Neural Network
parts
parts
This is the first section of a neural network that receives raw data like images, text, or numbers from the outside world.
what is input layer
What is a CNN
This traditional recommendation strategy relies on analyzing past interactions between users and items without needing detailed profiles of either.
What is collaborative filtering?
This error occurs when a model learns the training data too perfectly, including its noise, causing it to perform poorly on new, unseen data.
What is overfitting?
True or False: Adding more hidden layers to a neural network always guarantees a higher accuracy on the validation set.
What is False?
This basic unit or cell of a neural network receives inputs processes them and passes an output forward.
What is a neuron
This computer vision task goes a step beyond image classification by drawing a bounding box around specific items in a photo.
What is object detection?
This algorithmic technique breaks a massive user-item rating grid down into two smaller, lower-dimensional dense matrices to uncover hidden preferences.
What is matrix factorization?
This optimization algorithm calculates gradients and iteratively tweaks network weights to minimize the loss function.
What is gradient descent?
True or False: A neural network with purely linear activation functions across all layers can only solve linearly separable problems, no matter how deep it is.
What is True?
These layers sit between the input and output layers, doing all the heavy lifting and math to find patterns in data.
What are the hidden layers?
This common layer in a CNN downsamples an image to reduce its spatial size and computational load, usually by taking the maximum value in a window.
What is a max pooling layer?
Question: This problem occurs when a recommendation system struggles to suggest items to a brand-new user because there is no historical data available
What is the cold start problem?
This critical value controls how large of a step a model takes toward minimizing the loss function during each training iteration.
What is the learning rate?
True or False: Content-based recommendation filtering primarily relies on finding similarities between different users' behavior patterns.
What is False?
What is a weight?
This advanced task involves classifying every single pixel in an image to a specific category, tracing exact outlines rather than drawing boxes.
What is semantic segmentation
Question: In modern deep learning recommendation models, this technique converts categorical data into low-dimensional continuous vectors.
What are embeddings
This term represents one complete pass of the entire training dataset forward and backward through the neural network.
What is an epoch?
True or False: Dropout is a regularization technique that randomly deactivates a fraction of neurons during testing to speed up inference.
What is False?
This specific function is applied to a neurons output to determine whether it should fire or activate introducing non linearity to the network
What is an activation function
This architectural feature uses shortcuts to skip over one or more layers, helping train incredibly deep networks by preventing vanishing gradients.
Answer: What are residual connections
Often used in industrial pipelines, this type of neural network ranks candidate items by explicitly combining memorization of historical data with generalization of new features.
What is a Wide & Deep network?
This mathematical core of network training uses the calculus chain rule to calculate the gradient of the error from the output layer back to the input layer.
What is backpropagation?
True or False: If a neural network's loss stops decreasing, it means the model has definitely found the global minimum of the loss function.
What is False?