Sponsor
Statistical Science
Department of Statistical Science Friday Seminar - Homecoming Edition
Friday, September 18,
-
Speaker(s): Jerry Reiter and Yue Jiang
Speaker(s):
Prof. Jerry Reiter, Henry W. Newson Distinguished Professor of Statistical Science ('92 Mathematics):
Title: Is There a Future for Public Use Data?
Abstract: For decades, research and education in the social sciences have been facilitated by record-level, public use data files provided by statistical agencies and other data stewards. Today, however, the future of public use data is uncertain. The growth of readily available digital information and computational tools, while enormously beneficial for advancing research and education, has led to increased risks that ill-intentioned data users could mine the public use data to learn data subjects' identifies and sensitive information. I discuss the future of public use data in the social sciences, focusing on the risks to data subjects' privacy and confidentiality. Along the way, I discuss techniques that statistical agencies have used to reduce disclosure risks in public use data files. I conclude with a vision for the future of access to confidential social science data.
Prof. Yue Jiang, Associate Professor of the Practice of Statistical Science ('12 Statistical Science and Earth & Ocean Science)
Title: Computational methods to support creative musical decision-making
Abstract: I'll briefly present two projects (one with a current 2nd year MSS students and one with an alumnus) that use computational methods to support musical decision-making and creation. The first project looks at how complex polyphonic music (e.g., symphonies, piano scores) can be automatically adapted for monophonic instruments such as flute or saxophone. The challenge here is to reduce many notes and musical lines to a single line without losing what makes the original piece recognizable; our method is informed by music theory, additionally incorporating a graph-based optimization approach to determine which notes and motifs should be retained. The second project incorporates elements of music theory in a diffusion model to inform AI generation of music, additionally allowing for human control regarding harmony and melody. The use of music theoretic principles allows our method to use orders of magnitude fewer parameters to generate music that is comparable to current state-of-the-art methods and even human-composer work as evaluated in listener experiments.
Prof. Jerry Reiter, Henry W. Newson Distinguished Professor of Statistical Science ('92 Mathematics):
Title: Is There a Future for Public Use Data?
Abstract: For decades, research and education in the social sciences have been facilitated by record-level, public use data files provided by statistical agencies and other data stewards. Today, however, the future of public use data is uncertain. The growth of readily available digital information and computational tools, while enormously beneficial for advancing research and education, has led to increased risks that ill-intentioned data users could mine the public use data to learn data subjects' identifies and sensitive information. I discuss the future of public use data in the social sciences, focusing on the risks to data subjects' privacy and confidentiality. Along the way, I discuss techniques that statistical agencies have used to reduce disclosure risks in public use data files. I conclude with a vision for the future of access to confidential social science data.
Prof. Yue Jiang, Associate Professor of the Practice of Statistical Science ('12 Statistical Science and Earth & Ocean Science)
Title: Computational methods to support creative musical decision-making
Abstract: I'll briefly present two projects (one with a current 2nd year MSS students and one with an alumnus) that use computational methods to support musical decision-making and creation. The first project looks at how complex polyphonic music (e.g., symphonies, piano scores) can be automatically adapted for monophonic instruments such as flute or saxophone. The challenge here is to reduce many notes and musical lines to a single line without losing what makes the original piece recognizable; our method is informed by music theory, additionally incorporating a graph-based optimization approach to determine which notes and motifs should be retained. The second project incorporates elements of music theory in a diffusion model to inform AI generation of music, additionally allowing for human control regarding harmony and melody. The use of music theoretic principles allows our method to use orders of magnitude fewer parameters to generate music that is comparable to current state-of-the-art methods and even human-composer work as evaluated in listener experiments.
