Machine Learning Tech Talk for Startups
Date Updated: Saturday, May 17, 2014
Machine learning is a field of study that lets you use computers to discover structure and patterns in data that are hard to see with the human eye or make predictions based on historical data. You should consider integrating machine learning in your business model if you have the following problems (Not an exhaustive list!):
- You have a business that is easy for new competitors to enter (i.e. The barrier to entry is low). As a result, your business is just one of the many similar product/ services in the market. You are commoditized by other competitors doing the same thing. Machine learning can solve this problem by increasing the barrier to entry for your business especially when you make the quality of your product or service dependent on the quality of your data.
- You are tired of spending time, money and energy working on routine, boring human tasks that cannot be automated by computers. Machine learning lets you automate human operations to increase your business productivity and lower cost as it enables computers to learn without being explicitly programmed.
- You have a mobile app or web based business and you are struggling to improve your business metrics like sales generation or customer engagement rates. Machine learning can let you customize your product or service offering to increase customer engagement and profits.
In summary, the nature of the opportunity in machine learning is that when your competitors integrate machine learning into their business, everyone else in your industry should employ those machine learning techniques eventually because it is a competitive advantage.
To make it relevant to the average person, I discuss 3 use cases of machine learning based on my experiences:
- Mobile optical character recognition of Asian Text. The use case is when you want to scale up a product concept with a trade off on accuracy.
- Predict housing prices around Seattle using Craigslist data. The use case here is in making predictions based on historical data.
- Machine learning in stock trading. The use case opportunity is that machines see patterns in big data more easily and perform tasks faster than humans.
You will also learn about the best practices in machine learning to improve the computer’s ability to make fast, accurate predictions. More details are covered in the tech talk. In summary, these come under the following themes:
- Improve the training data
- Modify the objective function
- Increase, reduce or change the features used (i.e. Feature engineering)
- Change the optimization algorithm
Some free useful resource links are also discussed during the talk so that you can still apply machine learning to benefit your business even if you do not have deep technical expertise. In fact, I believe that the best machine learning practitioners are those that know how to combine human intuition and wisdom with a machine’s speed and pattern recognition capabilities. This is because humans are good at the things machines are bad at, and machines are bad at the things humans are good at.
You can also grab the slide from the workshop here.