Module Reviews: Y3S2
ALS1010: Learning to Learn Better
Lecturers: Robert Kamei, Magdeline Ng, Chng Wei HengAssessment:
30% Attendance
20% Survey Completion
15% Learning Plan 1st Draft
15% Finalized Learning Plan
20% Course Reflection Slide
Course Difficulty
The module is pretty much a series interactive and informative sessions on how to better your learning methods and discover some strategies to improve your learning. As different people have different learning methods, we get to explore each of these and the methods offered, such as blocked vs interleaved learning, doing a root cause analysis, chunking strategy, SMART goals, and many more.
The workload below is measured weekly.
Lectures - 2h
Compulsory participation, interactive PollEv sessions, I have not much comments on it.
Personal Opinion
Free 2MC if I would say, including the learning plan assignment itself. It is also an opportunity for you to reflect on your learning strategy so far because why not, who knows your productivity can really increase by employing these strategies!
Expected grade: CS
CFG1003: Financial Wellbeing-Introduction
Assessment:100% e-Learning Completion
Course Difficulty
Simply an OTOT e-learning session which covers many views, tips, and tricks on financial well-being, such as money narratives, forms of capital, principles on money management and budgeting, etc.
Personal Opinion
Pretty sure everyone takes this mod because of at least one of the two reasons: purely curious of financial well-being, or intending to take CFG1004 but this course is on the way.
Expected grade: CS? it's a 0 MC mod so
CFG3001: Career Advancement
Assessment:10% MCQ Quiz
10% conNectUS Profile
10% Informational Interview Invitation
30% Reflection Journal
10% conNectUS Forum Response
10% VIPS Worksheet
20% VMock Interview Report
Course Difficulty
The course covers different tips on advancing your career journey, starting from defining your brand, communicating it via resume and online profile, networking, gaining essential interview skills, knowing what to do on the first days of work, and getting to know some office etiquettes. One way to help one define his or her own brand is through the VIPS worksheet, which can be communicated via a conNectUS profile to gain new connections, which some may be available for an informational interview. Through this informational interview, you can learn new things that might go into your final reflection journal. Finally, VMock helps you to ace your interview skills as they can evaluate how well you answer the question as well as how well you behave during answering the question.
Personal Opinion
I feel that this course does give me some new things to learn. I had a great informational interview, overall an enjoyable course as I can actually practice and reflect on myself on how to go ahead in the future.
Expected grade: CS?
CS4225: Big Data Systems for Data Science
Lecturers: He Bingsheng and Ai XinAssessment:
25% Hadoop Assignment
25% Spark Assignment
25% Midterm Test
25% Final Test
Course Difficulty
The course covers different concepts about big data systems and how can one manage such vast amount of data, going through the pros and cons of different data systems one by one. It started with the 4 V's of big data (velocity, veracity, volume, and variety), principles of big data management, MapReduce and Hadoop, how it relates to relational databases and data mining. On the second half of the semester, the course covers NoSQL and Spark, explaining how BASE is used instead of ACID, as well as Spark's use cases in relational databases, streaming, and machine learning, wrapped up with delta lake(house), PageRank and how graph algorithms play a role in this course.
The workload below is measured weekly.
Lectures - 2h+1h
The lectures are recorded so I only managed to come face-to-face to less than half of the lectures. I like the pace of Prof He's lectures and how he regularly checks on whether we can keep up with it or not. I had no regrets coming to the venue to listen to it directly although most of the time I had errands to which I couldn't come for it. Prof Ai Xin's lectures are rather slower but it's not that slow, so some people may be able to keep up better with this pace. Clarity-wise they both did fine, so I had no issues understanding what was going on.
Tutorials - 1h
This is a rather quick tutorial so there is not much to cover. I never came to the live session of this so usually I would just skim through the questions and the solutions. Take note that this is not weekly, on average this is rather fortnightly.
Assignments
The first assignment is using Hadoop and MapReduce to implement something taught in the lectures. Code speed is not an issue in this assignment as long as you can produce the right answer. On top of this, you also have to submit a video explaining how your code works, so make sure you can manage your time to explain the possibly long code that you have.
The second assignment is using Spark to implement SQL-like queries and perform machine learning. This assignment is rather more straightforward than the previous one, but is given a reasonable time limit. Similar to the previous assignment, you also have to submit a video explaining how your code works.
Exams and Personal Opinion
I have a strong feeling that both midterms and finals will be left-skewed, which kind of explains why this module has a steep bell curve. That being said, I enjoyed doing the assignments a lot.
Expected grade: A-
DSA3102: Essential Data Analytics Tools: Convex Optimisation
Lecturer: A/P Tong XinAssessment:
10% Weekly Quiz Participation
15% Assignments (2)
25% Midterm Test
50% Final Test
Course Difficulty
The course covers various concepts usages of convex optimization, starting from the definition of convex functions, iterative methods to approach optimality, gradient descents (naive, SGD, projected), coordinate descent (Jacobi and Gauss-Seidel), KKT points, convex optimization in machine learning (linear regression and SVM), regularization, and finally subgradient and subdifferential.
The workload below is measured weekly.
Lectures - 3h
I enjoyed coming to his lectures. He will start his lecture slides and sometimes handwrites the workings on a piece of paper, making it a rather unique way of teaching. In the middle of the lecture, he would launch a PollEv on possibly funny or deep philosophical questions which is quite a nice intermezzo. His tone is rather calm, and that's what warms up the cold lecture theatre.
Tutorials - 1h
I never came to the tutorials and read the solutions directly instead, so I have nothing to say about the classes itself. The questions are well-designed, possibly foreshadowing what might come in the midterm and the final exam.
Assignments
Two assignments, rather straightforward, and you have plenty of time to do it. Not an issue.
Exams and Personal Opinion
Midterm's curve was supremely steep. Hopefully the final exam paid off?
Expected grade: A-
DSA4212: Optimisation for Large-Scale Data-Driven Inference
Lecturer: A/P Alexandre Hoang ThieryAssessment:
3x10% Assignments
30% Midterm Test
40% Final Test
Course Difficulty
The course covers many optimization methods that are initially similar to DSA3102 but eventually branches out differently, since it's more coding-oriented using Python, NumPy, SciPy, and JAX. It started with linear algebra and multivariable calculus reviews which are the fundamentals of this course, then gradient descent, introduction to neural networks and deep learning, second order methods like Newton's method, quasi-Newton methods, and gauss-Newton methods with Jacobian matrices. The second half of the semester involves mainly recommendation systems and how matrix factorization models work, regularization on function reconstructions, Bayesian statistics, and solving optimizations on expectations or integral forms with KL divergence and REINFORCE gradients.
The workload below is measured weekly.
Lectures - 2h+1h
This is thankfully (the only course that is) live on Panopto so I don't have to come directly or watch it at some other time beyond the lecture slot. Prof Alex usually goes through the lecture slides and discuss some of the in-slide exercise problems in a separate handwritten note. The pace of the lecture is good, so I have no complaints about it.
Tutorials - 1h
This does not happen every week but when it does, Prof Alex will just go through some notebooks prepared beforehand and implement codes related to what we learnt on the lectures. They mainly focus on the JAX version of the application.
Assignments
All three assignments are required to be done in a group of 1-4 students, up to your preference. Word of advice, be in a group of at least 3 or 4. The first assignment is about building a convolutional neural network (CNN) from scratch to classify images. To my surprise a part of the grading is on how the accuracy of the model is COMPARED to other groups. On top of building the model, we have to create a tidy report of up to 5 pages to explain the experiments done that led to the final model pipeline.
The next assignment is to predict anime ratings (yes, the dataset exists on Kaggle) using various methods including what we learnt in the matrix factorization chapter. Similarly to the first assignment, the model performance compared to other groups is still a part of the grading but this time it's the MSE since it's no longer a classification problem. The report requirement is still a thing.
The final assignment is about one of the topics selected by each group. This is rather an open-ended assignment and thus the page limit on the final report is slightly higher.
Exams and Personal Opinion
The midterm is well-made and the distribution was not so bad. However, half of the final exam consists of the same questions from midterm so I'm not surprised if the statistics are severely left-skewed. One complain I have from this course is that the assignments are all given after the recess week, meaning the three assignments have to be done in two weeks with no breaks in between, which is somewhat tight to my liking. Other than that, I absolutely enjoyed this course.
Expected grade: A
ST4248: Statistical Learning II
Lecturer: Lim ChinghwayAssessment:
4x5% Assignments
10%+18% Group Project
20% Midterm Test
40% Term Paper
Course Difficulty
The course covers the other half of ISLR and is a sole sequel to ST3248. This review heavily relies on my ST3248 review as it was also taught by Prof Chinghway. ST4248 covers more statistical tools uncovered in ST3248, such as non-linear regression such as polynomial regression and step+basis functions, regression splines and smoothing splines, local regression and GAMs, tree-based methods such as random forest and bagging, support vector machines or SVMs, kernel smoothing and KDE, neural networks (simple, CNN, RNN) and finally ensemble learning like stacking and boosting.
The workload below is measured weekly.
Lectures - 2h+1h
Very similar workflow to ST3248, so I'll just copy what I put there.
Basically Prof going through the slides. I found his teaching clear and concise, so I had no complaints whatsoever. Content-wise, you can technically learn from just ISLR and still do well in the course. I personally didn't read the book at all, but I can tell there are two parts of the lecture: the theory and the lab, where we get to touch with R codes.
Tutorials - 1h
Very similar workflow to ST3248, so I'll just copy what I put there.
Done right after the 1h lecture. The tutorials are pretty doable in my opinion, as Prof will go through it live anyways. There is solution PDF given, but he does have one or two things to comment on with regards to the question.
Assignments
Very similar workflow to ST3248, so I'll just copy what I put there.
The assignments are taken from one of the tutorial questions. This usually involves R coding, and explaining findings from our reports, be it a plot, a model fitting result, etc. There is a 4-page limit on your assignment report, so make sure to manage space well in your answers while still keeping the content clear and complete.
Group Project
In groups of 4, the project, like the previous semesters, is basically how you would apply what has been taught to solve a real-life problem. The project phases consist of the proposal, the live presentation, and the final report. During the live presentations, the other groups will peer-grade as well as the Prof himself. Just make sure you have a very good topic for the project and try not to simply touch-and-go. If you decide to employ various statistical learning methods, do explain thoroughly the reasoning of such choice.
Term Paper and Personal Opinion
The term paper is basically the same as the project but 4 days and you are on your own. The topic pool was given so you had to simply select one topic and one problem statement related to that topic. I personally didn't have enough free time for this during Week 13 which was a shame so I didn't think I did well for it. Probably my last ST-coded course for my entire undergrad journey so I'm glad that this course gives me sufficiently many statistical learning knowledge that I never heard of online or anywhere.
Expected grade: B