Machine Learning vs. Deep Learning [What's the difference?]

in this video we're going to define some buzzwords  you may have heard floating around the ai space   deep learning machine learning and how these fit  within the ai world so here's what you need to   know in a single sentence machine learning is  a type of artificial intelligence deep learning   is an especially complex part of machine learning  what is machine learning machine learning is the   general term for when computers learn from data  it describes the intersect of computer science   and statistics in this field algorithms perform a  specific task without being explicitly programmed   instead they recognize the patterns in data and  make predictions once new data arrives in general   the learning process of these algorithms can  be either unsupervised or supervised depending   on the data being used to feed the algorithms  a traditional machine learning algorithm can   be something as simple as linear regression for  instance imagine you want to predict your income   given your years of higher education in a first  step you have to define a function for example   income equals y plus x times years of experience  then give your algorithm a set of training data   this could be a simple table with data on some  people's years of higher education and their   associated income next let your algorithm draw the  line for example through an ordinary least squares   ols regression now you can give the algorithm  some test data for example your personal years   of higher education and let it predict your  income while this sounds simple it does count   as machine learning and yes the driving force  behind machine learning is ordinary statistics   the algorithm learnt to make a prediction without  being explicitly programmed only based on patterns   and inference to recap machine learning is at the  intersection of computer science and statistics   this is the means through which computers receive  the ability to learn without being explicitly   programmed there are two broad categories  of machine learning problems supervised and   unsupervised learning a machine learning algorithm  can be something as simple as an ols regression   now let's examine how the term deep learning  relates to all of this what is deep learning   deep learning algorithms can be regarded both as a  sophisticated and mathematically complex evolution   of machine learning algorithms this field has been  getting lots of attention lately in the mainstream   because there's been an explosion in deep learning  technology deep learning describes algorithms that   analyze data with a logic structure similar to  how a human would draw conclusions this can happen   both through supervised and unsupervised learning  to achieve this deep learning applications use a   layered structure of algorithms called an  artificial neural network a-n-n the design   of such an ann is inspired by the biological  neural network of the human brain leading to   a process of learning that's far more capable  than that of standard machine learning models   this is what a simple artificial neural network  looks like the leftmost layer is called the input   layer the rightmost layer is the output layer the  middle layers are called hidden layers because   their values aren't observable in the training set  in simple terms hidden layers are the calculated   values used by the network to do its magic the  more hidden layers a network has between the input   and output layers the deeper it is in general any  ANN with two or more hidden layers is referred to   as a deep neural network applications of deep  learning deep learning is used in many fields   in automated driving for instance deep learning  is used to detect objects such as stop signs or   pedestrians imagine the company tesla using a deep  learning algorithms for its cars to recognize stop   signs in the first step the a n would identify  the relevant properties of the stop sign   also called features features may be specific  structures in the inputted image such as points   edges or objects the military uses deep learning  to identify objects from satellites for example   to discover safe or unsafe zones for its troops  and the consumer electronics industry is full   of deep learning too home assistant devices such  as amazon Alexa for example rely on deep learning   algorithms to respond to your voice and know  your preferences while a software engineer would   have to select the relevant features in a more  traditional machine learning algorithm the ann   is capable of automatic feature engineering the  first hidden layer might learn how to detect   edges the next how to differentiate colours and  the last learn how to detect more complex shapes   catered specifically to the shape of the object  we're trying to recognize when fed with training   data the deep learning algorithms would eventually  learn from their own errors whether the prediction   was good or whether it needs to adjust what the  job of a neuron in an ann actually looks like   overall through automatic feature engineering and  its self-learning capabilities the deep learning   algorithms need only little human intervention  while this shows the huge potential of deep   learning there are two main reasons why it  has only recently attained so much usability   data availability and computing power firstly deep  learning requires incredibly vast amounts of data   we'll get to exceptions to that rule in a minute  tesla's autonomous driving software for instance   needs millions of images and video hours to  function properly secondly deep learning needs   substantial computing power however with the  emergence of cloud computing infrastructure   and high performing gpus graphic processing  units used for faster calculations the time   for training a deep learning network could be  reduced from weeks to mere hours but probably   one of the most important advances in the field  of deep learning is the emergence of transfer   learning that is the use of pre-trained models the  reason transfer learning can be regarded as a cure   for the needs of large data training sets that  were necessary for a ANN to produce meaningful   results these enormous data needs used to be the  reason why ANN algorithms weren't considered to be   the optimal solutions to all problems in the past  however for many applications this need for data   can now be satisfied by using pre-trained models  the main differences between machine learning and   deep learning this is a common question and if you  have watched this far you probably know it's not   the right one to ask deep learning algorithms are  machine learning algorithms therefore it might be   better to think about what makes deep learning  special within the field of machine learning   the answer the ANN algorithm structure the lower  need for human intervention and the large data   requirements first and foremost while traditional  machine learning algorithms have a rather simple   structure such as linear regression or a decision  tree deep learning is based on an artificial   neural network this multi-layered ANN is like  a human brain complex and intertwined secondly   deep learning algorithms require much less human  intervention remember the tesla example if the   stop sign image recognition was a more traditional  machine learning algorithm a software engineer   would manually choose features and a classifier  to sort images check the output is as required and   adjust the algorithm if this is not the case as a  deep learning algorithm however the features are   extracted automatically and the algorithm learns  from its own errors the deep learning algorithm   doesn't need a software engineer to identify  features but is capable of automatic feature   engineering through its neural network thirdly  deep learning requires much more data than a   traditional machine learning algorithm to function  properly machine learning works with a thousand   data points deep learning often times only with  millions due to the complex multi-layer structure   a deep learning system needs a large data set  to eliminate fluctuations and make high quality   interpretations don't forget to subscribe if  you're looking for more introductory level guides   to machine learning and get in touch with us to  learn more about creating your own ai without code

Channel: Levity

Views: 19,648

Keywords: automation, machine learning, artificial intelligence, levity, productivity, workflow automation, saas, deep learning, machine learning vs deep learning, what is deep learning, what is machine learning, ai, neural networks, structured vs unstructured data, requirements for deep learning

Id: cDTp-qXXZU0

Length: 7min 40sec (460 seconds)

Published: Tue Sep 14 2021