Machine learning seems to get all the interest and leap these days, and some even say it goes mainstream. ML has even been dedicated to conferences and summits such as 2021 AWS Machine Learning Meeting. For ML to move into the mainstream, from my point of view, there are still real-world lessons that we need to translate into ML production for companies, and I hoped to get some guarantees from this summit. I listed here some of the parts that affected me the most. We hope you find these useful when you plan to use ML:
- Pizza with the right amount of cheese, for example with Swami Sivasubramanian
- Computational humor conversation with Yoelle Maarek
- Dirty hands on conversation with Andrew Ng
Zince the discussion is hosted by AWS, the summit is leaning towards the use of their ML services, but takeaway applications can also be applied to other cloud service platforms, such as GCP, that offer their own ML services.
A possible indication that ML is moving towards the mainstream is its use in all kinds of businesses, including the pizza business. Swami Sivasubramanian was introduced Dafgårds as an example of a business that can apply ML even with limited training material. They wanted to use ML to improve quality and efficiency by placing the right amount of cheese in each pizza dough. Yes, what a nice way to start with things I like: ML and pizza.
They used visual quality control a few shots of learning who can learn a particular task with just a few examples. I think this is an appropriate use for an ML application with a limited set of training data. Gathering data – getting the right kind and amount – is usually a painful point when starting ML projects. Going further, I was interested AWS visual observation it was addressed from customer feedback along with the learning idea of these few shots.
There were other companies and companies you’ve probably heard of, like BMW and the New York Times, but the pizza idea got stuck.
For ML to be mainstream and stay in place, it is also important to look at User Experiences. And because of a rather interesting twist, AWS takes a serious look at the computational humor spent hours in this study. I liked their approach to embedding the ML team into the product team. In this example, a computational humor team has worked alongside the Amazon Alexa Shopping team.
Looking for ways to delight the customer, I can’t just think of the recent refreshing progress with artificial intelligence chat robots. Discussions with inanimate objects such as Pluto and paper planes in Google’s LaMDA in Google ME!
To get back into the ranks of computational humor, the team not only made AI fun, but the team wanted to know if the customer was funny and how the AI should react. In this case, the customer takes the initiative.
For detection and possible response, the team even tried to look at cultural references, sarcasm, or whether the client is in the mood to be playful. I like how the speech also included references to mixed-initiative theory, facilitation theory, and a few others.
For example, for the Nintendo Switch Gray Joy-Con, one customer responded and referred to some cultural references: “Can I use this to hack the matrix and save humanity?”
Another customer responded, presenting sarcasm with very expensive water coolers: “Will this thing fly me? It seems that the price needs to do something special. “
Yes, the current ML mode can already do so much, but there is still a lot to explore and new ways to use it.
And then, there’s a fire chat with Andrew Ng and Swami Sivasubramanian. There were quite a few interesting points in the less than 30 minute conversation, but the highlight for me was actually the first minutes!
The first part provided notes to companies and talked to their executives, such as CEOs and technology leaders, who want to start learning machine. When you jump towards ML for the first time, you have to get your hands dirty.
Too long to plan and start is probably a mistake. The data for the first project can be messy, but there can be workarounds, and the company should be able to start with small pilot projects or proof of concepts and make a quick profit. Learn from them to grow and expand your project.
And I liked the idea of trying out a pilot project before devising a long-term strategy for machine learning. POC’s studies help create a strategy that better fits the company’s goals and human culture.
I mentioned in a previous article Confirmation applied to business problems that the biggest hurdle of RL is in formatting the problem and the next one would formulate the next formatted problem. So you should be careful when formulating the problem. But even that said, it shouldn’t stop people from getting dirty and still give it a try. In the long run, there could probably be an ML platform, but always the first POCs and quick wins give insights into the direction of the ML journey.
And now, even for non-managerial positions, when you start an ML project, it can be difficult for an entirely new application to create target metrics to check if the ML is successful or not before the ML team has certified or if there are relevant projects in the literature which help define basic performance. The most important thing here is probably a fast and dirty prototype system, repeat and learn from these experiences.
More and more companies are adopting ML, but growth will not slow down soon. It may have started with technology companies, but ML projects are now penetrating through other companies, including pizza companies. A key part and recurring theme of the summit has been the transformation of ML into production.
Dealing with pain points, such as getting a large set of exercise data, is crucial, so getting techniques like learning a few shots is a welcome change. And for these ML projects to be successful and not just a hype, customer experience should be a priority. Investing in research that addresses customer experiences, such as those with computational humor, is not just a side project but an integral part of an ML product. And it’s just refreshing to see the occasionally different perspectives in the world of ML. Considering ways of working, such as embedding the ML team in a product or development team, can be key to the success of a project.
Turning ML into production would also mean going beyond the theoretical, and especially for companies embarking on their ML journey, getting their hands dirty with small pilot projects and where these quick profits can help give direction to the company’s longer strategy in the company.