Merge pull request #584 from esciencecenter-digital-skills/deep-learn…

…ing-terms

Deep learning terms

glossary.yml

@@ -10120,4 +10120,85 @@
     term: "non-parametric (statistics)"
     def: >
       A branch of statistical tests which do not assume a known distribution of the population which the samples were taken from (Kruskal-Wallis and Dunn test are examples of non-parametric tests).
-     
+- slug: artificial_intelligence
+  ref:
+    - nlp
+    - machine_learning
+  en:
+    term: "artificial intelligence (AI)"
+    def: >
+      Intelligence demonstrated by machines, as opposed to humans or other animals. AI can be 
+      exhibited through perceiving, synthesizing and inference of information. Example tasks include
+      [natural language processing](#nlp), computer vision, and [machine learning]
+      (#machine_learning).
+
+- slug: cnn
+  ref:
+    - deep_learning
+    - backpropagation
+    - perceptron
+    - neural_network
+    - machine_learning
+  term: "convolutional neural network (cnn)"
+  def: >
+    A class of artificial neural network that is primarily used to analyze images. A CNN has layers
+    that perform convolutions, where a filter is shifted over the data, instead of the general 
+    matrix multiplications that we see in fully connected neural network layers.
+
+- slug: rnn
+  ref:
+    - deep_learning
+    - backpropagation
+    - perceptron
+    - neural_network
+    - machine_learning
+  term: recurrent neural network
+  def: >
+    A class of [artificial neural networks](#neural_network) where connections between nodes can 
+    create a cycle. This allows the network to exhibit behavior that is dynamic over time. This 
+    type of network is applicable to tasks like speech and handwriting recognition.
+
+- slug: epoch_dl
+  ref:
+    - deep_learning
+    - backpropagation
+    - perceptron
+    - neural_network
+    - machine_learning
+  term: epoch (deep learning)
+  def: >
+    In [deep learning](#deep_learning), an epoch is one cycle in the deep learning process where all
+    the training data has been fed to the algorithm once. Training a deep neural networks usually
+    consists of multiple epochs.
+
+- slug: learning_rate
+  ref:
+    - deep_learning
+    - backpropagation
+    - perceptron
+    - neural_network
+    - machine_learning
+  def: >
+    In [artificial neural networks](#neural_network), the learning rate is a hyper-parameter that
+    determines the pace at which the network adjusts the weights to move down the loss gradient. 
+    A large learning rate can speed up training, but the network might overshoot and miss the 
+    minimum. A small learning rate will overshoot less, but will be slower. It can also get more 
+    easily stuck in local minima.
+
+- slug: class_imbalance
+  ref:
+    - machine_learning
+  def: >
+    Class imbalance refers to the problem in [machine learning](machine_learning) where there is an 
+    unequal distribution of classes in the dataset.
+
+- slug: hidden_layer
+  ref:
+    - neural_network
+    - machine_learning
+    - deep_learning
+    - perceptron
+  def: >
+    A hidden layer in a [neural network](#neural_network) refers to the layers of neurons that are
+    not directly connected to input or output. The layers are "hidden" because you do not directly
+    observe their input and output values.