在泰国曼谷举行的ACL 2024顶级会议是本周学术界的一大焦点，吸引了全球众多顶尖研究者的目光。会议期间，字节跳动公司不仅展现了其在人工智能领域的强劲实力，还成为了会议的一大亮点。
根据官方数据，本届ACL共收到近5000篇论文投稿，经过严格的评审，最终有940篇论文被主会录用，其中168篇工作入选大会口头报告。录取率低于3.4%，而字节跳动公司有5篇研究成果入选口头报告，表现突出。
在8月14日的Paper Awards环节，字节跳动的《G-DIG: Towards Gradient-based DIverse and high-quality Instruction Data Selection for Machine Translation》论文被评选为Outstanding Paper之一，这是ACL成立59年来，中国科学家团队第二次获得这一殊荣。
为了深入探讨今年的前沿研究成果，字节跳动公司将于8月20日下周二19:00-21:00，举办线上直播的「字节跳动ACL 2024前沿论文分享会」。届时，豆包大语言模型研究团队负责人王明轩，将携手多位研究员，分享部分中选成果，涉及自然语言处理、语音处理、多模态学习、大模型推理等领域。
此外，字节跳动公司还将在活动中分享两项创新研究：RepCodec和DINOISER。RepCodec是一种用于语音离散化的语音表示编解码器，旨在提高离散语音标记的性能，并通过增强信息保留能力，在语音理解和生成方面取得显著效果。而DINOISER则通过噪声操纵增强的扩散条件序列生成模型，解决了文本扩散模型在生成离散序列数据时的挑战，并在多个条件序列建模基准上取得了优异的成绩。
这些研究成果不仅展示了字节跳动公司在人工智能领域的创新实力，也为未来的语音处理和语言模型研究提供了新的思路和方法。
英语如下：
News Title: “ByteDance Continues to Achieve Great Success at ACL Conference”
Keywords: ACL Summit, ByteDance, Outstanding Research
News Content: The ACL 2024 top conference, held in Bangkok, Thailand, was a major focus of the academic world this week, drawing the attention of global top researchers. During the conference, ByteDance not only showcased its strong strength in the field of artificial intelligence but also became a highlight of the event.
According to official data, the conference received nearly 5,000 paper submissions, and after strict review, a total of 940 papers were accepted for the main meeting, with 168 works selected for oral presentations. The acceptance rate was below 3.4%, and ByteDance had 5 research achievements selected for oral presentations, performing exceptionally well.
On August 14, during the Paper Awards segment, ByteDance’s paper “G-DIG: Towards Gradient-based DIverse and high-quality Instruction Data Selection for Machine Translation” was selected as one of the Outstanding Papers. This is the second time in the 59-year history of ACL that a Chinese scientist team has received this honor.
To delve into this year’s cutting-edge research achievements, ByteDance will host a live online “ByteDance ACL 2024 Frontier Papers Sharing” event on August 20 from 19:00 to 21:00. At the event, Wang Mingxuan, the head of the Big Language Model Research Team, will share selected outcomes with multiple researchers, covering areas such as natural language processing, speech processing, multimodal learning, and large model inference.
In addition, ByteDance will also share two innovative studies: RepCodec and DINOISER. RepCodec is a voice discretization voice representation encoder-decoder, aimed at improving the performance of discrete voice markers and achieving significant results in voice understanding and generation by enhancing information retention capability. DINOISER solves the challenge of generating discrete sequence data in text diffusion models by using a noise manipulation enhanced diffusion conditional sequence generation model, and has achieved outstanding results on multiple conditional sequence modeling benchmarks.
These research achievements not only showcase ByteDance’s innovative strength in the field of artificial intelligence but also provide new ideas and methods for future speech processing and language model research.
【来源】https://www.jiqizhixin.com/articles/2024-08-15-5
Views: 3