Introduction to Machine Learning and Artificial Intelligence Technologies. Discover the basics surrounding this tech, including business uses and evolution over time.
2. Agenda
Part I - Business Needs
- Use Cases
- Why Machine Learning?
Part II - What is ML?
- Big Data Explosion
- Types of ML
- Specific Examples
Part III - AI
- Turing Test & History
- Neural Networks
- Demo!
- Applications
- IT Landscape
Part IV - Q&A
4. Optical Character Recognition (OCR)
Steps:
1. Recognize Pixel Positions
2. Map Pixels to Numbers
Challenges:
- Many styles
- Needs to be quick
- Some characters are similar
In this presentation, I will describe how machine learning and AI has become an inseparable part of our everyday living. In the next slides I will go over business cases where ML and AI are used, why is it necessary and dive into a few details about how it is done. Then I will go into a more in-depth discussion of AI and recent progress in that field.
Let’s start by looking at some business use cases where complex algorithmical approach is needed.
OCR or Optical Character Recognition is a good example where the process of pattern recognition requires an immense amount of processing. For each number which needs to be recognized, we need to be able to understand all possible ways it could be rendered, down to the pixel level. The process involves analyzing each pixel of written character and determining which number its position corresponds to.
Another example of statistical analysis is product recommendations used by many online retailers. In order to be able to recommend a product that a customer may be interested in, a large number of parameters need to be taken into account and compared to historical patterns. Machine learning can help this task by continuously learning about new patterns and growing a list of consumer paths over time.
Speech recognition is one of the more complex examples where Machine Learning is absolutely necessary. In order to recognize human voice, the system need to hear different dialects, pitches and resonances, then be able to recognize phonemes and separate them from each other. And this is only for word recognition, now when you add a layer of comprehension, this becomes a truly complicated task.
In today’s word, there many processes which require very complex logic and processing. These are tasks which at some point have been an exclusive domain of human mind. However as demand grows and technology evolves, we are more increasingly looking into computers to accomplish these tasks.
Here are just a few examples where complex logic is necessary and is often enabled through Machine Learning. ML technology allows us to perform a complex computational logic which can adjust over time as new solutions and outcomes are added.
Now let’s take a look at how Machine Learning works
So how can Machine Learning accomplish these tasks. There is one necessary component and that is data availability. If you were trying to predict what a consumer would but based on previous purchases and you had transaction history for a few hundred other consumers, you would not be very accurate. Luckily, we have been collecting data at an ever increasing rate for the past decade and the amount of information we store is continuing to explode. This allows us to have the needed data to make statistical predictions.
There are three main types of Machine Learning. Supervised, Unsupervised and Reinforced. Each method has its own applications and is best suited for certain use cases.
Now let’s take a closer look at one of the more common application of Machine Learning - Supervised Learning. This implies that we have historical data which we can use to train the model in order to enable it to predict outcomes when new information is fed into it. Once we have provided enough data, the output is manually verified and the model is then allowed to make predictions in production.
Classification is one example of supervised learning. It’s most known application is email spam filtering. The system is trained to recognize which emails are likely to be spam and which are not, then that model is used to filter other incoming emails.
Regression is another common method of supervised learning. It is often used in value predictions, like stock market or weather forecasts. Here the values are all plotted in a chart and a line of best fit is calculated using least squares method. This can then be used to predict future values based on available data.
Unsupervised Learning is almost the opposite of Supervised. Here, we start with the raw data not knowing what we are looking for. The most common method being Clustering takes raw data and splits it into groups of like nodes. This is useful for predictive analytics, where there are many possible attributes which could end up being clustered and only become evident after the model is developed.
Another type of Machine Learning is called Reinforcement Learning. This is very similar to animal training, a system is initially trained similarly to supervised learning, but is not given fixed parameters on which to operate, so it is able to make decision outside of what was learnt in the training data. If that decision leads to a successful outcome, like a consumer purchase, the system is rewarded by reinforcing that decision path.
Now let’s take a look at Artificial Intelligence.
What is it? This video demonstrates what is commonly referred to as AI.
AI technology has been growing over the last few decades and have now made a number of significant accomplishments. In 1997, Deep Blue beats the world champion in chess and last year a system called AlphaGo beat one of the best players in Go.
Meet Mitsuku
What you have seen in the previous slide is called a ChatBot. While the desired results can be achieved by a well built algorithm, a true human brain simulation would have to be based on neural network concept. This is similar to how our brains work and consists of multiple linked machine learning processes each analyzing that data being fed into it, in this case, our human language and then passing the result to another node which would evaluate the data against a slightly different algorithm. This allows for multiple decision making points to occur without sacrificing the performance.
In today’s world, Machine Learning is more frequently used as a supporting framework for another business process. This includes Business Intelligence where predictive analytics can help predict changes in financial models or product supplies. Customer Service with automated Q&A, where AI components can almost replace human interaction.
The current ecosystem for Machine Learning technologies is very complex and is increasingly being folded up into a broader AI category. Some of the more notable players in the field which you may have heard of are Hadoop, which is an Apache software library that is used to store and manipulate large sets of data, essentially an SQL Server equivalent used in Machine Learning. Spark, another Apache product, a cluster computing framework used to process large amounts of data up to 100 times faster than Hadoop can do natively. TensorFlow is a Google library often used in complex Machine Learning projects and Neural Networks.