Train Your Own Machine: A Collaboration between the Microsoft Student Partners of Stanford and Georgia Tech

As a Microsoft Student Partner (MSP), I've been hosting workshops on technology-related topics for other students at Stanford every month. This winter break, I came home to Atlanta, and thought: what better way to spend this time than to host a workshop? My twin brother, Arvind Ramaswami, is also a Microsoft Student Partner -- so we hosted a workshop on machine learning, along with MSP Kirtan Patel from Georgia Tech.

Why is this important?

Machine learning has been transforming all sorts of fields! For example, machines can now understand what your photos are about and label them. Here's an example label from the Azure Computer Vision API:

Another example: Woebot uses tools from AI and machine learning to be a natural chatbot and companion to people who need therapy, helping transform the future of mental health.

These are just a few of the possible applications, which include text-to-speech, natural language processing, and more. The possibilities really are endless!

AI vs ML

To start, it would help if we knew what all these acronyms meant! Whereas AI (artificial intelligence) refers to search-driven algorithms that do things such as finding the shortest or the most optimal path, ML (machine learning) fundamentally involves learning from data.

For example, a ML algorithm may learn how a faucet looks like by looking at thousands of pictures of faucets. It can learn from data and make decisions on it -- just like how our brain works. In contrast, a typical AI algorithm would find the cheapest store to buy household goods or the shortest route to drive to work. Both search-driven and data-driven algorithms, however, are included in the common usage of the term "AI" though (which is why it's such a buzz word).

Additionally, we discussed a high-level overview of different types of machine learning, such as supervised and unsupervised, and the algorithms used to implement them. The big picture: algorithms such as neural networks are used for different types of machine learning.

To learn more in detail about what we discussed, feel free to check out our slides from the presentation.

Microsoft ML Studio

Next, we did an interactive workshop in which we showed how to use the Azure Machine Learning Studio. Azure ML studio lets you play around with AI and use powerful algorithms just by dragging and dropping pieces together.

We walked through one example; from Census data, could we predict the income of people? To make it simple, we would predict just one binary value -- whether someone's annual income is greater than or less than 50K, based on other factors such as age, gender, race, and marriage status.

In the end, our entire program looked like this: (No code!)

The steps include:

1. Split Data: Split the data into a training data set and a test data set. This way, around 70% of your data is used to train your model -- your program is "learning" how to classify a person, and what features affect income. Finally, the 30% of the rest
2. Initialize Model: Initialize a model (in this case, a multiclass decision forest).
3. Train Model: Train the model with the training data.
4. Score Model: Score the model by using it to predict incomes in the testing data.
5. Evaluate Model: Finally, check the accuracy of the model's predictions. Would another model have worked better?

After running the program, the model was trained and evaluated; and we were able to see the accuracy for ourselves. For just a couple thousand rows of data, 88% accuracy is really good, especially given that we didn't do any coding!

Additionally, Azure ML studio also lets you add scripts written in Python or R, and create Web APIs from your machine learning "experiments"; so it integrates well with code, if you choose to do so. It really is an interesting and intuitive tool. Try out the Azure ML studio at https://studio.azureml.net/

Workshop attendees

This workshop was different than most other workshops we've done in the past. We decided to reach out to the entire local community -- not just college students. Whoever was interested in learning about AI and machine learning could come to our workshop. And the results were surprising.

A large chunk of our attendees were college students -- but the vast majority was either professionals or high school students.

We found that people from all ages were interested in education and these topics. Indeed, machine learning and artificial intelligence have been transforming all careers -- and it will continue to do so. It is ever more important both for working professionals to familiarize themselves with these new technologies and for aspiring students to learn about these to get ahead of the game.

So what can you do today to learn more?

Additional resources

• For a start, the Azure ML Studio is a great way to play with and learn about machine learning algorithms and datasets.
• The Introduction to Machine Learning course at Coursera may be one of the best places to start learning about the theory and programming ML algorithms in Python. It is taught by Stanford Professor Andrew Ng.
• Kaggle has lots of datasets and competitions in which you can practice your AI algorithm skills and put them to the test.

Here are our slides from the talk:

Learn more about the Microsoft Student Partners at Stanford by joining our Facebook group; you'll be subscribed to our updates and events.

If you're at Georgia Tech, join the Microsoft GT group to get the latest updates about events near your area.