AI a Brief Intro
Artificial Intelligence :
In simple terms the ability of a Computer to think and Learn like a human brain
Machine Learning : Subset of AI
It is a technique of Parsing the data and applying the conditions to perform any decision
Example : If you search for a Product In Amazon Shopping or Some other web results by keeping the keyword your's advertisement will be displayed
Deep Learning : Subset of Machine Learning
By keeping the data applying the conditions in all possibilities
Diff ML and DL
Example : Here - the scenario is If Fire Door needs to be opened
ML- Fire - The Door will be opened
DL - The cupboard is burning , the building is burning open door etc ..
Here the Fire Keyword is not mentioned but the door need to be opened using the analyzement
Language and Tools:
In simple terms the ability of a Computer to think and Learn like a human brain
Machine Learning : Subset of AI
It is a technique of Parsing the data and applying the conditions to perform any decision
Example : If you search for a Product In Amazon Shopping or Some other web results by keeping the keyword your's advertisement will be displayed
Deep Learning : Subset of Machine Learning
By keeping the data applying the conditions in all possibilities
Diff ML and DL
Example : Here - the scenario is If Fire Door needs to be opened
ML- Fire - The Door will be opened
DL - The cupboard is burning , the building is burning open door etc ..
Here the Fire Keyword is not mentioned but the door need to be opened using the analyzement
Language and Tools:
The below
would be languages and tools which we would using,
Basic:
a.
Python
b.
Jupyter notebook
c.
Numpy
d.
Pandas
e.
Matplotlib
f.
Seaborn
Machine
learning library:
Deep
learning library:
Brief
overview:
a.
TensorFlow
b.
Keras wrapper with TensorFlow
backend
Basically
90%-95% of the course would be covered using the Sklearn Machine learning and
Pytorch deep learning framework.
Prerequisite:
- Python
(Duh)
- Working
knowledge of iPython notebook, Numpy, Pandas, Matplotlib and
Seaborn.
I will
recommend buying and reading the book “Python for Data Analysis: Data Wrangling
with Pandas, NumPy, and IPython, 2nd Edition, Wes McKinney”.
- I will
also recommend a Macbook or a DevDesktop running Linux.
I can try
supporting Windows and Machine learning libraries should work but it would make
logistics easier if everyone is in Linux or Mac. Deep learning library
(Pytorch) will _not_ run in Windows .
If you are
serious about Deep Learning you will need either a AWS GPU instance (
p2.xlarge) or home rig with GPU. Please refer the below link for more details,
Why and
what GPU for deep learning?
Current
hardware status of Deep learning. (read the comments also)
For this class, a Nvidia GTX 960/GTX1060 desktop class card (or) in AWS a
Nvidia Tesla K80 should suffice. Ideally you should be able to share a single
AWS instance across members or borrow some of your friends gaming rig ☺.
One
is TensorFlow from Google and the other is PyTorch from FAIR (Facebook AI
Research). Tensorflow works in creating a static graph (a DAG) which is then
pushed to the GPU after which it becomes a black box which means that we cannot
use Source level debugging and put breakpoints in-between to figure out what’s
going on in any easy manner. This is not a problem for someone who is well
versed in the theoretical aspects of Deep learning. In our class, we will go
with a code first approach which will be difficult to do with TensorFlow.
Second
reason is that, Tensorflow has “Tensor flow” routines which are equivalent of
Python routines though not exactly same. This means a bit of learning is needed
before we can start using the same.
Third, if
you take any Kaggle deep learning competitions, more than 80%-90% of the
winning entries would be based on Pytorch. Considering that Pytorch has not
even completed its first birthday yet, it’s is just remarkable.
Pytorch
avoids all these problems by building “Dynamic” graphs and the speeds are more
or less comparable. This means we put breakpoints and debug stuff with a
debugger if needed. Also most of basic python could be reused without
needing to refer to the documentation to find some equivalent Tensorflow
routines.
Don’t get the me wrong, Tensorflow is a brilliant framework with some cool
stuff like Tensor board and is more suitable for production. It’s just not
suited for our class.
- What
about libraries like Keras with Tensorflow backend:
Keras is very east for certain subset of work and very difficult to get it
working for a lot of other problems. The book in have given below should
help you do the same thing in Keras or Tensorflow if you do not like or
want Pytorch for some reason.
- Can I
read these in my free time and not attend the class: By
all means please do the same. All my material is taken from internet and
nothing exclusive expect some of the data which I use for examples which
is also available online. Having a class structure only enforces some sort
of discipline, nothing else.
Books:
Mandatory
reading:
Python for
Data Analysis: Data Wrangling with Pandas, NumPy, and IPython, 2nd
Edition, Wes McKinney
Good to
read:
Python
Machine Learning 2nd Edition, Sebastian Raschka
Deep
Learning with Python, Francois Chollet
To be
read after gaining experience: (but the book is free)
Deep
Learning book by Ian GoodFellow, Youshua Bengio and Aaron Courville
For math’s
Khan academy
Algebra
Linear Algebra
Advance deep learning
machine learning into deep learning
Deep learning
Pandas
Pytorch
Seaborn
Numpy
Matplotlib
ScikitLearn
Jupyter
nvidia cuda
GPU graphics Processing unit
Machine Learning
DeepLearning
Re enforcement learning
alexnet
universal approximation theorem
Markov chain
vector
matrix
tensor
n dimensional matrix
jupyter why we are here
DSP
Convolution
Correlation
Butterworth filter
Transfer function
chebysev filter
fourier transform
band pass filter
pass band
stop band
Neural network
task
Edge detection
Linear algebra
Factorization
graph theory Adjacency list
sparse matrix
eigen value ,vector
google page rank uses 1^
similar valley
non factorization
y= mx+c
z=m1x1+m2X2
orthohanal
orthonormal
single valley decomposition
Derivative
Calculus Probability and Statistics
Multivariable Calculus
Linear
Regression
Data
CostFunction --> how much correct or wrong in it
--> concave functions
Simple linear regression model
ML - you have features in that regression
DL - There is no features example Speech alexa
RSS - Residual some of square
Concave Convex function
andrenngin video
chain rule
in ML
Gradient
Slope
Descendent making closure to 0
Partial Derivatives
Louis hemalton
Algebra
Linear Algebra
Advance deep learning
machine learning into deep learning
Deep learning
Pandas
Pytorch
Seaborn
Numpy
Matplotlib
ScikitLearn
Jupyter
nvidia cuda
GPU graphics Processing unit
Machine Learning
DeepLearning
Re enforcement learning
alexnet
universal approximation theorem
Markov chain
vector
matrix
tensor
n dimensional matrix
jupyter why we are here
DSP
Convolution
Correlation
Butterworth filter
Transfer function
chebysev filter
fourier transform
band pass filter
pass band
stop band
Neural network
task
Edge detection
Linear algebra
Factorization
graph theory Adjacency list
sparse matrix
eigen value ,vector
google page rank uses 1^
similar valley
non factorization
y= mx+c
z=m1x1+m2X2
orthohanal
orthonormal
single valley decomposition
Derivative
Calculus Probability and Statistics
Multivariable Calculus
Linear
Regression
Data
CostFunction --> how much correct or wrong in it
--> concave functions
Simple linear regression model
ML - you have features in that regression
DL - There is no features example Speech alexa
RSS - Residual some of square
Concave Convex function
andrenngin video
chain rule
in ML
Gradient
Slope
Descendent making closure to 0
Partial Derivatives
Louis hemalton
gen.lib.rus.ec
ReplyDeleteto download books