Last week, Apple revealed more updates to its iPhone which specially target Neural Networks and Machine Learning as well as updates to ARKit. As more companies invest in Machine Learning we decided to talk with Kevin Scott who talks about Machine Learning on his website: thekevinscott.com.
How did you get into machine learning?
I’ve had my eye on the field for many years. However, the large learning curve and hardware requirements had put me off. Recent amazing advances in the field, particularly with reinforcement learning and creative computation, made me decide I had to get involved.
I come from a design background, and I’m particularly interested in how AI and machine learning can help augment human creativity and intelligence. There’s a huge opportunity for folks not coming from Computer Science backgrounds – people who are designers, artists, and humanists – to get involved and help build the next generation of AI tools.
What are you most excited about in regards to the future of machine learning?
We are still in the very early days of this technology coming into wide usage.
Andrew Ng often refers to AI as “the new electricity” and I think it’s an apt metaphor. I believe Machine Learning will be ubiquitous in every aspect of our digital lives.
Machine Learning has traditionally required huge investments in both computation and data, giving large technology companies insurmountable advantages. In the past few years, that’s begun to change.
On the hardware side, a number of startups (and more recently, Google) have begun offering more and more powerful GPUs for cheap or free. On the data side, we’ve seen progress with techniques like synthetic data generation, few-shot learning, and transfer learning, allowing programmers to achieve extremely accurate results with very little data.
There is a ton of opportunity to shape and contribute to the field, and we’re at a unique time where you can have an outsized impact.
I think there’s a huge opportunity for folks not coming from CS backgrounds – people who are designers, artists, and humanists – to get involved and help build the next generation of AI tools.
Do you have a simple explanation of machine learning for people who are unfamiliar with it?
Machine Learning is the technique of teaching computers to recognize patterns in data.
It has its roots in the field of statistics and math, and in fact statistical methods continue to be used for many problems in the field of machine learning.
Deep Learning is a subset of machine learning, and refers to neural networks that have many hidden layers (aka, “deep” networks). Deep learning models, when given access to lots of data, can become extremely accurate. They also have the ability, to some extent, to “figure out” the patterns in the data themselves (as opposed to more traditional machine learning techniques like feature engineering).
There is a ton of opportunity to shape and contribute to the [Machine Learning], and we’re at a unique time where you can have an outsized impact.
How do you see big technology companies taking advantage of machine learning?
Every technology company is investing massive resources in machine learning. They all accurately recognize it as the next major sea change, and recognize that investment in AI technology will result in better products in both enterprise and consumer software.
Apple has been baking machine learning capable tools into XCode and Swift for years now. CoreML and create-ml are two examples of this.
Google has probably the most recognizable efforts in AI today. You can see AI at work in Gmail’s auto suggestions; in Google Photos’ assistant, automatically suggesting styling options and tagging; in Google Translate; and in Google Home, answering your queries.
What specific area do you think is the most exciting?
I’m most excited about transfer learning and one shot encoding. Both techniques allow regular folks to get great results without access to powerful hardware or large data sets.
Technology-wise, I’m most excited about the idea of GANs. GAN stands for Generative Adversarial Network, and the idea is to take two neural networks and pit them against each other. For instance, one application has a neural network attempting to generate realistic celebrity faces, while another attempts to pinpoint the fakes. The final result is indistinguishable from real photos. These kinds of tools will enable great new bounds in human creativity and innovation.
What is the best way someone can get started in learning more about bots and machine learning?
Fast.ai and Andrew Ng’s Coursera course are probably the most popular beginner intros to deep learning. They’re both wonderful introductions to the field.
From there, it’s largely up to your specific interests. Images and vision, natural text, reinforcement learning, and others are all differing specialties that require their own sets of skills. Kaggle is a great data science website that posts challenges you can compete in as well.
My recommendation would be to figure out what excites you about the field and focus on a single area. I’m particularly interested in applied machine learning, and so my focus has been on inference in the browser, leveraging transfer learning.
How can the average businessperson leverage the power of machine learning?
If you’re sitting on a large amount of data there are likely opportunities for machine learning. A good rule of thumb is to look for things having to do with prediction. Anywhere there’s patterns, it’s likely that a machine learning model could generate predictions automatically.
If you are interested in learning more about Machine Learning, I highly recommend subscribing to Kevin’s blog to learn more.