Vector Institute Researchers Present Over 65 Papers at NeurIPS 2023
Researchers from the Vector Institute for Artificial Intelligence are presenting more than 65 papers at the 2023 Conference on Neural Information Processing Systems (NeurIPS). The conference, held in New Orleans and online from December 10 to 16, features work by Vector Faculty, Affiliates, and Postdoctoral Fellows. Their research spans diverse AI applications with potential impacts on health, chemical materials discovery, data privacy, music, and biodiversity. Key highlights include the introduction of the BIOSCAN-1M Insect Dataset, a large-scale resource for biodiversity assessment using computer vision and genomic data. Another significant contribution is GRACE, a lifelong model editing method that allows AI models to be fine-tuned without disrupting their overall structure, addressing issues of outdated or unexpected behavior. Additionally, the team presented AlpacaFarm, a simulation framework designed to reduce the high costs and inconsistencies associated with collecting human feedback for large language models. By simulating human responses, AlpacaFarm offers a cost-effective alternative for training and evaluating AI systems. These presentations underscore Vector Institute's commitment to advancing artificial intelligence research across various domains, demonstrating innovations that address both technical challenges in machine learning and practical applications in science and daily life.
Wire timeline
Vector Institute Researchers Present Over 65 Papers at NeurIPS 2023
Researchers from the Vector Institute for Artificial Intelligence are presenting more than 65 papers at the 2023 Conference on Neural Information Processing Systems (NeurIPS). The conference, held in New Orleans and online from December 10 to 16, features work by Vector Faculty, Affiliates, and Postdoctoral Fellows. Their research spans diverse AI applications with potential impacts on health, chemical materials discovery, data privacy, music, and biodiversity. Key highlights include the introduction of the BIOSCAN-1M Insect Dataset, a large-scale resource for biodiversity assessment using computer vision and genomic data. Another significant contribution is GRACE, a lifelong model editing method that allows AI models to be fine-tuned without disrupting their overall structure, addressing issues of outdated or unexpected behavior. Additionally, the team presented AlpacaFarm, a simulation framework designed to reduce the high costs and inconsistencies associated with collecting human feedback for large language models. By simulating human responses, AlpacaFarm offers a cost-effective alternative for training and evaluating AI systems. These presentations underscore Vector Institute's commitment to advancing artificial intelligence research across various domains, demonstrating innovations that address both technical challenges in machine learning and practical applications in science and daily life.
Vector Institute for Artificial Intelligence