Accelerating machine learning adoption

Last week I had an amazing opportunity to present about machine learning and innovation to 110 people leaders at my company. It was a mixed audience of technical and non technical people.

My message is that machine learning needs to be understood by everyone in the business, not just the data scientists. Domain knowledge with machine learning will really enable successful data projects.

Here are my slides and some notes

A few months ago I followed a tutorial on http://course.fast.ai and entered a Kaggle.com competition. The competition problem was to label 20,000 images of cats and dogs using computer vision. There were 1300 entries from around the world.

I downloaded an existing, freely available algorithm – vgg16 and slightly modified it to get 87% accuracy.

I thought this was a great result. It’s REALLY hard for computers to tell what is in these pictures. 5 years ago a team of scientists got 57% accuracy with this same dataset.

However it turns out I didn’t do very well:) I came 600th! The winners got closer to 97% accuracy.

We’ll never all be data scientists but the technology is at a point where anyone new to machine learning can download great solutions and start solving these problems. You as domain experts are in the best place to see these opportunities and start experimenting.

What is machine learning and why is different to what we do now? This is very simplistic but with traditional computing we would tell the computer exactly what result we wanted for a given set of inputs. With machine learning we give the computer a large amount of information and we ask the computer to give us insights in to the data.

We don’t write explicit programs. The ‘program’ is an output from the data and will change based on the data.

It does this using some well known and well studied mathematics. Data scientists even have a cheat sheet for which algorithm to use. For more difficult aspects of machine learning like Deep Learning, there are some very good models available for free online. I downloaded one of these for cats and dogs.

But we’re not here to learn the cheat sheet so forget about the detail.

Just remember that the algorithms are well known for a given problem.

What gives companies an advantage in machine learning is their data.

We have an incredible set of users here. And they’re giving us some great data. Crunching all this data costs money.

One of the reasons you hear a lot about machine learning recently is that computing power has gotten very cheap. I spent just $150 for a few hours of computing from amazon for cats and dogs.

Why now? Exponential innovation…

Every few years for the past 100 years the amount of computing power you can buy for $1000 dollars has doubled. We are just at the tail end of the most recent technology advance – semi conductors. This pattern means that right now for roughly $1000 dollars you can purchase the same amount of computing power as a mouse’s brain.

If this trajectory continues then by 2024 for that same $1000 dollars you will be able to purchase the same amount of computing power as the human brain.

Now this is a wacky idea and I don’t believe it myself. But that’s perfectly normal! Humans are really bad at thinking exponentially.

 

If I ask you to walk 30 steps linearly then that’s easy to picture, 30 meters. However if I ask you to walk 30 steps exponentially, doubling every step – 1m, 2m, 4m, 8m. Then by the 30th step you will step billions of meters. The final step will take you 26 times around the world!

We can’t think this way but this is how fast and how cheap computers are becoming.

There are three things we can do to help accelerate adoption of machine learning throughout the business.

Realize that machine learning is absolutely accessible and it’s not magic once you know what types of problem can be solved.

There are 5 major types of problem…

v
Classification
Regression
Regression
Clustering
Clustering
Ranking
Ranking
Anomaly Detection
Anomaly Detection

Think about problems in your part of the business that can be phrased this way.

We need to collect better data, not just more data. We need to collect relevant data and this is where your domain knowledge is vital.

You are also in the perfect position to identify gaps in our current data. We should find these as soon as possible and start plugging them.

We need to identify any possible external sources of data, council data for example.

You should identify areas of the business where we are making subjective decisions. If we can eliminate ambiguity and subjective decisions from the business we can make better decisions.

Collaborate with data scientists – your domain knowledge combined with the skills of our data science is what will produce the best results

Don’t silo data. ask your data team where you should push data so everyone in the business can access it.

Don’t be afraid of sensitive data. We can anonymise the data and still get great insights from it.

We have some slack channels. Anyone can join the channels. We are all learning and these are a safe space for any level of knowledge in machine learning.

 

We will be running classes ranging from this type of over view information all the way to implementing real solutions.

So don’t be afraid of machine learning. Here we have a huge loyal user base and they’re generating amazing data, we have a group of the top technical and business talent in the country.

But our industry is changing faster than we can imagine and we need to use every tool available to keep our advantage in the future.

Think of machine learning as another technology or tool like Word, Excel or Photoshop. Learn about it. Get involved.

Please get in touch if you would like more information.

Some images from this amazing article on wait, but why: http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html

The slides and idea for exponential innovation from this great talk by Kaila Colbin: https://www.youtube.com/watch?v=XwxwVSJcOGU

Course: http://course.fast.ai

Startup Weekend Auckland May 2017

Wow, it’s a week after Startup Weekend Auckland. I’ve had time to get my thoughts straight about the whole experience. I learned a hell of a lot of positive lessons about idea validation, about myself and about working in teams.

Friday Night

We started the event on Friday night with a brief introduction to the format and some key organizers. Thirty enthusiastic 60 second pitches quickly followed. There were ideas for B2B services, tournament platforms, monetizing Facebook, recipes and shopping.

Andres and Billie Friday night
Andres and Billie Friday night

The problem that resonated with Billie, Andres and I was presented by Steffi who spoke about infertility and how it could be reduced with simple education. I was interested because I know a family that is affected by infertility and it was the only idea that wasn’t immediately commercial.

After the pitches we went out to meet Steffi and ask her if she would like to have us on her team. Our team was quickly joined by Andy who works in property and Ally who works in marketing. By the end of the voting we had a very well rounded team.

We stayed at the event until 1am Friday night researching the idea. We learned about existing solutions and tried to figure out which part of the huge demographic we would try to target. It was a productive evening learning about the problem.

Saturday

We got to the second day and immediately got in to working out an informational site. We got strongly challenged by the mentors about not thinking through the greater problem.

We got some great feedback on the first pitch. One mentor warned that saying “ins and outs” may cause laughter issues” – We would turn this in to a positive by the final pitch.

Re finances – “If you can only crowdfund $3,000 in the first year no one gives a shit”. I didn’t think this way before the weekend but it’s so true. $3,000 these days isn’t much at all.

Another mentor said we should have more optimism – “What beyond a website would help these people?”. This really resonated with me and got me thinking that an informational website wasn’t a great idea.

Some problems started to develop in our group. There were passive aggressive comments around the table and it was a terrible environment. I found it very difficult to confront strangers. This made me realize how great it is to work in such an honest, direct workplace at Trade me.

Lesson 1 – There’s no time for bullshit, be honest and up front

At around 3pm on Saturday I lost passion for the informational site. I felt like any criticisms of it weren’t being listened to so I decided to use the time to learn some technology and just settled in to coding a template Billie had designed.

Lesson 2 – The person who pitched doesn’t own the idea any more, give up your plan and let the idea evolve

Our second pitch didn’t go very well because we had run out of energy and enthusiasm. The judges called us out on it and we all felt pretty miserable coming out. We left at 11pm absolutely drained, physically and emotionally.

There were some wonderfully positive outcomes from Saturday. I learned about the fertility problem in general and I learned I could work well with Billie.

Sunday

I started the last day feeling a bit apprehensive about the whole weekend but Steffi had really opened up to changing the plan and the rest of the team were keen too.

We went back to basics and started an ideation session in one of the lecture theaters. I had some ideas but mostly I just kept firm about the not so good ideas that were being bandied about. We came up with a range of ideas but didn’t settle on one for the morning pitch. It was really hard tearing up everyone’s ideas but I’m glad I did to get us where we eventually got to.

Andy gave a great pitch on what we had done for the whole morning and where we were going with ideas. The mentors saw that our energy was back and helped us out by giving us the Logic Model.

We used this model and realized we would have a better impact by focusing on young people. We argued back and forth about what to actually build while hanging out around a table outside. There were sperm iOS games and egg tamogotchi ideas. After lots of thinking we thought of a Movember clone – a month of no sex and named it Sexless September.

I thought this was an awesome idea and felt I had been negative enough all morning so after one final argument with Steffi to not have too much information on a Movember type site I checked out and got some lunch.

The team was buzzing again and I quickly built a site that Billie had designed in square space. I was surprised how awesome and easy squarespace is.

Sexless September site design

Lesson 3 – Persevere, It will come together at the end

We pulled it together by the end. We announced the new site on our Facebook pages and got 60 or so signups to our email list by the end of the weekend.

Steffi and Andy gave an incredible pitch. It started off quite somber and then Andy brought in the humorous side. What really struck me was the audience was our intended target audience and they LOVED it. Billie’s tee shirt idea went down a treat.

Giving the final pitch
Giving the final pitch

So overall the weekend was like a roller-coaster. It started high with a surprise not for profit project, dropped very low on Saturday, climbed Sunday morning with our ability to change output, dropped off again for me having to be negative to focus the output and finally finished on a super high because of where we got to.

I would highly recommend going to a startup weekend. It opened my eyes to many more of the factors involved in building a product. The mentors were awesome. Our team was awesome and the problem we had to solve was awesome. It was organised very well. It was a pressure cooker but I came out better for having gone.