What it means to study AI and machine learning: What I gained from taking the GCI (Global Consumer Intelligence) course

Reasons for taking GCI

The reason I took this course was that I was interested in AI and machine learning. During graduate school, I learned and used machine learning and deep learning.
My research at the time involved using AI to move robots. This is a hot field now, like physical AI. (I want to write about this in detail another time.)
—This is a slightly geeky part, so please feel free to skip it.—
The machine learning models and methods I have actually implemented include SVM (Support Vector Machine), MLP (both classification and regression tasks), CNN, DCGAN, metric learning, semantic segmentation, image recognition, and deep reinforcement learning (I played around with things like robot autonomous driving using Actor-Critic and DQN with ROS and Gazebo in a virtual space).
(I used Chainer for deep learning implementation. Those were the days.)
————————————————————

Even after graduating from graduate school, I remained interested in AI and continued to read AI papers and articles. On the other hand, my work was quite mechanical, and even when I explained AI enthusiastically to my boss, it didn’t get through well, making it difficult to actually connect it to my work. (Around 2022-2025)
However, recently the accuracy of generative AI has improved dramatically, and even those who are not familiar with AI have started using generative AI like Gemini, ChatGPT, Copilot, and Claude on a daily basis. And many companies have started to consider whether they can incorporate AI into the workplace. My company is no different, and we have started thinking about what we can do using AI. So, I strongly felt that I wanted to do something related to AI again.
I was somewhat confident in my knowledge of machine learning, but I had never actually implemented a system using AI in society, and some time had passed since I was heavily involved with AI, so I wanted to relearn the basics of AI. This GCI course allows you to learn the basics of machine learning, data analysis, and business, and since it matched my goals, I decided to take it.

What is GCI?

1. Explanation of GCI

GCI stands for Global Consumer Intelligence, and it allows you to acquire data analysis skills that can be a weapon in various fields. The course covers techniques for effectively visualizing analysis results, the basics of machine learning, and how to handle databases, aiming to help you stand at the entrance to becoming a full-fledged data scientist.
The course consists of 15 lectures, held once a week for 1 hour and 45 minutes each. The content comprehensively covers how to use the Python programming language, an overview of machine learning (especially logistic regression and decision tree-based methods), business applications of AI, and the basics of databases. In addition to lectures using materials, you can learn while actually running sample code.
In addition to the lectures, there are attendance questionnaires to summarize what you learned each time (15 times in total), 7 assignments, 2 competitions, and a final project.
(It is possible to complete the course without submitting everything perfectly. Since you’ve come this far, I hope you will tackle everything seriously, though.)
With generative AI now widespread in the world, a great many people are taking this course.
Although it is designed for beginners, it is a versatile learning system that can be educational even for those who already have some knowledge of Python and machine learning. It is recommended for those who want to learn about AI and machine learning but don’t know where to start. A wide range of people, from elementary and junior high school students to retirees, are taking this course.

2. Appeal: GCI’s famous competitions

One of the interesting things about this GCI is the two competitions held in total. There are competitions where participants compete against each other using what they learned in the lectures and assignments. I cannot go into details, but you compete on the accuracy of tasks such as predicting or classifying something using data. In doing so, it is a practical format where you analyze data using Python’s numpy, pandas, and matplotlib, and train models using machine learning. Sample code is provided so that even beginners can easily practice, so if you just want to predict and submit, you can learn in a practical format without any burden. However, it is difficult if you aim for the top in the competition. In addition to what you have learned, you need to do your own extra research. The level of the top participants is quite high, so even those who are not beginners can enjoy learning and improve their skills.

3. Appeal: The final and greatest hurdle, the final project

To complete this course, you must submit a final project. This requires you to make full use of almost everything you have learned in the lectures so far. I cannot go into details about this either, but it tests not only Python, machine learning, and data analysis, but also business perspectives and presentation creation skills. It is a final project that makes you feel like you have actually become a data scientist. It is a bit difficult and demanding, but it is a very rewarding project. And if you meet the requirements and submit it, you can complete the course.

4. Completion requirements

There are two types of completion requirements. I was able to achieve Advanced completion.
[Standard Completion] Course completion and understanding
1. Attend a certain number of lectures (or watch the archives) and submit the questionnaire. (Attendance registration is possible for 1 week after the lecture)
2. Submit the final project and receive a certain evaluation (Planned content of the final project: Create a business proposal based on what you learned in this course. Details will be explained when you take the course.)
[Advanced Completion] Skill acquisition
1. Attend a certain number of lectures (or watch the archives) and submit the questionnaire. (Attendance registration is possible for 1 week after the lecture)
2. Submit homework and exceed the set threshold.
3. Submit all competitions (Kaggle format) and meet the set requirements.
4. Submit the final project and receive a certain evaluation (Planned content of the final project: Create a business proposal based on what you learned in this course. Details will be explained when you take the course.)

What I gained from GCI and the fun of GCI

Since I had originally studied AI extensively and had experience implementing it, I thought I was in a position to review the basics of AI rather than learn new things. However, I was able to learn much more than I had expected before taking the course.

1. Learning about numerous methods

I think many people who think they know AI to some extent actually started studying from deep learning and know almost nothing about methods other than neural networks. In fact, generative AI like ChatGPT and Gemini is based on these neural networks. The accuracy of other things like image generation, physical AI, video generation, image captioning, meeting minutes generation, and voice generation has also improved dramatically, but neural networks are used in most of these models. I was one of those people; I had implemented neural network-related things and studied them quite a bit during graduate school, but I only vaguely understood the names and rough algorithms of methods before that (at a level written in the E-qualification).
This course uses other machine learning models rather than neural networks. What I learned in particular was decision tree-based methods. I knew that Gaussian process regression, SVM, and genetic algorithms could have high accuracy depending on the task you want to perform, but I had assumed that decision trees were not very useful indicators. However, in this course, decision trees such as RandomForest, ExtraTrees, XGBoost, LightGBM, and CatBoost are often handled, and these methods have short training times and are very accurate. On the other hand, because neural networks have too much expressive power, they are basically not handled in this course, and even when I actually implemented and compared neural networks, the prediction accuracy was not very high.
In this course, you can learn various methods such as decision trees and logistic regression in a practical format.

2. Adding features (Feature Engineering) ← One of the important things in this course

In this lecture, you can learn in a practical format not only machine learning but also what kind of features (explanatory variables) to use as input to train a model. You can input the data you have as is, but by performing appropriate feature transformation, the prediction accuracy of the model improves. In this course, you can acquire these feature engineering skills. It was very good to be able to learn again about basic but very important things, such as generating new features from original data (generating meaningful features, dimensionality reduction, etc.), handling missing values and outliers, converting text data (categorical variables) into numerical features (encoding), and scaling features through standardization.

3. Being able to learn about implementation and business aspects

Through assignments, competitions, and the final project, you can gain practical implementation experience, which you can immediately apply to your own work or research after completing the course. Also, since there are opportunities to think from a business perspective during lectures and the final project, another appealing aspect is that you will be able to think about what kind of things AI can be used for.

4. Competition

You can learn even more deeply by competing with other students to improve the accuracy of your machine learning models in the competition. As I mentioned above, if you want to aim for a top spot in the competition, you need to study beyond what you learned in the lectures. This was extremely interesting and I got so absorbed in it that I sometimes found myself working on it from morning until night. Nowadays, pre-trained AI is widespread and taken for granted, and just being able to use it can make people treat you like an AI expert, but in this competition, you have to think about the learning methods yourself, so it is a great opportunity to learn about the depth of AI and machine learning.

Difficulties I faced in GCI

What I struggled with in GCI was balancing it with other things. I had a lot of things I wanted or needed to do, such as work, housework, strength training, English study, and hanging out with friends, so it was hard to find the time. In particular, the competition and the final project were open-ended and could be worked on indefinitely, so it was busy. During the GCI course period, I made sure to set detailed plans and goals. By doing so, I was able to balance it with other things without wasting time.

Important things for completing this course

It is to learn proactively. Your passion for learning AI and machine learning will be directly reflected in your learning results. If you are planning to take this course in the future, I hope you will have high aspirations and learn actively. However, the learning system is designed so that even beginners can learn, so if you are interested, I think you should definitely take it. Also, it is important to work according to a plan. There are many things to do, such as lectures, assignments, competitions, and the final project. If you work according to a plan, it is not that difficult. I also worked by setting detailed deadlines and goals.

This might be a bit geeky, but I plan to write passionately about the appeal of AI at a later date.
Thank you for reading this far.