Zobrazeno 1 - 10
of 25 549
pro vyhledávání: '"feng Yang"'
Weak supervision combines the advantages of training on real data with the ability to exploit signal properties. However, training a neural network using weak supervision often requires an excessive amount of signal data, which severely limits its pr
Externí odkaz:
http://arxiv.org/abs/2412.00198
Iterative retrieval refers to the process in which the model continuously queries the retriever during generation to enhance the relevance of the retrieved knowledge, thereby improving the performance of Retrieval-Augmented Generation (RAG). Existing
Externí odkaz:
http://arxiv.org/abs/2411.19443
Streaming generation models are increasingly utilized across various fields, with the Transducer architecture being particularly popular in industrial applications. However, its input-synchronous decoding mechanism presents challenges in tasks requir
Externí odkaz:
http://arxiv.org/abs/2411.17170
Inference acceleration of large language models (LLMs) has been put forward in many application scenarios and speculative decoding has shown its advantage in addressing inference acceleration. Speculative decoding usually introduces a draft model to
Externí odkaz:
http://arxiv.org/abs/2412.00061
Autor:
Zhang, Shaolei, Zhang, Kehao, Fang, Qingkai, Guo, Shoutao, Zhou, Yan, Liu, Xiaodong, Feng, Yang
Large language models (LLMs), with their powerful generative capabilities and vast knowledge, empower various tasks in everyday life. However, these abilities are primarily concentrated in high-resource languages, leaving low-resource languages with
Externí odkaz:
http://arxiv.org/abs/2411.16300
Reconstructing the 3D anatomical structures of the oral cavity, which originally reside in the cone-beam CT (CBCT), from a single 2D Panoramic X-ray(PX) remains a critical yet challenging task, as it can effectively reduce radiation risks and treatme
Externí odkaz:
http://arxiv.org/abs/2411.03725
Byte-based machine translation systems have shown significant potential in massively multilingual settings. Unicode encoding, which maps each character to specific byte(s), eliminates the emergence of unknown words, even in new languages, enabling br
Externí odkaz:
http://arxiv.org/abs/2411.01474
Autor:
Wu, Hsin-Yeh, Besançon, Marc, Chen, Jia-Wern, Chen, Pisin, Glicenstein, Jean-François, Liu, Shu-Xiao, Lu, Yu-Jung, Navick, Xavier-François, Paganis, Stathes, Tuchming, Boris, Tsionou, Dimitra, Tsai, Feng-Yang
A dual-operation mode SNSPD is demonstrated. In the conventional Geiger SNSPD mode the sensor operates at temperatures well below the critical temperature, Tc, working as an event counter without sensitivity to the number of photons impinging the sen
Externí odkaz:
http://arxiv.org/abs/2410.10280
Models like GPT-4o enable real-time interaction with large language models (LLMs) through speech, significantly enhancing user experience compared to traditional text-based interaction. However, there is still a lack of exploration on how to build sp
Externí odkaz:
http://arxiv.org/abs/2409.06666
The Rust programming language has garnered significant attention due to its robust safety features and memory management capabilities. Despite its guaranteed memory safety, Rust programs suffer from runtime errors that are unmanageable, i.e., panic e
Externí odkaz:
http://arxiv.org/abs/2408.03262