vid_L2: Deep Learning Tools and Frameworks Flashcards

Flashcards about deep learning tools, frameworks, and technical setups.

Keras

A high-level framework used for defining neural networks and setting up their architecture.

TensorFlow

A framework used for extra computation, often alongside Keras; alternatives include PyTorch and new objects.

NumPy

A library for basic number manipulation in Python, useful for multi-dimensional array operations.

What is the backend language commonly used for the actual computation in deep learning frameworks?

C++ and CUDA

What should you do to ensure compatibility and prevent version conflicts when installing necessary libraries?

Use a separate virtual environment

GPUs have a large number of what relative to CPUs?

Cores

What free cloud computing resource can you use with a Google account to run deep learning code?

CoLaboratory

What type of file contains both text and code and is used for the content and exercises in the course?

Jupyter Notebook

TensorFlow

What supports GPU which speeds up computation (as comapred to NumPy)?

Tensor

A multidimensional array in TensorFlow is called a what?

In TensorFlow, what are the two types of objects to consider when defining an array?

Constants and Variables

Which operation multiplies elements in tensors whereas another performs full matrix multiplication?

Element-wise multiplication (using *) vs. Matrix multiplication (using @ or tf.matmul)

What do the reduce functions (like reduce_sum) do related to rows?

It can sum down the rows, and get rid of the rows.

What is the shape of a single number (scalar) in TensorFlow?

An empty tuple ()

What are the typical dimensions for images processed in deep learning?

Height, Width, Channels

What is the fourth dimension added in deep learning image processing?

Batch dimension

Why is mini-batch gradient descent used?

Because it is fast and enables one to fit all the data into the GPU memory.

What functions add a dimension of one to the front end/gets rid of the dimesions?

Expand_dims and squeeze

Gradients

The directions of change are computed at this points.