2015: three more in computer science,
2020: HTML, CSS, and JavaScriptTML and JavaScript,
2021: materials science course, and
then a couple three unfinished efforts after that.
This course went well and I mostly enjoyed taking it. Sure, pros and cons as always, but far more pros.
The course was first ‘published’ in the fall of 2017. Nine years is a very long time in the tech world yet not so long in the math world, since NNs are mostly math I don't think I suffered any loss. I bet, rather, that I gained in simplicity of presentation.
In June 2017, a landmark research paper titled “Attention Is All You Need” [AIAYN] was published by a Google research team.
This paper introduced the Transformer architecture, which completely revolutionized the field of artificial intelligence. Modern LLMs and generative AI would not be possible without transformers. Note: every LLM is a form of generative AI, but not all generative AI systems are LLMs, according to Toloka AI.
The paper was authored by eight researchers:
Noam Shazeer (born in the USA),
Niki Parmar (born in India),
Jakob Uszkoreit (born in Germany),
Llion Jones (born in Wales),
Aidan N. Gomez (born in Canada),
Lukasz Kaiser (born in Poland), and
Illia Polosukhin (born in the Ukraine).
Five years later, in November 2022, OpenAI introduced ChatGPT, a generative AI combined with large language models. ChatGPT, and its variations, have achieved a new level of artificial intelligence.and serves as the foundational backbone for modern large language models (such as GPT-4, Claude, and Gemini).
PROS of the course
- Dr. Ng is an excellent teacher.
- The course pace is just right (for me). It’s organized and presented as a 4-week course, probably presumes you are employed full time while doing the course on the side. As a retired person, I finished in 8 full days and I think I’m slow at this point. Yes it is self-paced learning.
- The material is presented, then reinforced verbally, then reinforced via the graded tests and lab exercises, one of each per week number.
- Beyond the lectures et cetera, the course was augmented by interviews with AI researchers.
- The difficulty level registered medium. The material itself is not difficult to understand. Overcoming the complexity of the matrix dimensions provides a mental workout.
- Coursera itself is good enough but it isn’t slick. The interface for doing the work is say A minus. The coordination between signing up, taking lectures, labs, and tests is more like C minus. Annoying at times.
- Dr. Ng’s audio is of mediocre quality. Hard to understand why that is so. It takes some time getting used to. I’m hard of hearing and I often use captioning but captions were not available in this course (as far as I know).
- A forum is advertised wherein a student might get help from others past and present. It didn’t work at all thanks to a revoked security certificate. So you’re on your own.
- The background of machine learning
- The basics of machine learning
- The meaning of forward and backward propagation
- Avoiding for loops: vectorization !
- Activation: Sigmoid vs Tanh vs ReLU vs Leaky ReLU
- Hyperparameters (the art in AI)
- Several NumPy
trickstechniques - Insight: Interview with Geoffrey Hinton. And let’s not forget John Hopfield.
- Viewpoint: Interview with Ian Goodfellow (invented GANS)
- How to recognize cat pictures: final lab project
