Kristian Lum
Kristian Lum | |||
|---|---|---|---|
| |||
| Nationality | American | ||
| Citizenship | United States of America | ||
| Alma mater | Duke University | ||
| Known for | Algorithmic bias Fairness (machine learning) | ||
| Scientific career | |||
| Fields | Statistics Machine Learning | ||
| Institutions | Twitter Human Rights Data Analysis Group University of Pennsylvania | ||
| Doctoral advisor | Alan E. Gelfand | ||
Kristian Lum is an American machine learning researcher with a focus on responsible use of AI systems, particularly predictive models used in a criminal justice setting. She is founding member of the Association for Computing Machinery Conference on Fairness, Accountability, and Transparency.[1] Lum received a PhD from the Department of Statistical Science at Duke University.[2] Lum is currently a Senior Staff Machine Learning Researcher at Twitter.[3] She was previously the Lead Statistician at the Human Rights Data Analysis Group[4] and Research Assistant Professor at University of Pennsylvania. [5]
Career and research
Much of Lum’s work concerns predictive policing models, including the predictive policing ProPublica COMPAS (software), and how to incorporate fairness into machine learning algorithms and avoid machine learning models amplifying bias that exists in the training data. Her research looks at the statistical framework needed for fair predictive algorithms[6]. This method removed bias through removing all information regarding protected variables from the permitted training data. This approach was novel and more effective than the existing common approach. The most common approach to creating neutral models was to omit race as a covariate, but that approach still leads to racially disparate predictions when including other variables with a strong correlation to race. Lum’s suggested approach showed the racial disparities were removed with minimal impact of predictive accuracy. While this approach applies to predictive policing models, it is also general enough that the concept can also be applied to other models. This work was continued in a 2021 paper currently under review detailing quality concerns regarding the data collected about individuals in the criminal justice system. It was found that pretrial risk assessment instrument datasets contain numerous measurement biases and errors inherent to criminal justice pretrial evidence. The impact is that algorithms using criminal justice system data have limited real-world impact.[7]
She has applied a number of statistical and probability models to the criminal justice system. This includes causal inference methods to explore the causal impact of setting bail on the likelihood of pleading or being found guilty. Another method is agent-based modeling methods, originally in epidemiology, to study a disease-like spread of incarceration through a social influence network.[8]
Lum is an author of the dga package, an open source software for population estimation. The package focuses specifically on Bayesian model averaging for capture-recapture. This is a CRAN package built for the R (programming language) in 2015 as part of Lum’s work at Human Rights Data Analysis Group.[9]
Honors
Lum was profiled in the American Statistical Association magazine article, A Statistician's Life, Celebrating Women in Statistics.[10]
References
- ↑ "Kristian Lum". University of Washington. University of Washington Department of Statistics. Retrieved 21 November 2021.
- ↑ "How Kristian Lum M.S. '08, Ph.D.' 10 makes the data matter". Duke Alumni. Duke Magazine. Retrieved 21 November 2021.
- ↑ "Kristian Lum". University of Washington. University of Washington Department of Statistics. Retrieved 21 November 2021.
- ↑ "A Statistician's Life, Celebrating Women in Statistics". Amstat News. American Statistical Association. Retrieved 21 November 2021.
- ↑ "Welcome Kristian Lum!". University of Pennsylvania. University of Pennsylvania Department of Computer and Information Science. Retrieved 21 November 2021.
- ↑ Lum, Kristian; Johndrow, James (25 Oct 2016). "A statistical framework for fair predictive algorithms". Retrieved 21 November 2021.
{{cite journal}}: Cite journal requires|journal=(help) - ↑ Bao, Michelle; Zhou, Angela (10 Jun 2021). "It's COMPASlicated: The Messy Relationship between RAI Datasets and Algorithmic Fairness Benchmarks". Retrieved 21 November 2021.
{{cite journal}}: Cite journal requires|journal=(help) - ↑ "Kristian Lum, PhD Lead Statistician". HRDAG. Human Rights Data Analysis Group. Retrieved 21 November 2021.
- ↑ "dga: Capture-Recapture Estimation using Bayesian Model Averaging". The Comprehensive R Archive Network. Human Rights Data Analysis Group. Retrieved 21 November 2021.
- ↑ "A Statistician's Life, Celebrating Women in Statistics". Amstat News. American Statistical Association. Retrieved 21 November 2021.
External links
This article "Kristian Lum" is from Wikipedia. The list of its authors can be seen in its historical. Articles taken from Draft Namespace on Wikipedia could be accessed on Wikipedia's Draft Namespace.