Convolutional Neural Networks (ConvNets) and Recurrent Neural Networks (RNNs)
The Future of Deep Learning
Training Multilayer Architectures
Irrelevant Randomness
100
A new area of machine learning.
Deep learning.
100
A recent major success of ConvNets.
Facial recognition.
100
How is human and animal learning unsupervised?
We learn by observation, not by being fed data.
100
A common method for training a neural network.
Backpropagation.
100
Javier's last name.
Weddington.
200
The science of getting computers to act without being explicitly programmed.
Machine learning
200
Since when have ConvNets been successfully applied?
Since the early 2000s.
200
Another area in which deep learning is expected to have a large impact over the next few years.
Natural language understanding.
200
First major application in the pre-training approach.
Speech recognition.
200
A branch of physics which deals with the nature and behavior of matter and energy on an atomic and subatomic scale.
Quantum mechanics/physics.
300
The most common form of machine learning where both input and desired data are provided.
Supervised learning.
300
Networks designed to process data that come in the form of multiple arrays.
Convolutional neural networks.
300
Procedures that could create layers of feature detector without requiring labeled data.
Unsupervised learning procedures.
300
When did researchers find out that they could use Stochastic Gradient Descent (SGD) to train multilayer architectures?
Mid 1980s.
300
The molecular shape of a double-stranded DNA molecule.
Double helix.
400
Requires very little engineering by hand.
Deep learning.
400
Networks that are often used for tasks involving sequential inputs like speech and language.
Recurrent neural networks.
400
What systems have been used for speech and handwriting recognition for a long time?
Deep learning and simple reasoning.
400
What allowed researchers to train networks 10-20 times faster?
Graphics Processing Units (GPUs).
400
Put the taxonomic ranks in order from broadest to most specific.
Kingdom, Phylum, Class, Order, Family, Genus, Species.
*50 point bonus if Domain is added*
500
A set of methods that allows a machine to automatically discover the representations needed for detection or classification.
Representation learning.
500
What happens to the backpropagated gradients in RNNs after many time steps?
They either explode or vanish.
500
What systems are expected to make major progress in Artificial Intelligence?
Systems that combine deep learning with complex reasoning.
500
What was introduced by Canadian researchers in 2006?
Unsupervised learning.
500
How can you throw a ball as hard as you can and have it come back to you even if it doesn't hit anything, there's nothing attached to it, and no one else catches it or throws it back?