Artificial Intelligence For PMs
We, Product Managers, are always trying to make our products better, more user-friendly, and better experience altogether.
In the pursuit to achieve this incrementally better product offering, we are always trying to predict, analysis, optimize, and automate our product offerings; this is where Artificial Intelligence or the more apt word Machine Learning comes in.
Machine Learning is the field of study that gives computers the ability to learn without being explicitly programmed — A typical text book definition
Well, that's all good, you make the computer learn! But how do you do that?!
It all starts with basic maths and statistics and I am not going into details about that but let me enumerate the different machine learning methods:
1. Supervised Learning
It is the simplest prediction you can do. You have data for a range, and you predict the next set of ranges. Remember, Y = mX + c, the straight-line formula we used in our schools?; It is all that. You have Y and X for a range, you find the slope and the constant, and use that data to predict that next set of ‘Y’s. By the way, this is called linear regression, and you guessed it because it is a linear form of data.
What if you don’t have a linear set of data, but there are only two sets of data, like: Email — Spam/Not Spam, Online Transaction — Fraud/Not Fraud, you get the picture; this is called a classification problem, solved by a Logistic Regression.
I hope you’re getting the picture, in supervised learning, you have some data and you find some analogy to it by building models. Now you use that model to predict (get output).
2. Unsupervised Learning
When you have only one set of data, that is all of them unlabeled, we try to find patterns or clusters among them to make sense of the data. This still requires learning algorithms other techniques; just that we only have one set of data — Xs and not Ys, so there is one more extra step to make sense of the data and find those Ys.
Learning Algorithms for supervised mainly are Linear Regression, Logistics Regression, and Support Vector Machine, and for unsupervised — K Neighbors Classifiers, Random Forest, Clustering, etc are used. Statistical techniques like gradient descent, backpropagation, AdaBoost, etc are used to build that predicting model. There are also techniques to simplify these equations by using — Principle Component Analysis, Regularization, Rectified Linear Units, etc.
Typical Use Cases of Supervised/Unsupervised Learning:
1. Predict Data — Change in prices, change in the economy or any form of linear prediction
2. Analysis Data — Spam Filters, Medical History Segmentations, Fraud Detection or any other form of segmentation
3. Optimize Data — Detecting User Preferences, Recommendation Engines, Delivering Personalised Experiences, etc
3. Neural Nets — Deep Learning
Now in the real world, we don’t have a small set of data, i.e. only X and Y; but have a huge set of data with a lot of inputs/features (Xs) and a lot of outputs (Ys), we can’t be using simple statistics to solve them. We use something called neural nets — similar to our brains; we use layers of equations to simplify the output to finally get that single output — A lot of Xs converting to one Y.
You can read a simple XOR Example using neural nets — https://medium.com/@jayeshbahire/the-xor-problem-in-neural-networks-50006411840b
There are many Deep Learning Algorithms, but the most widely used are — Convolutional Neural Networks for images and video learning and the Recurrent Neural Network for language understanding.
Typical Use Cases of Neural Nets/Deep Learning:
1. Computer Vision: Making computers understand images/videos as we do
2. Language/Text Analysis: Making computers understand language and semantics like us
3. Self Driving Cars: Use of above understanding and building self driving cars and other automations
This blog was to make everyone understand the nuances of Artificial Intelligence — Machine Learning and not go deep but if you still want to read more about Machine Learning and Deep Learning you can look at my past projects on GitHub — https://github.com/angshu-min-js/MachineLearning_Basics & https://github.com/angshu-min-js/machine_learning.
For Further Reading:
- http://cs109.github.io/2015/
- https://machinelearningmastery.com/basic-concepts-in-machine-learning/
- https://github.com/angshu-min-js/MachineLearning_Basics/tree/master/Machine_learning_complete_Notes
- http://cs229.stanford.edu/materials.html
- https://www.tensorflow.org/tutorials/
- https://github.com/angshu-min-js/python-machine-learning-book
- https://github.com/angshu-min-js/Machine-Learning-Tutorials
- https://github.com/angshu-min-js/awesome-machine-learning
