Here are some of the key takeaways from the summit:
1. Machine Learning is already here
Machine Learning (ML) was the main focus in the whole summit. A few customers are already running production workloads on various ML products offered by Google Cloud. All of the products are in General Availability with the exception of newly launched Video Intelligence API which is in beta.
All products have a demo page where you can play with the API. No need to sign up or login or anything. Check out these demos:
Cloud Vision API
I dragged and dropped a meme and got this as the output:
Cloud Natural Language API
I tried the demo with this text, a restaurant review I wrote on seamless earlier tonight:
The Good: Naan was good. It was soft and light. The Bad: Very few pieces of Paneer in Palak Paneer. Also, the Palak was not pureed/blended as one would expect in Palak Paneer, it was more like Saag Paneer. The Ugly: Basmati rice was stale, and smelling bad. Verdict: Not ordering from here again.
And this is the output I got:
2. The algo is free, but the data is not
When TensorFlow was first announced as an open source project, Wired magazine nailed it when they said “the algorightm is free, its the data that is precious”. They could not have said it better. This is apparent in the product offerings.
ML as a Service. Google Cloud Machine Learning Engine is a compute engine with a custom GPU based chip. With this machine, you can write your own TensorFlow based programs and run your training set. This is the free part where the software is free. But to do this, you will need ML engineers and a decent amount of training data.
Models as a Service. On the other hand, they have ready-made pre-trained models (such as the vision model and the natural language model) which you can consume via HTTP REST API. The data used for training the models, as well as the model itself are proprietary. You can only use them via API. This is where the barrier of entry is very low. Most web developers are familiar with REST APIs and can easily build applications based on these APIs. A lot of speakers were using curl command on their terminal for demos, showing how easy it is to use these APIs.
3. ML in customer service
Two case studies were presented showcasing companies using ML in customer service. The first case study was Ocado, a UK based online grocery store. They were using the Natural Language Processing (NLP) capability to do three things:
- Analyze the message and classify the ticket as high or low priority
- Analyze the message and route it to correct customer service team
- Analyze the message and close the ticket if it appears that the issue has been resolved
Read more about the Ocado case study.
Another case study involved a company that uses NLP on customer emails to decide:
- Which field technician (based on skill set) to send out to fix a problem
- What tools they should carry
- What spares they should carry
Obviously they are sitting on a mountain of emails linked to work orders to track which emails resulted in which spares etc. This was used as a training set to predict what spares might be required based just on the email from the customer reporting the problem.