In the fifth course of the Deep Learning Specialization, you will become familiar with sequence models and their exciting applications such as speech recognition, music synthesis, chatbots, machine translation, natural language processing (NLP), and more. By the end, you will be able to build and train Recurrent Neural Networks (RNNs) and commonly-used variants such as GRUs and LSTMs; apply RNNs to Character-level Language Modeling; gain experience with natural language processing and Word Embeddings; and use HuggingFace tokenizers and transformer models to solve different NLP tasks such as NER and Question Answering. The Deep Learning Specialization is a foundational program that will help you understand the capabilities, challenges, and consequences of deep learning and prepare you to participate in the development of leading-edge AI technology. It provides a pathway for you to take the definitive step in the world of AI by helping you gain the knowledge and skills to level up your career.
Great hands on instruction on how RNNs work and how they are used to solve real problems. It was particularly useful to use Conv1D, Bidirectional and Attention layers into RNNs and see how they work.
Such a nice instructor and very good course material to understand the basics of Deep learning. I really enjoyed this course , Thanks for making such online course for us. Once again a big thanks.
the assignments were a really good format for someone who hasn't learned how to derive wrt multiple variables. It made sense to have the formulas provided to introduce a context for me: a developer.
brilliant course, great quality instruction from Andrew Ng. The only faults are that some of the labs have not been supervised properly being a but buggy and a couple of later lectures were very dry.
So many possibilities will be presented in front of you after this course. The only limit is the boundary of my imagination and creativity, that is how I feel now upon the completion of this course.
One of the best thing from this class is not only we can understand the concept of RNN, LSTM, etc, but also I also get the idea about how these technique can be used in many daily life applications
The course is very good and has taught me the all the important concepts required to build a sequence model. The assignments are also very neatly and precisely designed for the real world application.
The previous courses raised the bar and expectations. The assignments for Week 1 and Week 2 were a bit unclear. Lectures for Week 1 and Week 2 can be improved as well. Besides, this is a great course!
Hope can elaborate the backpropagation of RNN much more. BP through time is a bit tricky though we do not need to think about it during implementation using most of existing deep learning frameworks.
Loved the course - it was very interesting. It is also pretty complex, so will probably go through it again to review the concepts and how the models work. Thank you for this wonderful course series!
The lectures covers lots of SOTA deep learning algorithms and the lectures are well-designed and easy to understand. The programming assignment is really good to enhance the understanding of lectures.
Cant express how thankful I am to Andrew Ng, literally thought me from start to finish when my school didnt touch about it, learn a lot and decided to use my knowledge and apply to real world projects
Showing: 20 of 3,850
Tons of editing errors in lectures, and the programming problems rely more on knowledge of Keras (essentially untaught throughout the course) than they do on understanding of lecture material. A disgraceful end to an otherwise solid course sequence.
This course is by far the weakest out of the 5 course sequence. I did well in it (96.8%) but I think the programming exercises did not help build understanding of sequence models. Often I found myself just trying to get through the programming because I felt it was more an exercise in reading Keras documentation. I think you can pass this course without a solid understanding of what is going on in the sequence models. The programming exercises should be revamped to focus more on understanding what is happening in the program rather than trying to figure out Keras syntax (which is also useful, but perhaps better suited for a prep course).
Full of appalling errors that have been present for over 1 year. No one fixes it. It is clear that since Ng was let go by Stanford and Baidu, he is trying to earn a living with deep learning_ai. This apparently is not working as the small income from Coursera is not sufficient. As a result the prerecorded classes remain on Coursera to accrue some residual income. But, Andrew Ng and the staff are apparently gone.
Sadly, since these classes are no longer based on REAL Stanford classes the quality has gone downhill. I would recommend not taking the deeplearng_ai classes. Stick to classes offered by currently employed professors at established universities--preferably classes that ARE the same as the university classes or, at least, those derived from actual classes.
I really like the deeplearning.ai specialization. And also I like the Sequence Models course. However, I feel that I have learned less during this course comparing to the other ones in the specialization. First, I believe it was an extensive use of Keras. Whereas the framework is great, it would be much better for understanding if all the exercises were in numpy, whereas Keras tween-projects be optional. Doing both numpy and Keras versions would allow to better understand the material and learn through repetition. Second, even though the course is great, I perceived the number of errors/typos was much higher than in other courses. Is that true? For example, the Jazz Improvisation exercise was a nightmare. Overall, thank you for the course. Despite those problems, I would still recommend it.
Keras is required to pass the assignments but no training provided for it. I can learn it myself of course but then the question is this - what am I paying for?
The lectures were great. Andrew is a wonderful teacher, but the assignments were beyond miserable. Jupyter notebook is probably the least stable, most infuriating piece of software that I've been forced to interact with. I spent countless hours trapped, not able to perform the most basic of operations, such as saving my work or submitting. I lost work innumerable times. I, like others, eventually resorted to "saving" my work by copying it into a text editor on my system for fear of Jupyter sabotaging me. Even if the system was stable, most of the assignments were worthless as learning experiences. The majority of "programming" boiled down to playing a cryptic game of fill in the blank. Bottom line, I wouldn't recommend this class to anyone in its current state. Especially as a paid service. I really expected better quality.
To set the context, I have a PhD in Computer Engineering from the University of Texas at Austin. I am a working professional (13+ years), but just getting into the field of ML and AI. Apologies for flashing this preamble for every course that I review on coursera.
This course is the 5th and final one in a 5 part series offered by Dr. Andrew Ng on deep learning on coursera. I believe it is useful to take this course in order and it makes sense to study it as a part of the series, though technically that is not necessary.
This is one of the best courses to take if you want to understand the basics of Sequence Models (Recurrent Neural Networks). RNN is a technically-difficult-to-understand, still-evolving field of Neural Networks, and it has thus far found remarkable uses in a wide variety of field, ranging from Natural Language Processing (NLP) to Voice-to-Text conversion and Music Synthsis, to name a few. Dr. Ng really exposes us to this cutting edge research, by explaining research papers that were only recently published. By now, I could see how the problems would be tackled. However, there are several subtle aspects, such as the optimal metrics to use, the clever modification in the NN architecture, etc. which Dr. Ng drew attention to and made clear.
The instructor videos are very good, usually 10 min long, and Dr. Ng tries hard to provide intuition using analogies and real-life examples. The quizzes that accompany the lectures are quite challenging and help ensure that the student has understood the material well. As with the other courses, the programming exercises are the best part of the course. You get to practice, (1) Music synthesis, (2) NLP and Sentiment Analysis, (3) Trigger Word Detection (Hello Google, Hey Siri, Alexa!), ... All these problems are actual, real-life projects, which are extremely difficult to solve. They help the student practice the strategies and also provide a jump-start for the student to use the code for their own problems at work or in school.
Overall, this is an excellent course. Thank you Dr Ng and the teaching assistants, Thank you coursera.
I HAVE A HUGE GRIPE WITH COURSERA's TECHNICAL SUPPORT. THEY DO NOT HAVE A READILY AVAILABLE TECH-SUPPORT EMAIL ID. YOU HAVE TO SEARCH THROUGH THE WEBSITE CAREFULLY TO FIND A CHAT LINK. I HAVE RECOMMENDED COURSERA TO SEVERAL FRIENDS AND MANY OF THEM ARE VERY UPSET AT THE SUBSCRIPTION POLICY, WHICH IS SNEAKY. IN FACT THE WORDING IS ALMOST DESIGNED TO CHEAT YOU.
I have been a huge supporter of Coursera and hate to give this negative feedback here. I would have easily subscribed to many other useful and important coursers that Coursera offers, but will now be doubly careful about doing so.
I am so grateful that Andrew and the team provided such good course, I learn so much from this course, I am so excited that see the wake word detection model actually work in the programming exercise
5 stars for the very informative lectures. I especially appreciated the amount of content related to de-biasing models. Such an important step to take!
2 stars for the HW assignments, however. I felt like I was just translating comments into code and that I didn't really learn enough to do sequence model development on my own without the hand-holding.
I went through all course of DeepLearning.ai Specialization. While I was more, or less happy with first 4 courses in this specialization, I have a very bad impression regarding "Sequence Models" course. Actually, quality of the courses is gradually declining, starting from 5 stars for very first course in specialization, ending with 2 stars for "Sequence Models".
"Sequence Models" course is *disappointing*. It leaves you with bunch of scratches on the surface of technology without any details and/or understanding of how technology works. In simple words, it is not the course, it is bad overview of just few technologies. In order to get similar comprehension of technologies delivered in first four courses of this specialization you will need to spend a lot more time digging for information elsewhere.
Same issue with homework assignments. They are a bit helpful for technology understanding to some small extent, but they do lack depth.
Whole "Sequence Model" course looks like compromised/failed attempt to explain fairly complex material to newcomers. As a result, newcomers won't understand anything due to complexity of the matter, experienced engineers won't take away anything due to simplified explanations and absence of details.
If you do not plan to get a DeepLearning.ai Specialization certificate, I do not recommend to buy this course. Unfortunately, it will be a waste of your time and money.
Videos are okay, but exercise is just debuging!
I want to thank Andrew Ng and his team for the amazing work. You definitely make the world a better place sharing this knowledge, and it is an inspiration.
To the contents: the course covers many uses of sequence models, for many different formats (many-to-one, many-to-many...), the questionnaires are focused but comprehensive and the programming exercises cover a wide range of difficulty levels, from no-brainer-one-liners (most of them) to implementing LSTM backprop by hand (optional). They take away the dirty work from you but make sure you get how you would do it. At the end you get to work with pretty complex setups like the attention model, but you still get the feeling of knowing how it ticks from the very bottom up.
The actual merit is that, even if it feels simple, it actually does work and is a takeaway knowledge that can be directly applied for personal setups. And mr Ng's videos are a charm, you can totally feel the care. Glad to see him back after so many years :)
Cheers
The course is very good and has taught me the all the important concepts required to build a sequence model. The assignments are also very neatly and precisely designed for the real world application.
Many topics are only quickly skimmed over. Serves as an overly brief introduction to RNN.
Professor Ng seems a little bit confused about the subject and is making unnecessary analogies rather than going deep into the algorithm and explaining the context as he did in Convolution Neural Networks course. I hope that the videos are revised and professor explains the topic more clearly rather than depicting himself to be confused as well on the topic.
Learnt a lot about new concepts in RNN and LSTM. Really wanted to learn about these models. This course helped a lot. Everything was new and so fascinating. Loved this course and our teach Andrew NG.
Loved the course - it was very interesting. It is also pretty complex, so will probably go through it again to review the concepts and how the models work. Thank you for this wonderful course series!
The lectures covers lots of SOTA deep learning algorithms and the lectures are well-designed and easy to understand. The programming assignment is really good to enhance the understanding of lectures.
Great hands on instruction on how RNNs work and how they are used to solve real problems. It was particularly useful to use Conv1D, Bidirectional and Attention layers into RNNs and see how they work.
super unorganized!
really really bad