Thursday, February 29, 2024
HomeAILearn deep learning with the G test [2022| !

Learn deep learning with the G test [2022| !

 G test

I think there are many people who have heard the word “deep learning” on TV more and more. Some of them may be thinking about taking the G test.

Therefore, in this article, we will introduce the deep learning G-test.

Table of Contents 

  • What is deep learning
  • What is the G-test?
    • Overview
    • Test range
    • Exam schedule
    • Difficulty/pass rate
  • Contents of the G test
    • G-test example
  • Advantages of taking the G test
    • Gain systematic knowledge about deep learning
    • Can be used for work
    • Advantageous in finding a job or changing jobs
    • Can participate in the community
  • G test acquisition is recommended for the following people
    • People involved in data science projects
    • Beginners learning data science
  • Recommended study method
    • Learn from books
    • Learn online
  • Recommended books for the G test
    • Deep Learning Textbook Deep Learning G Test (Generalist) Official Text
    • Textbook for deep learning utilization
    • Shortest Breakthrough Deep Learning G Test (Generalist) Workbook
    • Will artificial intelligence surpass humans? What lies ahead of deep Learning?
  • Recommended courses for the G test
    • Qualification square
    • AVILEN
    • Agarute
  • Summary

What is deep learning

Deep learning is a machine learning technology that automatically analyzes a large amount of data and extracts the features of the data.

It is called deep (multilayer) learning because there are multiple intermediate layers of a neural network (NN) modeled on neurons, which are the brain neural circuits of humans and animals.

This multi-layered structure enables deeper learning of data features.

What is the G-test?

Here is an overview of the G-test.

Overview

The ‘G’ in G-test is the ‘G’ in ‘generalist’. A generalist in the G test is a person who has basic knowledge of deep learning, and has the ability and knowledge to determine appropriate utilization policies and apply it to business.

Those who pass the G test are given the JDLA (Japan Deep Learning Association) logo to show that they have passed the test. It can be said that it is a test to produce human resources who can propose solutions to problems in combination with business after understanding the above.

Test range

The test scope is as follows. You will be asked whether you have various knowledge about a very wide field.

①What is artificial intelligence?

Definition of artificial intelligence

(2) Trends in artificial intelligence

Search/Inference, Knowledge Representation, Machine Learning , Deep Learning

(3) Problems in the field of artificial intelligence

Toy Problem, Frame Problem, Weak AI , Strong AI , Physicality, Symbol Grounding Problem, Feature Design, Turing Test , Singularity

(4) Specific methods of machine learning

Questions about representative methods ( supervised learning , unsupervised learning , reinforcement learning ), data handling, and evaluation indicators

⑤ Overview of deep learning

Neural networks and deep learning, problems with existing neural networks, deep learning approaches, CPUs and GPUs, deep learning data volumes, activation functions, learning rate optimization, further techniques

⑥ Deep learning method

CNN, deep generative model, application in image recognition field, speech processing and natural language processing field, RNN, deep reinforcement learning, robotics, multimodal, model interpretability and its correspondence

⑦Toward social implementation of deep learning

AI project planning, data collection, processing/analysis/learning, implementation/operation/evaluation, law (personal information protection law, copyright law, unfair competition prevention law, patent law), contracts, ethics, current discussions (privacy, bias, transparency, accountability, ELSI, XAI, deepfake, diversity)

From Japan Deep Learning Association

Exam schedule

The G test is conducted two to three times a year. There are three exam dates for 2022:

  • March 5, 2022
  • July 2, 2022
  • November 5, 2022

Difficulty/pass rate

The degree of difficulty is relatively high, partly due to the wide range of exams. There are many questions that require college-level mathematics and an understanding of theoretical backgrounds, so a certain amount of study is essential to pass.

Now let’s look at the success rate. As you can see from the chart below, the pass rate for the G test is not low at 60-70%.

Contents of the G test

The G-test asks questions such as:

There are three main types of questions, all of which are multiple-choice questions, such as four-choice questions that fill in the blanks, questions that select one word from a group of words and fill in multiple blanks, and questions that ask whether a sentence is correct or wrong.

G-test example

Next, I will introduce the specific questions asked in the G test.

Trends in artificial intelligence

▼ ProblemRead the following sentences and select the option that best fits each of the blanks.

The first AI boom occurred in the 1950s. Around this time, programs called artificial intelligence were solving problems based on (a).

In particular, (B), developed by IBM in 1996, is famous for its victory over world chess champion Garry Kasparov.

However, there was a problem that only problems called (C), such as mazes and puzzle games with fixed rules and settings, could be solved, so research declined.

(ah)

  1.  knowledge representation
  2.  expression learning
  3.  machine learning
  4. Exploration/Inference

(stomach)

  1. deep blue
  2. Bonkras
  3.  Ponanza
  4.  Sharp

(hare)

  1.  A/B testing
  2. pattern matching
  3.  toy problem
  4.  dartmouth workshop


▼Answer

(A) 4. Exploration/Inference, (B) 1. Deep Blue, (C) 3. Toy Problem

 

▼ ProblemSelect all that are correct about the international image recognition competition “ILSVRC2012”.

  1.  Image recognition is the task for which deep learning can achieve the highest accuracy as of 2017.
  2.  ImageNet is a dataset for handwriting recognition.
  3.  The winning team is SuperVision, led by Professor Jeffrey Hinton of the University of Toronto.
  4.  The results of this competition were called “a breakthrough in artificial intelligence research in 50 years.”


▼Answer

1.3.4

Problems in the field of artificial intelligence

▼ ProblemThe terms listed below are issues that were raised during the second AI boom. Pick one that best describes each issue.

(a) Frame problem
(b) Symbol grounding problem

  1.  It is difficult to systematize the vast amount of human knowledge.
  2.  It is difficult to select and consider only the necessary information from a huge amount of information.
  3.  It is difficult to associate symbols such as words with the meaning they represent.
  4.  It is difficult to develop a computer to process a huge amount of knowledge.
  5.  It is difficult to maintain the Internet to obtain sufficient data.


▼Answer

(a) 2, (b) 3

 

▼ ProblemChoose the two most appropriate explanations for ” strong AI / weak AI “.

  1.  ” Strong AI ” is called an expert system and is still widely used today.
  2.  What is called AGI (Artificial General Intelligence ) is closer to ” strong AI “.
  3.  The development of a “computer that thinks like a human” in the original sense of the term triggered the third artificial intelligence boom.
  4.  In international image recognition competitions, ” weak AI ” has achieved identification performance that surpasses that of humans.


▼Answer

2, 4

 

Concrete method of machine learning

▼ ProblemChoose the one that best matches the combination of words that fill in the blanks.

Problems in supervised learning can be broadly divided into two types according to the type of output value. (A) The problem is used when the output is discrete and we want to predict a category. On the other hand, problem (B) is used when the output is a continuous value and we want to predict the continuous value itself.

  1.  (A) LIMITED (B) GENERAL
  2.  (A) Partial (B) Full
  3.  (A) Classification (B) Regression
  4.  (A) linear (B) nonlinear


▼Answer

3

 

▼ ProblemThe following sentences describe various machine learning methods. Choose one of the word groups that best fits the blank.

There are several methods in machine learning, and it is necessary to understand the meaning of the terms correctly. A method that uses missing data is called (a). There is also a method (c) in which correct labels are given only to some samples.

  1.  unsupervised learning
  2.  supervised learning
  3.  reinforcement learning
  4.  expression learning
  5.  multitask learning
  6.  semi-supervised learning
  7.  manifold learning


▼Answer

(a) 2, (b) 1, (c) 6

 

▼ ProblemChoose one of the following words that best fits in the blank.

There are various performance indicators for classification problems. Here, we consider binary classification that divides samples into two classes, positive and negative. Also, if you want to focus on reducing false positives (FP), use (b) to reduce false negatives (FN). (c) should be used when focusing on .

  1.  Correct answer rate
  2.  realization rate
  3.  Cooperation rate
  4.  Harmony rate
  5.  precision
  6.  Recall
  7.  f-value
  8.  p-value
  9.  t-value
  10.  z-value


▼Answer

(a) 1, (b) 5, (c) 6, (d) 7

 

▼ ProblemIn machine learning, the principle is to divide the teacher data into several parts and use only some of them for learning. Select the one that is most appropriate for the purpose of adopting such an approach.

  1.  To save computational resources in the initial stage by learning with a small amount of data once.
  2.  To remove samples with outliers in the data.
  3.  Semi-supervised learning can be done even if some of the data is unlabeled.
  4.  To correctly estimate the performance that the model will exhibit when deployed.


▼Answer

Four

 

Deep learning overview

▼ ProblemSelect all that apply as reasons why deep learning has rapidly achieved high results in recent years.

  1.  This is because it has become possible to perform learning in a realistic amount of time thanks to improvements in computer performance due to advances in semiconductor technology and high-speed parallel processing using GPUs.
  2.  The development of neuroscience has made it possible to reproduce the structure of the human brain that corresponds to the task, such as visual and language areas for image recognition and natural language processing.
  3.  Due to the spread of the Internet, it has become possible to obtain a large amount of data that allows highly expressive models to avoid overfitting.
  4.  The invention of backpropagation made it possible to train multilayer neural networks, which had been difficult until then.
  5.  Many frameworks for deep learning have been developed, making implementation easier.


▼Answer

1, 3, 5

 

▼ ProblemRead the following sentences and select the option that best fits the blank.

The steepest descent method, an optimization method used in conventional machine learning, is called (a) because it uses all the data for one learning. Therefore, a method called stochastic gradient descent is often used. The method that uses only one sample is called (a). Both have advantages and disadvantages, and it is recommended that (c), which uses a certain number of sample groups, be adopted.

  1.  set learning
  2.  batch learning
  3.  online learning
  4.  point learning
  5.  sampling learning
  6.  Mini-batch learning


▼Answer

(a) 2, (b) 3, (c) 6

 

▼ ProblemWhen I trained a neural network model, I observed the error against the test data. At that time, the error continued to decrease smoothly until the number of times of learning exceeded 100, but after that the error gradually increased. Choose the one that is most appropriate as the reason.

  1.  This is because as the number of learning times increases, the value of the error function is less likely to be updated.
  2.  This is because as the number of learning times increases, the optimization will be performed only for the learning data.
  3.  This is because as the number of learning times increases, the number of parameters that must be updated at once also increases.
  4.  This is because as the number of times of learning increases, the time required for calculation processing also increases.


▼Answer

2

 

Deep learning method

▼ ProblemRead the following sentences and choose one of the words that best fits in the blank.

One of the international image recognition competitions is ILSVRC (ImageNet Large Scale Visual Recognition Competition). In 2014, (a), which uses a structure called the Inception Module, won the championship, and (c) also achieved an excellent achievement close to it. (d), which enabled learning of a deep network called learning, won the prize.

  1.  AlexNet
  2.  Elman Net
  3.  GoogLeNet
  4.  ImageNet
  5.  LeNet
  6.  ResNet
  7.  VGGMore
  8.  WaveNet


▼Answer

(a) 1, (b) 3, (c) 7, (d) 6

 

▼ ProblemRead the following sentences and choose the option that best fits the blank from each word group.

In neural networks, (a) was used as the activation function in the middle layer in the early days. This is an important problem called the vanishing gradient problem.
Compared to (a), (c), which is often used as an activation function in deep learning, is less prone to this problem. It is also characteristic that the amount of calculation is small. On the other hand, it is known that learning progresses quickly by using (d) even when (a) is used as the activation function.

(ah)

  1.  step function
  2.  ReLU
  3.  sigmoid function
  4.  softmax function

(stomach)

  1.  The output becomes constant when a negative value is input
  2.  The mean value of the output is not 0 and the standard deviation is not 1
  3.  Function has non-differentiable points
  4.  If the absolute value of the input is large, the output becomes almost constant.

(hare)

  1.  step function
  2.  ReLU
  3.  sigmoid function
  4.  softmax function

(workman)

  1.  Dropout
  2.  Batch normalization
  3.  Regularization
  4.  Weight Decay


▼Answer

(a) 3, (b) 4, (c) 2, (d) 2

 

Deep learning research field

▼ ProblemSelect one of the following statements that best fits in the blank.

The application of machine learning is also being promoted in the field of robotics. For example, there are many examples of using algorithms (a) such as Q-learning and the Monte Carlo method for robot motion control. In addition, since the robot has a system that can collect different sensor information such as cameras (visual), microphones (auditory), pressure sensors (tactile), etc., this information is processed in an integrated manner by DNN. There is also research (c) that attempts to generate a series of robot motions with a single DNN.

(ah)

  1.  End to End Learning
  2.  Supervised learning
  3.  Motion learning
  4.  Adaptive Learning
  5.  Reinforcement Learning
  6.  Representation Learning

(stomach)

  1.  multimodal
  2.  Inception
  3.  Cognitive
  4.  full scratch

(hare)

  1.  End to End Learning
  2.  Supervised learning
  3.  Motion learning
  4.  Adaptive Learning
  5.  Reinforcement Learning
  6.  Representation Learning


▼Answer

(a) 5, (b) 1, (c) 1

 

▼ ProblemSelect the most suitable reason why RNN (Recurrent Neural Network) has contributed to the improvement of accuracy in the field of natural language processing.

  1.  By performing convolution processing in the convolution layer, it became possible to read the context from the appearance position of the word.
  2.  Because it became possible to retain past information in the hidden layer, and it became possible to extract the meaning from the sequence of characters.
  3.  By providing a memory part outside the network, it became possible to easily refer to the pattern of sentences.
  4.  This is because it can automatically learn repeatedly until it can output correct sentences.


▼Answer

2

 

Advantages of taking the G test

Next, I will introduce four benefits of taking the G test.

Gain systematic knowledge about deep learning

The first is that you can gain systematic knowledge about deep learning.

As we have seen above, the G-test tests various knowledge related to deep learning and AI. Therefore, you can inevitably learn these knowledge while studying for the G test.

Can be used for work

The second is to be able to use it at work.

As introduced, the G in the G test means generalist. In other words, the G test requires a person who has basic knowledge of deep learning, the ability and knowledge to determine appropriate utilization policies and apply it to business.

Therefore, the knowledge gained from the G test can be used in business.

Advantageous in finding a job or changing jobs

The third is that it can be advantageous in finding a job or changing jobs.

I introduced the merits that can be used at work, but it means that a person who has the G-test qualification has the ability to utilize deep learning in business. If you pass the G test, you can put the logo on your business card, so if you want to find a job or change jobs to a company that wants this kind of person, just having the G test qualification will be an advantage.

Can participate in the community

The final advantage is that those who pass the JDLA exam called “CDLE” can participate in a community established in 2018 for the purpose of exchanging information.

If you join, you will be able to participate in study sessions and events held in this community, so you can deepen your knowledge.

G test acquisition is recommended for the following people

Next, we will introduce what kind of people we can recommend to take the G test.

People involved in data science projects

The first is a person involved in a business related to data science.

The G test measures the ability to utilize basic knowledge about AI in business. . Therefore, the knowledge gained through the G-test will also be useful for data science projects.

Beginners learning data science

And they are the first to learn data science with little knowledge of AI.

Since the G test has basic content, it is a perfect qualification for those who have never studied AI or deep learning before. Through studying the G test, you can acquire basic knowledge about AI and deep learning, as well as the ability to utilize it.

Recommended study method

I think there are many people who want to take the G-test but have never studied deep learning and don’t know how to study.

So, here are some recommended study methods for the G test.

Learn from books

The first method is to study by yourself using books.

The advantage of reading books is that they are cheaper than taking courses. Currently, many books have been published for preparing for the G test, including some published by the Japan Deep Learning Association.

With books, even people who usually don’t have time to study can study in their spare time, such as commuting to work or school.

Learn online

Another method is to study online.

There are various types of G test preparation courses, and you can choose the course that suits you. Also, some courses have a curriculum, so it is recommended for those who do not know what to study.

Recommended books for the G test

Here are some recommended books.

Deep Learning Textbook Deep Learning G Test (Generalist) Official Text

This book is the official text of the Japan Deep Learning Association. If you are thinking of taking the G test, I recommend reading it once.

Textbook for deep learning utilization

This is also a recommended book for the G test by the Japan Deep Learning Association. We will explain the impact of deep learning based on case studies.

Shortest Breakthrough Deep Learning G Test (Generalist) Workbook

This book is a collection of problems that study the G-test and cover important points. Recommended as a textbook for practice.

Will artificial intelligence surpass humans? What lies ahead of deep learning?

It describes how deep learning will change AI in the future. The points pointed out here are important perspectives for taking the G test.

Recommended courses for the G test

Below is a list of recommended online courses.

Qualification square

In this course, even those who are studying AI and deep learning for the first time can study using easy-to-understand materials.

AVILEN

At AVILEN, if you meet the completion conditions of this course within one year and fail, you will receive a full refund of the course fee.

Agarute

This course uses teaching materials written by the instructor and is very easy to understand.

Summary

In this article, we introduced in detail the content of the test, example questions, and study methods for the Deep Learning G Test.

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Recent Posts

Most Popular

Recent Comments