| Contents |
|---|
| Getting Started |
| Mathematics |
| Deep Learning |
Most of the greatest feats of AI today (Alpha Go, Google Search Engine and etc) are possible due to a subdomain in AI called Machine Learning. This specific domain focuses on machines teaching themselves how to do a certain task (as defined by the programmer) based on input data i.e. the machine learns from the data hence the name "Machine Learning". Due to the mathematical nature of ML this guide involves mathematical courses as well. It is strongly recommended that students complete them for a better understanding of the subject.
Machine Learning on Coursera by Andrew Ng (highly recommended)
Access: Audit / Financial aid (2-3 weeks approval time) / Payment
This course is considered the best introduction to machine learning by the entire global community for its simplicity and concept-based learning. Course notes contributed by Anas Ayubi and scanned by Umair Hanif
Advantages:
- Andrew Ng dives into the algorithms used in machine learning by keeping mathematical involvement to a minimum whilst delivering all the intuition behind the mathematics used
- Students are expected to code the machine learning algorithms themselves
- Andrew Ng delivers practical advice on how to approach and solve machine learning problems throughout the course
Disadvantages:
- Delivering mathematical intuition becomes difficult due to the increased mathematical complexity in certain algorithms
- Octave/MATLAB is used for programming assignments in the course (less recognised programming language within the machine learning community)
Access: Audit / Financial aid (2-3 weeks approval time) / Payment
This course can be considered as a less intensive and shorter substitute for Andrew Ng's machine learning course.
Advantages:
- Algorithms are explained but students are not required to code them themselves - focus is mainly on learning how to use libraries
- Python, a well used language in the data science community, is used throughout the assignments
Disadvantages:
- No practical advice on how to approach and solve machine learning problems is given throughout the course
Although Andrew Ng, Carlos and Emily intentionally attempt to cover up the mathematics in order to not overwhelm beginners, students must not shy away from attempting to understand these topics in order to deepen their understanding regarding the subject. As the method of teaching in mathematics within Pakistan requires great improvement, it is strongly recommended that the following courses are taken by all participants. One more thing to note - our focus within mathematics courses is mainly towards rigorous university courses rather than "not so rigorous" MOOC courses. This is due to the fact MOOCs tend to simplify mathematical content which leads to limited applicability of that content to real life situations. We at Pakistan.ai can't emphasis how important it is to pick the correct mathematical courses hence we dedicated an entire file to this topic. Read more on this here.
Access: Free
This course delivers an entire undergraduate experience of studying at Harvard and introduces a rigorous and intuition-based introduction of probability and how it is applied.
Advantages:
- Great insight is given into how probability exists in the real world and how it must be correctly applied to real life situations
- Exercises are provided
Disadvantages:
- Course is considered challenging by most students
- Difficult to follow the course if exercises are not attempted by students
Access: Free
For those who find Probability at Harvard by Joe Blitzstein challenging Khan Academy's courses serve as a less rigorous alternative with an increased focus on intuition.
Advantages:
- Highly conceptual videos with focus on intuition
- Exercises are provided
Disadvantages:
- Exercises are oversimplified and lack in real world applications - course does not deliver the skills required to apply concepts to real world applications
Access: Free
This course delivers an entire undergraduate experience of studying at MIT and introduces a rigorous and intuition-based introduction of linear algebra and how it is applied.
Advantages:
- Great insight is given into how linear algebra is relevant and how it must be correctly applied to real life situations
- Recitation videos are present - these videos walk you through a solution to a specific problem
- Exercises are provided
Disadvantages:
- Course is considered challenging by most students
- Difficult to follow the course if exercises are not attempted by students
Access: Free
For those who find Linear Algebra at MIT by Gilbert Strang challenging Khan Academy's courses serve as a less rigorous alternative with an increased focus on intuition.
Advantages:
- Highly conceptual videos with focus on intuition
- Exercises are provided
Disadvantages:
- Exercises are oversimplified and lack in real world applications - course does not deliver the skills required to apply concepts to real world applications
Access: Free
This Youtube playlist discusses the intuition of linear algebra from different perspectives and builds up understanding. It acts as great supplementary content to a course that a student is already currently doing in linear algebra.
Deep Learning (DL) is recongnised as a subdomain of ML and a wide majority of all modern applications of AI lie within this specific domain. DL can be considered as an extension of Artificial Neural Networks (ANNs) where the architecture of the neural network is tweaked. The greatest benefit of DL is that it eliminates the need of feature engineering as required in ML with the unfortunate drawback that increased computing power and more data is required. Most of today's computer vision (CV) and natural language processing (NLP) is based entirely on DL. An understanding of ML is highly recommended before attempting DL courses.
Access: Audit / Financial aid (2-3 weeks approval time) / Payment
This specialisation (a sequence of many courses) is considered as a logical extension from Andrew Ng's machine learning course and is recognised as the best introduction to deep learning by the entire global community for its simplicity and concept-based learning. This specialisation contains 5 courses.
Note: You will have to apply for financial aid for each individual course
Advantages:
- Andrew Ng dives into the algorithms used in machine learning by keeping mathematical involvement to a minimum whilst delivering all the intuition behind the mathematics used
- Students are expected to code the machine learning algorithms themselves
- Andrew Ng delivers practical advice on how to approach and solve machine learning problems throughout the course
- Python is used for programming assignments in the course (widely recognised programming language within the deep learning community)
- The most relevant python libraries (within the deep learning community) are used in assignments
- Computational power is provided to the participants of the course
Disadvantages:
- Delivering mathematical intuition becomes difficult due to the increased mathematical complexity in certain algorithms