Ch1 & Ch2 End-of-Chapter Questions Flashcards
What was the the name of the first device that was based on the principle of the artificial neuron?
Mark I Perceptron (Rosenblatt)
What were the two theoretical misunderstandings that held back the field of neural networks?
- Marvin Minsky and Syemour Papert wrote book called Perceptrons, where they showed that a single layer of these devices was unable to learn even simple mathematical functions. They also showed that adding more layers solved the problem, but only the first insight became widely known.
- In the 80s, people started building models with two layers, which theoretically was enough to allow any mathematical function to be approximated. However, these networks were too big and too slow to be useful in practice.
What is a GPU?
Graphics Processing Unit, a.k.a. graphics card. A kind of computer processor that can handle thousands of single tasks at the same time. GPUs can run neural networks hundreds of times faster than regular CPUs.
Why is it hard to use a traditional computer program to recognize images in a photo?
Because the steps we take to recognize an image are ambiguous since they happen in our brain. Without knowing the steps, we can’t write a program.
What is the name of the theorem that shows that a neural network can solve any mathematical problem to any level of accuracy?
The universal approximation theorem
What is a metric? How does it differ from loss?
A metric measures the quality of a model’s predictions using the validation set, and it’s chosen to be something easy for people to understand. Loss is a measure of performance that the training system can use to update weights automatically, and it’s chosen to be something easy for SGD to use.
What is the “head” of a model?
The head of a model is the part that is newly added to the pretrained model to be specific to the new dataset.
What is an architecture?
It’s the functional form of a model. Model and architecture have been used interchangeably.
What is segmentation?
Creating a model that can recognize the content of every individual pixel in an image
Where do neural networks used for text models have a major deficiency?
Deep learning is good at imitating human text, but it is not good at generating correct responses. So, it sounds like a human, but it’s saying the wrong things.
What are possible negative societal implications of text generation models?
Neural networks could be used to generate text that sounds good, but that could spread disinformation, create unrest, and encourage conflict.
In situations where a model might make mistakes, and those mistakes could be harmful, what is a good alternative to automating a process?
Start out with a manual process and the model running at the same time. A person would check the results of the model and fully control any decisions until we know the model is doing what we want.
What’s a key downside of directly using a deep learning model for recommendation systems?
The model will tell you which products the customer likes (based on past purchases), but it is not good at giving recommendation of new products (ones that are not the same as what she has bought already) that would be useful for the customer.
What does the splitter parameter to DataBlock do?
It defines how to split the data into a training and validation sets.
What 4 things do we need to tell fastai to create DataLoaders?
- What kinds of data we have
- How to get the list of items
- How to label these items
- How to create the validation set
