In the Machine Learning, Say What? training course, you'll learn:
High-Level Introduction
- What is Machine Learning?
- What are the three major types of Machine Learning?
- Supervised learning
- Unsupervised learning
- Reinforcement learning
- How are companies using these technologies?
- How is this software being used?
Strengths and Weaknesses of Machine Learning
- How to evaluate if machine learning is a good fit for a particular problem
- What are the major sources of risk in machine learning projects?
- "Black box" versus interpretable models
- Data quality and bias
What Goes into Making a Machine Learning Product?
- Describe the parts of the Machine Learning life-cycle
- Define the objective
- Acquire & explore the data
- Execute modeling
- Interpret and communicate the results
- Implement, document, deploy, monitor and maintain the system
What Went Wrong? (Case Study)
- Amazon attempted to use machine learning as a tool for resume screening, but ultimately shuttered the program before deployment after determining it was biased against women
- Discuss the following:
- What are some few hypotheses about why their system ended up with this bias?
- What measurements could be used to quantify the model's bias?
- What corrections could be made to correct this bias in the model?
- What are the possible classifications of this tasks as a good fit for machine learning?
How to Simulate the Product Life-Cycle?
- Discuss what the machine learning product life-cycle
- What data is needed and how will it be retrieved?
- What metrics should be tracked to measure the success of this system?
- How to define "success" and "failure" with respect to these metrics
- What are the risks in a system like this?
- What are possible failsafes to design to mitigate these risks?
- What might the infrastructural requirements be?
- What systems generate the input on the live system?
- How will it be monitored? How will it alert if it's no longer working?
- What controls does the algorithm need to be successful?
- (e.g. it probably needs to block a transaction... what else?)
- What needs to be done if the system starts failing?
- Is this a good fit for machine learning?
- Is machine learning the best way to do this?
- Is there something simpler that works just as well?
Group Discussion and Recap
