NeurIPS 2018 Breaks Attendance and Submission Records in Montreal
The 32nd NeurIPS conference draws a record 8,300 attendees and accepts 1,010 papers out of nearly 5,000 submissions. Reinforcement learning and generative models dominate the event, while speech processing earns notable spotlight recognition.
The 32nd Conference on Neural Information Processing Systems (NeurIPS) takes place in Montreal, Canada, attracting a record-breaking 8,300 registered attendees. The conference officially transitions its acronym from NIPS to NeurIPS and receives an unprecedented 4,854 paper submissions, ultimately accepting 1,010 papers for an acceptance rate of 20.8 percent. The massive event spans six days and features a tutorial day, three days of main conference sessions, and two days of around 40 parallel workshops.
The Test of Time Award goes to Léon Bottou and Olivier Bousquet for their 2007 paper on the tradeoffs of large-scale learning, which highlights the effectiveness of stochastic gradient descent (SGD). Over a thousand accepted papers cover a wide array of AI fields, with reinforcement learning, generative adversarial learning, and optimization emerging as the most active research topics. Transfer learning, unsupervised learning, causal learning, and network interpretability also represent strong themes throughout the various presentations.
Computer vision leads the application areas, but the speech processing community maintains a solid presence with several noteworthy contributions. Baidu earns a NeurIPS spotlight for its paper on neural voice cloning with minimal samples, while other speech papers explore time-domain music generation, multi-speaker text-to-speech synthesis, and on-device automatic speech recognition. Additionally, two dedicated workshops on audio interpretability and conversational AI bring together diverse research communities to discuss the future of voice technology.