AI/ML has now come of age and proliferating at an unprecedented pace across Industries. The convergence of low-cost computing capability (thanks to powerful chips and cloud), ubiquitous network connectivity, and low cost of technology have propelled AI/ML into every aspect of our daily lives. From powering our smartphone decision engines to voice assistants to Autonomous cars to even simple automation for productivity apps, algorithms today power our daily schedules in ways we cannot even fathom. A simple scroll on Facebook/Instagram or even on your phone is powered by an algorithm, offering you the right content of your interest based on your past behaviors.
This is a great time for software developers to build algorithms and self-learning models that can drive significant impact and showcase their competency. The only challenge is AI/ML implementations remain a complex domain to navigate. Today most organizations need a combination of data scientists, Developers, and infrastructure engineers to bring the real power of AI/ML to life. These implementations need to converge statistical, data analytical knowledge with software development and eventually the capability to deploy and manage models on the cloud infrastructure. Hardly a task for a single individual. Hence in most corporates, AI/ML deployments are moving at a sluggish pace given the challenges of bringing multidisciplinary teams to move at high velocity. But wait there is an alternative for developers to drive the full power of AI/ML on their own.
What is Auto/ML?
Automated machine learning, also referred to as automated ML or AutoML, is the process of automating the time-consuming, iterative tasks of machine learning model development. It allows developers to build ML models with high scale, efficiency, and productivity all while sustaining model quality.
How is it different from the traditional approaches?
AutoML takes away the development and deployment-intensive tasks from the humans and provides automated tools and platforms for most use cases.
Even business users or data scientists who are not developers can utilize some of these platforms.
Time to market is rapid as most of the iterative tasks of model development, feature engineering, model training are all automated.
Today platforms are also offering end-to-end managed services.
Best illustration is this write-up written by a 17-year-old Alexander Mamaev on how to use AutoML from scratch!
Why should you use AutoML?
Most AutoML platforms take a low code approach with a Web GUI to do most of the model prep and parametrization.
AutoML platforms and projects significantly enhance time to market for AI/ML deployments. This is one of the biggest pain points for most engineers and corporates.
The platform approach gives you scalability options on demand, akin to cloud, that makes it easy for you to focus on the outcomes rather than the engineering complexity of development and deployment issues.
It is a great way to start your journey to demonstrate early wins and learn at low cost rather than do a big program with high risks
Provides velocity and flexibility reducing team dynamics across functions
Variable cost structure with the ability to scale up and down and pay for use.
How do I get started?
The best place to start learning as a beginner would be here. Although this site directs you to packages I would encourage you to look at the platform-centric approaches first that I have enumerated below.
Great news is that there are now a plethora of AutoML platforms you can start playing around with. Few are listed below:
1. Google AutoML: Freemium model. You can use the cloud AutoML offering with some pre-built AI models and then scale up with a paid version. Google is leveraging the power of its vast AI engines behind the scenes. Worthwhile if the use cases are what you need to get started.
2. If Google is offering, Microsoft and Amazon cannot be far behind with their wares, can they? Both offerings are some versions of the Freemium models but the paid versions are what they will push you towards if you want to scale up in any measure.
3. BigML: Free for workloads up to 16MB per dataset. Yes, it feels like a small but great place to start if you are a developer just learning your way into AI/ML. Feature rich platform that offers quick access, models that can be understood, exported, collaborations, automation, flexible deployments, and more.
4. H2O.ai: Opensource platform with a very large community that continues to enhance the models and use cases. Great place to get your hands dirty, partner with other developers from around the world and grow your expertise. H2O is one of the most mature and leveraged platforms with use cases across most industries. So if you get serious about scaling up your deployments this is the platform to go to. Remember that you will have to be your support system at scale.
In addition to the platform plays there are many AutoML projects and some of them that you can access on Github are:
1. Auto-Sklearn: Mechanized ML programming package based on Scikit-learn.
2. MLBox: Robust Python package
3. Pycaret: Open-source low code Python Package machine learning library.
4. AutoKeras: Simplistic approach to learning models and applications. Built on the Keras platform.
These are just some of the platforms and packages/libraries in a growing list. More and more players are launching AutoML offerings s they realize the potential of the market. There is no clear leader. In addition, there are a number of open source projects mushrooming to address these needs.
Things to keep in mind:
1. If you are worried if your role is at risk, chill!. Data researchers and Engineers will continue to play a pivotal role in the AI/ML journey. AutoML is just an enabler approach.
2. Be thoughtful of the choice of the platform/package you choose. As you start to scale you will have to live with the features or lack thereof given the investments you would have made.
3. Leverage the use cases and platform features to the maximum. Engage and leverage the vendor support team or the community. There is significant body of knowledge and experience available so don’t try to do it yourself. Avoid the mistakes the early adopters have made.
4. Do! Jump in and get started. We as engineers love to debate the feature sets or the endless debates on which product is better. The key is to take an agile way of working. Do, try, finetune and repeat.
Hope this note will get you started on your AutoML journey and provide a faster path to learning, developing, and making an impact. Go for it, give it a try! Good luck.