Fast AI Flashcards
What did Frank Rosenblatt build?
perceptron
What was Marvin Minsky’s innovation?
Using multiple layers could solve for limitations
What is the general method for updating weights in a neural net?
Stochastic gradient descent
What is ImageNet?
The ImageNet project is a large visual database designed for use in visual object recognition software research. More than 14 million images have been hand-annotated by the project to indicate what objects are pictured and in at least one million of the images, bounding boxes are also provided
What’s a tensor?
collections of numbers (vector, matrix, and higher dimensions)
Explain gradient descent
At each step we try and work out the gradient and based on updating the parameters to minimise the loss
What function can be used to describe any input-output? In neural nets
Combinations of Rectified Linear functions
Describe deep learning mathematically…
Combinations of ReLu
Gradient descent
Inputs and outputs
When i adjust parameters (in line with gradient descent), the amount i adjust by is called the what?
The learning rate
the foundational mathematical operation in deep learning is …
Matrix multiplication
Describe solving the Titanic (kaggle) problem using regression in Excel
- Normalize all the variables
- Random coefficients
- Loss for each row, and then the mean of the sum
- “solve” function to minimimise the Total Loss by changing the coefficients
What is the method known where you split data by one variable to determine the outcome?
1R
How is the 1R made more sophisticated?
Turning it into a Decision Tree
What is the best package to use in Python for non-deep (aka classical) machine learning?
Scikit Learn
What’s the name of the measure for how good a split is?
Gini
(how likely if you pick from a sample and then pick another that you’ll get the same thing)
What’s the main way of improving decision trees?
Random Forest
What’s the idea behind “bagging” (used in Random Forest)?
Take average predicition of uncorrelated models - that should reduce the error to zero
What’s a good first model to use when given a data set?
Random Forest - get to see which columns are the most important
What is an Out of Bag Error?
Predictions made on the rows not used in the particular decision tree (subset)
What is gradient boosting?
Boosting is one kind of ensemble Learning method which trains the model sequentially and each new model tries to correct the previous model
what is TTA (machine learning)?
Test Time Augmentation
(passing along to the model multiple augmented images and getting back the average result)
What does Jeremy Howard recommend for GPU?
Nvidia
RTX cards are for consumers (NOT for data centers) e.g. 3080 is about $1,000
Recommendation systems are known as
Collaborative Filtering
What’s a convolution?
sliding matrix (/kernel) of values from o.g. image dot-product-ed with a filter to pick out features such as vertical edges, horizontal edges etc
How is overfitting avoided in CNNs?
By introducing a drop out layer - some of the pixels are zeroed (randomly)
