Deep Learning
Machine
Learning
Artificial Neural Networks
Modeled after biological
neuronal networks
Stochastic Gradient Descent
Used to calculate error and
tune weights in a network
Supervised Learning
Providing labeled data at the
input layer as training
Unsupervised Learning
Allowing the machine to find
patterns in unlabeled data
Deep Learning
Using non-linear transforms
to correlate data through
complex layers and
abstractions
Just as general computing changed the way society operates, so
too will general learning methodologies revolutionize the way
software is written and the way information is processed.
The rapid evolution of machine learning
techniques is heralding in an age of digital
thinking such that the world has never
seen before.
With the ability to abstract information at
multiple layers and draw correlations
autonomously, self-learning software
algorithms are making pattern
recognition, image recognition,
translation, and many other high-order
functions a task no longer limited to the
domain of humans.
Deep learning is a powerful example of
one such series of techniques.
By consuming an input of vectors and
processing them across astronomical
numbers of interconnected units and
layers, a deep learning algorithm can
teach itself how to detect edges, motifs,
shapes, and entire objects inside of an
image.
Utilizing widely available, inexpensive,
powerful, parallelized computer chipsets
called Graphical Processing Units
(GPUs), the performance of processing
these once unconscionably massive sets of
calculations has exploded.
Though creating a deep learning machine
is no secret, the cutting edge of these
algorithms often require access to
extremely large archives of labeled
images, sounds, and text so that they are
able to be trained or train themselves
when making the connections necessary
to form the desired output.
After realizing the potential of modern
self-learning machines, one can surmise
that the future of artificial intelligence is
right around the corner and that the
creative, problem-solving faculties of the
human mind are one step closer to being
replicated algorithmically.