AI 101
With the right data and the right model, AI can solve many problems. But choosing the problem, finding the data, and training the model is actually difficult part.
- All AI today are narrow forms of artificial intelligence
- Methods of implementing AI
- Expert systems (if/else statements)
- Tree search (state tree)
- Machine learning (learning from data)
When to learn from machine learning
- Classification: Given an input, output the class of the input
- Clustering: Given an input, output a groups of similarities in the data
- Regression: Given an input, output a prediction
Types of learning
- Supervised learning: Training data, compare to known answer, repeat
- Unsupervised learning: Unknown data, this method gathers information about the data
- Reinforcement learning: Aiming to reward a type of behavior. The correct answer is not provided, but the goal is known
7 Steps Of Training An Algorithm
- Define a problem: Is the picture a cat or a dog?
- This step is extremely important. AI cannot solve all your problems
- Find the data: A shit ton of labeled pictures of cats and dogs
- Clean data: Make sure the data is good, clean, and representative
- 80% of data science and machine learning
- Choose a model
- How do you want to represent the data
- In other words, select the features of the data
- Train the model
- Test the model
- Deploy the model
4 Questions You Should Ask Any AI You Encounter
- What is the goal?
- What is the training data?
- What model is being used?
- What is the accuracy when tested on new data (testing)?