In order to use Wide Learning™, data for learning must be input into the AI. Upload the data for learning for identifying mammals that you just downloaded to the Wide Learning™ Trial Tool to see what it contains.

When you upload data for learning, the contents of the data are displayed in a table. Each row represents the data for one species of animal. The data for learning includes a total of 88 species of animals: 40 mammals and 48 non-mammals.

In the Trial Tool, you can view the contents of the uploaded data for learning as a bar graph.

As a test, click "1 -3. Check the data for learning", click "Item" on the left, and then click “Hair/fur” in the row on the right. A bar graph is displayed showing how many species of "animals with body hair" and "animals with no body hair" there are.

Now let's look up "whether or not they have wings". Click "Wings" on the left.

Now, if you look at the data for learning for identifying mammals, you can see that there are some items that are likely to be useful and others that are not.

But now we are deciding "for some vague reason" whether or not it will help us determine if an animal is a mammal.

But “how much” would each item help you determine if an animal is mammal? Also, "having wings" does not seem to help you determine if an animal is a mammal, but how about combining it with other items?

It is not easy to think about these things in your head, but Wide Learning™ can automatically derive the items and combinations that can help you determine if an animal is a mammal.

Click the "Generate combinations of important items" button. (Generation takes some time)

When generation is complete, Wide Learning™ will display the combinations of items that have been determined to be important (useful in determining whether an animal is a mammal or not) in a table.

Each line displays the details of a single combination. The leftmost column, "Combinations of Important Items", indicates the combinations of important items found by Wide Learning™.

Here, let's check the statistics for each of the items of the data for learning.

Click “Step 1 Let’s Prepare Data for Learning!” and then "1-3. Check the data for learning".

A screen showing the statistical information for each of the items you checked earlier is displayed. Here, click "Item" on the left, and take a look at the graphs for when "Wings" is "Does not have", "Spine" is "Has", and "Breathes using lungs" is "Yes".

It seems that it is not possible to determine whether an animal is a mammal only using the features of "has no wings", "has a spine", and "breathes using lungs".

However, if we take these 3 features into account, we can see that while 40 species are mammals, there are only 7 animals that are not mammals, meaning there is a higher chance animals with these features will be mammals.

In this way, Wide Learning™ finds items that are difficult to identify by themselves, but can be identified by combining several items.

Click “Step 2 Let's check the combinations of important items!” and return to the screen showing the list of combinations of important items. Currently, they are arranged in order of the number of mammals.

Now click "NEG (Negative) Count" twice. This sorts the combinations of important items that you have discovered using Wide Learning™, starting with the ones that are most likely to identify animals that are not mammals.

This way, you can see combinations of items that might be useful in identifying animals that are not mammals.

Let's get the AI to learn from the combinations of key items we have found. Press the "Weight important combinations" button.

When learning is complete, you can see how much weight is attached to the important combinations you have identified.

Here, "Weight" is a numerical expression of how important a combination of items is in determining whether an animal is a mammal.

If the weight is positive and high, the combination is an important basis for determining if animals are mammals.

On the other hand, combinations with negative weight provide an important basis for determining that animals are not mammals.

Click "NEG" at the top right of the table. Now you can see the combinations with negative weight.

Click "3 -2. Individual Learning Results".

Here you can see both the positive combinations and negative combinations for each animal in the learning data and their weight, and which one has more weight.

For example, check the radio button to the left of the "Aardvark" row in the table to the left.

A doughnut chart and table appear on the right.

The doughnut chart shows the combinations of important items in the Aardvark data and their weight percentages.

Pink represents the combination of the key items on which "is a mammal" is based, and blue represents the combination of the key items on which "is not a mammal" is based.

Let's choose "Carp" from the table on the left. This time, a doughnut chart which almost entirely blue is was displayed. It seems that most of the important combinations of items for carp are learned as effective combinations to distinguish them from mammals.

Please check some other animals.

Although there are differences in the ratios of pink and blue, roughly speaking, it seems that the AI has learned about all animals correctly.

Let's use the AI to make some new predictions. Click the "Proceed to Next Step" button.

To make predictions, you must upload the target data to the Wide Learning™ Trial Tool.

The method is the same as for uploading data for learning, so you should be OK.

Let's upload the data for learning to identify mammals that you downloaded earlier and check the contents.

Once you have successfully uploaded the prediction data, you will be able to see the prediction data statistics and so on, just as you would when uploading data for learning.

Here we will omit the explanation of confirmation, but please note that there is only one difference from the case of data for learning. There is no solution label for "whether it's a mammal". The purpose of the next step is to predict this solution label.

Click the "Start Prediction" button.

The prediction result confirmation screen is similar to the screen of "3 -2. Individual Learning Results".

The difference is that to the right of the "No." column are the "Result" and "Score" columns.

This "Result" is the result of the prediction, meaning that "POS (Positive)" means "is a mammal" and "NEG (Negative)" means "is not a mammal".

"Score" is the value calculated by the AI to determine if each animal is a mammal. If it exceeds 0.5, the animal is determined as "is a mammal".

First, let's look at a dolphin.

Let's look at a vulture. The vultures have a lower score because of the weight of the combination of items used to determine if they are not mammals.

Let's look at a platypus. Platypuses has been identified as mammals with a score slightly above 0.5.

In fact, platypuses actually are mammals. You can see that the presence of body hair and a spine, the lack of wings, and breathing using lungs provide a basis for identifying mammals.

Click "NEG" to see a combination of criteria to determine if an animal is not a mammal. This shows that platypuses are very rare mammals with the features of being oviparous and toothless. Nevertheless, the AI was able to correctly identify platypuses as mammals because there were many combinations of features that led to the identification of them as mammals.

That concludes the explanation of this case study. Please try it with your own data.