Add all required tags in the tag option (top-right of the post):
The platform (e.g., dl-ai-learning-platform or Coursera-platform.)
The course week/module
Write a detailed description of your issue.
Note: Screenshots are welcome, but please do not post full assignment code.
For more information, please check the FAQ category.
Hi, I really like the MLS courses taught by Andrew Ng as it allows me to gain better understanding of the math behind the codes to execute the models training/evaluation; building intuition of the “why” and “how it works” behind the “what” is incredibly rewarding in itself.
However, I do want to suggest revamping the course 2 and 3 content using Pytorch as the coding example instead of Tensorflow/Kera, due to the broader adoption of Pytorch my most AI platform and frontier labs.
I am concurrently taking Pytorch Fundamental to fill this gap myself, but really think the two courses - theory + practical application, can be efficient/effective. Thus beneficial to learner like me. Thanks
PyTorch Foundational course is purely focusing on learning the framework, and basic building blocks for NN training/evaluation/deployment. It is practical but feeling “robotic" without getting to first understand the “why” we use ReLu or Sigmoid etc, for example
Machine Learning Specialization course goes deeper behind the scene to establish the foundational knowledge, or at least a good introduction, into the math behind neuro-network. Personally, this is how I learn: theory first, code walkthrough, hands-on project to harden the knowledge retention.
While the theory I learn from MLS is helpful for me while taking PF concurrently, I had to mentally convert between Tensorflow and PyTorch in my mind’s eyes because MLS is taught in Tensorflow.
It might not be a big deal for more experienced MLE or DS, it is a friction point for a less experienced learner like me. Hence my suggestion to “unify” the two courses - either by updating the MLS with PyTorch code example, or literally combining the two courses for an intro to ML with PyTorch .
This is a valid suggestion, but don’t hold your breath for them to implement this. Rewriting a course at the scale of MLS or DLS in the way that you suggest would be a lot of work. And it’s really not clear how much value there is in that. TensorFlow is also very widely used in industry, so it is a valid choice to teach both platforms. Historically PyTorch was more popular in academia, but it has really gained ground over the last 5 years in industry as well. Up until quite recently with the introduction of the new PyTorch Specialization from Laurence Moroney, all the courses here used TF except for the GANs Specialization.
The choice of platform is essentially the same as choosing a programming language. Note that the first version of what is now MLS, which was called Stanford Machine Learning and was originally published in 2012 (I think), was taught using MATLAB instead of python. The concepts are the same whether you implement them in python or MATLAB or TF or Torch.
If your goal is to be an active participant on an ongoing basis in the ML space, it doesn’t hurt to learn both TF and torch.
@paulinpaloalto Appreciate the response and as a product manager myself, I can totally understand the level-of-effort (LOE) and impact trade-off point. And the analogy of programming language is valid.
My quest is exactly because I am aware in many tech companies, big or small, especially startups, DS and AIE (AI engineers) are increasingly shifting towards Pytorch. As a learning platform, I would assume it is the intention to prepare learners to tackle real world problems as closely as possible. Hence the suggestion; mainly, to check if there might even be work undergoing in the direction I desire.
That being said, I am aware it is a daunting overhaul It doesn’t help that the field also advances very fast with new architecture, models and techniques popping up literally every month. Typically, I don’t want to prescribe a solution as I am not sure how feasible this is but if I recall correctly, in MLS courses, the volume of coding examples are relatively low. Perhaps, only updating only the sections where there are coding examples as well as the labs.
In any case, I will continue taking both courses for the time being. Thanks again for everyone’s response.
It doesn’t help that the field also advances very fast with new architecture, models and techniques popping up literally every month. Typically, I don’t want to prescribe a solution as I am not sure how feasible this is but if I recall correctly, in MLS courses, the volume of coding examples are relatively low. Perhaps, only updating only the sections where there are coding examples as well as the labs.