June Edition: Bias in the machine
Fairness, bias, and interpretability in artificial intelligence and machine learning models
As machine learning applications become more widespread, there has been great interest in implementing algorithms to transform usual business processes to realize efficiencies. From loan approvals to judicial sentencing, consumers and citizens face the reality of a black-box model as the final arbiter behind some of the most important decisions and events in our lives. It has become more critical than ever to understand the question of bias and fairness in the models we create and ensure that they don’t create unintended and/or discriminatory outcomes.
Machine learning bias is not well-addressed, or even well-understood, in data science. But researchers and other practitioners have taken steps to highlight the importance of mitigating sources of bias and finding solutions to prevent harms resulting from them. The following articles are a selection of our best stories on bias, fairness, and interpretability. We hope that they enrich your understanding and act as a resource of best practices when you encounter biases in your own models.
Elliot Gunn, Editor at Towards Data Science.
What is AI bias?
By Cassie Kozyrkov — 4 min read
The AI bias trouble starts — but doesn’t end — with definition. “Bias” is an overloaded term which means remarkably different things in different contexts.
Reducing AI Bias with Synthetic Data
By Alexander Watson — 5 min read
Generate artificial records to balance biased datasets and improve overall model accuracy
Explaining Measures of Fairness
By Scott Lundberg — 11 min read
Avoid the black-box use of fairness metrics in machine learning by applying modern explainable AI methods to measures of fairness.
Interpretable Machine Learning
By Parul Pandey — 10 min read
Extracting human understandable insights from any Machine Learning model
Guide to Interpretable Machine Learning
By Matthew Stewart, PhD Researcher — 28 min read
Techniques to dispel the black box myth of deep learning.
Interpretable AI or How I Learned to Stop Worrying and Trust AI
By Ajay Thampi — 13 min read
Techniques to build Robust, Unbiased AI Applications
Algorithmic Solutions to Algorithmic Bias: A Technical Guide
By Joyce Xu — 16 min read
I want to talk about technical approaches to mitigating algorithmic bias.
New paper: The Incentives that Shape Behaviour
By Ryan Carey and Eric Langlois
How causal models can describe an agent’s incentives.
New videos
- Eduardo Coronado — Don’t be Afraid of Nonparametric Topic Models
- Logan Yang — Why Sigmoid: A Probabilistic Perspective
- Sheldon Fernandez and Michael St. Jules — Explaining With Impact
- Nick Frosst — Certifiable Robustness to adversarial Attacks; What is the Point?
New podcasts
- Bahador Khalegi — Explainable AI and AI interpretability
- Roland Memisevic — Machines that can see and hear
- Denise Gosnell and Matthias Broecheler — You should really learn about graph databases. Here’s why.
- Rubén Harris — Learning and looking for jobs in quarantine
We also thank all the great new writers who joined us recently Jean-Matthieu Schertzer, Ravi Charan, Artur Jurgas, Ayush Thakur, Brandon Lockhart, Michael Datz, Matt Maciejewski, Shunit Haviv Hakimi, Günter Röhrich, Meredith Wan, Lee Medoff, Frank Schilder, Kiana Go, Sebastian Schaal, Gianluca Gindro, Axel Thevenot, Matt Britton, Florian Wetschoreck, Thomas Hikaru Clark, Christopher Luc, Jin Pu, and many others. We invite you to take a look at their profiles and check out their work.
