Source: https://drive.google.com/file/d/1gnDD548o2F8g_O__RfkRAQzWnkguGpe_/view
What to do about bias
- Anticipate and plan for potential biases before model generation. Check for bias after.
- Use machine learning to improve lives rather than for punitive purposes.
- Revisit your models. Update your algorithms.
- You are responsible for the algorithms you put out into the world, unintended consequences and all.
Deep Learning
How is deep learning different from classical ML?
Machine Learning involves feature extraction/selection (manually picking parts of your data that you want to run an ML algorithm on)
Machine learning involves trying different algorithms to find out which works best (dependent on the data you have/are using)
Most algorithms treat the data linearly
Deep learning does feature selection for you, just pass in all the data (great for high-dimensional data)
1 type of neural network tends to work for all sorts of data; small tweaks (rather than changing the whole algorithm)
Treats data points nonlinearly; algorithm will learn nonlinear function (i.e. curve) to classify the data
Where do we encounter DL?
- Neural nets used to solve many problems today
- Autonomous driving
- fintech/stock market predictions
- Google search results, google ads, Google Assistant, Siri, Alexa, etc.
- FB NewsFeed, Recommendations on YouTube, Amazon, Netflix, Spotify, TikTok, etc.
- Healthcare, Cancer Diagnostics, Genetics
If deep learning is so great, why don't we only use neural networks and deep learning for everything?