learning

What is learning for you?

Please pause for a second and define it for yourself before reading on. Is it about books?

English was difficult

Let me start with a story. When I was in the secondary school I was poor at English. I could not understand it why. I was very good at math and chemistry, and I was able to understand structures well. Such skills should have helped me to master languages. Unfortunately, this was not the case. Languages posed a challenge for me and I found English especially difficult.

At some point during my study I had to learn the difference between the Simple Past and Present Perfect tenses. This was difficult, despite my best efforts and long hours spent on studying the examples. The teacher made a test and I failed. I was very disappointed because I had the best intention to understand and I put the time into learning.

So, I said to the teacher “I disagree with the note. I have been studying hard for a few days.” Yet, she answered, “Perhaps you have been studying for days, but you have not learned it yet.”

The moral is this:
Learning is more than memorizing. Learning is more than studying examples. Learning is more than understanding structure.

What about a change in your life?

Let me now ask you. What would you like to change in your life?

Do you want to solve your health problems?
Do you want to loose weight?
Do you want to become fitter?
Do you want to start a new carrier?
Do you want to become an entrepreneur?
Do you want to find a mate?

How successful have you been so far?

Let’s say that you want to solve your health problems, e.g. heal from the irritable bowel syndrome.

How much have you thought about it? Perhaps a lot.
How much have you educated yourself on the subject? Perhaps a lot.
How deep is your understanding? Perhaps deep.

Perhaps you have read many books or articles on diet, nutrition, digestion, elimination, stress and so on. Perhaps you have inquired people around, asked for suggestions, listened to seminars, and so on.  Let me now ask you.

How much have you changed as the result? Perhaps not much.

Can you think your way out of poor health? Can you learn driving by reading a book or watching a video? Can you reason your way out of debt?   No.

The moral is this:
Thinking is not enough. Understanding is not enough. They are helpful, even necessary, but insufficient for learning to occur.

As long as you stay on the level of thinking, there is little chance for growth. Why? Because you need to integrate the knowledge into the working of your body. You need a direct experience.

Pattern recognition – how we learn from data

Let me now briefly tell you about pattern recognition. In a basic scenario you want to discriminate between two classes, say apples and pears moving on a conveyor belt. The task is automatic sorting. A camera makes a photo of a passing fruit and the system needs to detect which fruit it is and sort it accordingly.

To solve the problem you first start with the labeled data: a set of raw images with individual apples and pears. They are labeled. In addition, you may be given other measurements such as weight or size.

You start by finding a meaningful representation of the raw images in the form of mathematical descriptors. These are often characteristic features, e.g. related to various shape characteristics. Next, the challenge is to choose a function that will discriminate between two classes based on the extracted features. The task is to learn this function.

There are multiple approaches and models possible and within each approach there are multiple, even infinite, candidate functions. How do you start?

You make a selection of a few promising models based on your experience, literature (what other people reported that worked for them for similar problems), understanding of the problem and initial data analysis. Then, for each approach the labeled data is used to train the related discrimination function, which means that the data is used to define the parameters of that function.

But it is not as easy as it sounds.

Some measurements of particular apples are more important than others as they influence the parameter values more strongly. Perhaps these are examples of typical apples.

Some measurements may also be faulty and lead to poor estimation. Perhaps these are examples of other fruits such as small mangoes that were mistakenly labelled as apples.

Some features may be meaningless (unspecific), giving similar response to both apples and pears.  Ideally, these concerns should be incorporated in the way the parameters are estimated.

Let’s say we have learned the function. Now, it is the test time.

This is often the most time-consuming step as various scenarios, strategies and sub-strategies are tested with respect to a number of selected functions. During this validation time, the function is being tested on the labeled data but unused in the learning of the function. In this way, we evaluate how good the function is at predicting the correct labels for fresh data.

In real problems, the results are sub-optimal, even poor, at this stage. It is just the first reference.

What you do next is to go through a repetitive cycle of small improvements on all levels. You investigate multiple factors. For instance, you look at the data to understand the incorrectly assigned cases. Perhaps these are border cases that need a separate treatment (i.e. another function to be learned).

You also look at the appropriateness of the data used for training, the usage of atypical or problematic examples, choice of discriminative features: adding new ones, extracting better ones or removing some, the usage of multiple functions that focus on specific aspects of the problem and use a voting scheme for the final labeling and so on. Or perhaps you even abandon the model you have chosen first and select another one.

And you test extensively all simple updates made. This all happens because a function can be learned perfectly on the given training data (give zero error) yet behave poorly on new, unseen data.

It is a tedious process in which you go through the cycle of sequential changes to find small improvements at each stage so that the overall performance is greatly enhanced. Yet, this process still requires a conscious human partner – the one to set criteria, observe, make choices and decisions.

If you now think that sorting apples and pears is an artificial example I can assure you that similar tasks are being automated on all levels in real life. These include sorting luggage on airports, detecting faulty planks in a factory, automatic recognition of post codes, computer aided diagnostic for malignant tissues in X-rays or ultrasound, speech recognition and so on.

How do automatic systems learn from data?

The moral is this:
Learning is a process in which informative data is collected and represented for the task (data and knowledge organization), the discriminating function is chosen and its parameters are well estimated such that the whole system performs well on new data, i.e. the discriminating function makes little or no error. This is achieved through a repetitive cycle of improvements.

The development of pattern recognition techniques was inspired by human learning. Isn’t now the time for us to be inspired back? Learning is practised through little updates.

What is learning?

In the light of personal growth, learning is about the change in behavior. We now understand behavior broadly as abilities, skills, habits, practices or actions to be taken.

There is no learning if there is no change in the behavior. We often make the mistake thinking that we are learning when we are reading books, memorizing techniques, analyzing problems or thinking about them. But we are merely collecting information and organizning it into structures or perhaps knowledge. Even understanding complex phenomena is not yet learning.

Learning truly occurs when there is an update or a change. Think about it. We can study all the books on driving and analyze road scenarios, but unless we simply start driving and practice, there is no way to develop this skill. The same principles apply to all areas in our lives. Be it becoming healthy, loosing weight or running a business.

We can understand the problem and know what to do to solve it. But unless we do what needs to be done, observe the results, reflect, draw the conclusions and update our approaches for the better results, our situation will not change much, even though we hold the best intention possible. So, let me summarize.

True learning is about

  1. Information collection. (Data collection)
  2. Observation and thinking. (Visual inspection of the data.)
  3. Understanding and knowledge organization. (Problem understanding. Data representation. Selection of features.)
  4. Action or practice. (Learning the discrimination function)
  5. Reflection. Drawing conclusions. (Testing! Analyzing what needs change. )
  6. Change.
  7. REPEAT the steps 1/2 – 6 until satisfied.

In order to learn we need all the steps above. We may also need to go through these steps multiple times to find a satisfactory change.

***

So, I’ve learned to discriminate between Simple Past and Present Perfect when I started to practise their usage in real scenarios. I’ve learned about the impact of green smoothies by making them and drinking them daily. I’ve learned to listen actively by coaching others especially in conflict situations. I’ve learned cooking simply by doing it.

What about you?

Are you a data collector, knowledge organizer or a learner?

If you choose to be a learner, it makes sense to adopt a practical approach. Select one aspect that you want to change in your life. Choose one book or one idea relevant to your situation. Which one to choose? Take the one which is either the most attractive or the easiest to implement.

Then simply do what they suggest. If there are multiple choices – simplify as much as possible. Your goal is the first direct experience. You will improve later on.

Test the idea and reflect on how it affects you – your energy levels, your emotions, your thinking, your physical being, your relations and so on. See what the results are. Make your best guess how to update the idea towards an improvement in the final result. Apply. Test it. Reflect. Repeat.

If the idea works, improve it further on. If it doesn’t work and you see no way of improvement, abandon the idea and move on to another one. If it works well, make it a habit.

If you want to create a change in your life, learning is the necessary step for transformation.

You need to make the ideas tangible by integrating them with your being on the physical plane. There are so many insights and observations you can gain from a simple application that no reading or thinking will ever provide. Testing gives you the real taste. It is much better to take one idea and test it than staying knowledgable but stuck.

Embrace change. You need to become it in order to be it.

***

Photo copyright by Moyan Brenn. Photo available under the Creative Commons license on Flickr.

***

Learning and generalization posts:

Print Friendly, PDF & Email

Categories: LearningNerdyPattern recognitionPerformance

Like this post? Subscribe to my RSS feed and get loads more!