Machine learning is used so often in tech now that it’s nearly impossible to say where it isn’t used. Google uses it. Amazon uses it. Facebook uses it. Everyone wants a piece of it. So, what exactly does a machine learning product need? What does it require from the Product Manager?
Product Manager at Uber shared what they use machine learning for at Uber, and talked about the landscape of machine learning in consumer products.
Product Manager at Uber
Devdatta Gangal is a computer scientist turned product leader with diverse experience from Yahoo Mail & Messenger. He’s generated hundreds of millions of dollars for Zynga, leading mobile apps and platform for Groupon, and currently leads the charge of scaling sensor inferences at Uber. Their mission is to understand the real time physical state of the world Uber operates in.
His team uses signal processing, machine learning, computer vision on raw location and sensor streams from millions of cars in the world. They then help partner teams understand driving safety, detect fraud, improve pick-ups & deliveries, and build better mapping services.
Machine Learning Products
Devdatta talked about Machine Learning and how it has gone from research and engineering to consumer products. He discussed the role of a Product Manager in building machine learning-powered consumer products, and what those products need. He shared his experiences and explained how there is no playbook for a Machine Learning Product Manager.
He talked about how to prepare for a machine learning filled product manager interview and product role on machine learning-heavy products. He also shed light on the role a Product Manager plays amongst smart engineers & researchers.
- What is Machine Learning? 3 Simplistic views of advancing machine learning techniques:
- Linear models.
- 1st generation nonlinear models.
- Deep learning.
- Linear regression can solve many elementary prediction problems.
- Other techniques solve for data not linearly separable, but require smart feature engineering.
- Deep learning uses stacks of linear models with nonlinear connections. “DNN’s are hungry for data but learn the features from data itself!”
- Evolution of machine learning in consumer products.
- The Business behind the scenes: computational marketing, credit risk (investment), high-frequency trading, fraud detection. Around 2-3 decades.
- Consumer assist and designs: web search, automated voice response, personal assistants, image, voice and face detection. Last 5 years.
- Medical decision assist, agriculture, self-driving cars, smart homes and smart bodies. The future.
- Machine learning at Uber. “The real uber experience is beyond the phone screens.”
- Destination prediction.
- Fair and price estimation.
- Location accuracy.
- Navigation (for drivers).
- Customer tickets.
- Fraud detection.
- How are machine learning products built?
- Building machine learning is a multi-step iterative process with many issues to be resolved and trade-offs to be made along the way.
- It involves various stakeholders need to safeguard everyone’s interests.
- What does the product manager do exactly?
- Identify the problem, and whether it needs to be solved, what is the estimated impact, how to measure it, and whether it needs machine learning.
- What is important with machine learning products?
- Quality of understanding the data.
- Latency of decisions.
- Cost of data collection.
- “Data is the new oil – handle it with care.”
- Who does the data belong to? Gmail and Google? Alexa and Amazon? Facebook?
Not sure about how Artificial Intelligence is different from Machine Learning? Check here.
Have any comments? Tweet us @ProductSchool
The Product Book has arrived! Learn how to become a great Product Manager. On sale for a limited time. Get your copy here http://amzn.to/2uJqg9A
We teach product management courses in San Francisco, Santa Clara, Los Angeles, and New York. To learn more about our upcoming courses and how to apply click over to our course page.