motivate audience to participate in the action to prevent climate change. Because of COVID, only fourteen visitors as a single group were allowed to participate in real-time and posted to YouTube, where at the time of submission, 141 people have viewed the performance. The participants provided positive comments on the performance and showed their willingness to participate in the movement to prevent climate change, and expressed their further interest in STEM learning. This performance is expected to contribute to enhancing informal STEM and robotics learning, as well as advancing robotic arts.

Occluded faces-as a common scene in real life-have a significant negative impact on most face recognition systems. Existing methods try to remove the occlusions by a single-stage generative adversarial network (GAN), which is unaware of the occlusion and thus has difficulties in generalizing to a large variety of occlusion types, e.g., different objects at various positions. To this end, we propose the two-stage Occlusion-Aware GAN (OA-GAN), where the first GAN is for disentangling the occlusions, which will be served as the additional input of the second GAN for synthesizing the final de-occluded faces. In this way, our two-stage model can handle diverse occlusions in the wild and is naturally more explainable because of its awareness of the occluded objects. Extensive experiments on both synthetic and real-world datasets validate the superiority of the two-stage OAGAN design. Furthermore, by applying the generated de-occluded faces to facial expression recognition (FER) systems, we find that our two-stage de-occlusion process significantly increases the accuracy of FER under occlusion.