#论文笔记 #MLfairness On how to measure discrimination: "a methodology for analytically quantifying explainable
and illegal discrimination: conditional discrimination-aware classification." By Kamiran, F., Žliobaitė, I., & Calders, T. (2013). Quantifying explainable discrimination and removing illegal discrimination in automated decision making. Knowledge and information systems, 35, 613-644.

3.3 Mesuaring discrimination in classification. Measuring it is to determine, which part of this difference is explainable by program, and which part is due to
illegal discrimination, that is, the difference in the probabilities as a sum of the explainable and
illegal discrimination

D_all = D_expl + D_illegal.

See Table 3: the decision
making is biased in favor of males, P(+|m, ei) > P(+|f, ei), where ei is a program.

See Formula (4), the formal definition of the explainable discrimination is the difference between acceptance
of males and females.

#论文笔记 #MLfairness “Using sensitive personal data may be necessary for avoiding discrimination in data-driven decision models” by Zliobaite & Custers, 2016.

TL;DR: What results in unfair regression models: Omitted variable bias.

It happens when a regression model is fitted without an important causal variable.
When the bias happens, the unfair regression model would allocate more weights to the remaining, legitimate variable, thus incorrectly and unfairly punish or rewards people more than deserved by the legitimate variable. See the example in Chap 3.1.

🔥⏲️ Fudge Sunday "Who Said The AI ML Was Fair?" A look at at fairness in A.I. and M.L. and F.A.I.R. data.

#fairness #fairnessmatters #aifairness #mlfairness #aiops #mlops #critique #fairdata #artificialintelligence #ai #aiforgood #aiforall #aiandbusiness #gerrymandering #transparency #reproducibility #ethics #ethicsandai #ethicsinai #ethicsandcompliance #empathy #stem #esteem #platformengineering #watsonx #devx #developerexperience #newsletter #newsletters

https://fudge.org/archive/who-said-the-ai-ml-was-fair/

I am happy to say that our paper was accepted at #neurips2022 and we released not 1, not 2, but 6 companion datasets!🎉

The paper, "Turning the Tables: Biased, Imbalanced, Dynamic Tabular Datasets for ML Evaluation" presents a suite of datasets to evaluate ML fairness.😎

We are releasing these datasets because the normally currently used datasets to test ML fairness are small (less than 50k rows), are old (e.g., the frequently used "UCI Adult" is from 1994), or have documentation issues or data leakage.

On the contrary, our datasets are realistic, large (1 million rows), contain 5 different types of biases, have real-world properties (imbalanced, time shifts), and are privacy-preserving (via GANs, Laplacian noise, and filters and transformations).

Paper, datasets, datasheets, and code are available at:
https://lnkd.in/ddDm4Y64

Thank you co-authors Sérgio Jesus, José Maria Pombal, Duarte Alves, André Cruz, Pedro Saleiro, Rita P. Ribeiro, and Joao Gama, in another collaboration between Feedzai and Universidade do Porto.

Thank you also to colleagues and helping friends João Veiga, Joao Bravo, Catarina Belém, and Bruno Cabral and to sponsoring agency ANI - Agência Nacional de Inovação and Carnegie Mellon Portugal Program.

#feedzai
#frauddetection
#ResponsibleAI
#MLfairness

I constantly say #rightwingcontentisspam but realize I should explain:

I'm not trying to throw shade, rather, objectively, I am shining a light on the ecosystem that has captured right leaning audience attention.

Pretend this is a comparison matrix for [Spam , Right wing content]:

Deceptive content: [✅|,✅]
Manipulates emotions to get you to click: [✅,✅]
Shameless ads and monetization: [✅,✅]
Low quality info: [✅,✅]
Filtered by most platforms [✅,❌ ]

#mlfairness #spam #propaganda #fakenews