The next wave of innovation is making the power of machine learning available to everyone. AutoML tools can be tailored to your business needs to identify errors while it increases accuracy and turnaround time to production-ready models.
In 2018, Google launched Google Cloud automated machine learning (AutoML), a service that helps businesses tap into its machine learning (ML) infrastructure, albeit built with the data its users generated, to train and build their own AI models.
Since the traditional ML process is tedious, human-dependent, and repetitive, AutoML solves the real-world challenges by automating Artificial Intelligence (AI)-based ML. It uses statistical techniques or algorithms to enable a computer to become better at what it does, running the entire ML gamut end-to-end, from raw dataset to deployable ML model.
It is a tool that accelerates the process in different ways to create greater efficiency — regardless of whether the user is new to data science or a data scientist with years of experience.
To announce Google’s AutoML, Google CEO Sundar Pichai wrote, “Designing neural nets is extremely time-intensive and requires an expertise that limits its use to a smaller community of scientists and engineers. That’s why we’ve created an approach called AutoML, showing that it’s possible for neural nets to design neural nets.”
Google’s researchers have taught ML software to build ML software. In some instances, what it comes up with is more powerful than the best systems the researchers themselves could design. But the expertise needed to build cutting-edge AI systems is in scarce supply — even at Google.
“Today, these are handcrafted by machine learning scientists, and literally only a few thousands of scientists around the world can do this,” said Pichai, briefly name checking AutoML, at a launch event for smartphones in 2017. “We want to enable hundreds of thousands of developers to be able to do it.”
Initially, researchers developing AutoML focused on automating the model selection and model learning steps of the ML workflow. But these solutions have matured now, offering improved scalability, flexibility, versatility, and transparency. AutoML algorithms work best when direct measurement of model accuracy such as model fit error, or silhouette distance, and so on are available.
A growing number of researchers are working on AutoML, and it is making its way into enterprises, too. If AI-made AI becomes practical, ML could spread outside of the tech industry, such as healthcare and finance, much faster.
“AutoML drastically boosts productivity in building complex machine learning models in healthcare. Only 0.5 per cent of the global data are ever analysed; in healthcare, this implies missed opportunities to save lives and learn new medicine,” says Ioannis Tsamardinos, CEO and co-founder at JADBio.
“It protects against methodological errors during the analyses to reduce wrong medical decisions. It constantly improves in quality and often surpasses the models built by human experts. It empowers life scientists to have direct access to the knowledge in their data without the data scientists mediating. It is used to discover new knowledge in the form of predictive factors that provide insight into the underlying medicine or biology,” adds Tsamardinos.
Interestingly, besides beating humans at games, Google’s DeepMind helps reduce the energy used by Google’s massive data centres. It even sped up Google’s ability to map new cities. AutoML is changing the way businesses approach ML problems due to its many benefits.
Also Read: Implementing Machine Learning for Data Analysis
Time-saving, reduce errors
Data engineers and scientists manually test models, tune hyperparameters, and evaluate models to arrive at the best model for a particular problem. However, the implementation process is prone to human-made errors and bias. AutoML tools can automate this process and run a broader set of ML algorithms to select the best one, which data scientists might not consider before.
It can transfer data to the training algorithm and search for the best neural network architecture for a problem in a few minutes instead of hours if done manually. It also reduces the possibility of inaccuracies that arise mainly due to manual steps, helping businesses achieve a higher degree of accuracy and achieve higher ROI on ML projects.
Facebook trains around 300,000 ML models to improve its ML processes and created its AutoML engineer, named Asimo, to automatically generate improved versions of existing models.
Democratising, reduce the skill gap
It’s rapidly democratising ML tools by making it accessible for all users, enabling ML engineers and data scientists and non-technical users. A survey of about 80 data scientists conducted by CrowdFlower, found data scientists spend only 20 per cent of their time on actual data analysis and 80 per cent of their time finding, cleaning, and reorganising huge amounts of data.
According to a report by McKinsey, autoML will have an impact on hiring patterns. For instance, AutoML practitioners, like biochemists in drug research, will be able to perform simpler data-science tasks. Given the scarcity of ML experts, AutoML will empower lesser AI-skilled persons to build and deploy AI or ML-based models and monitor its performance too. In short, autoML will help bridge the talent gap with its enormous wide-ranging capabilities. Though it would not replace data scientists, it puts the power of advanced ML directly in the hands of business users.
Also Read: Top 5 Trends In Data And Analytics
Improve scalability, increase productivity
It increases productivity for data scientists, as AutoML enables them to devote more time to business problems rather than iterative modelling tasks. AutoML tables help to deploy state-of-the-art ML models at increased speed and scale. It also simplifies the process of applying ML to real-world problems, reducing the complexity of developing, testing, and deploying ML frameworks, thus boosting productivity. It provides a UI for non-technical users and a complete set of APIs that can be used in automation.
Furthermore, AutoML automates the data storage, identifies leaky spots and misconfigurations to ensure accuracy and precision in the result. Besides, it can open up new opportunities for businesses, which are restricted by time constraints and resources to create ML models.
Rolling out more models with increased accuracy can improve other, less tangible business results as well. For example, models lead to automation which improves employee engagement allowing them to focus on more interesting tasks.
In the last few years, Google Cloud AutoML allows users to tap into Google’s own ML capabilities, build custom models better suited for business needs, and help users perform text analytics and video intelligence, and other tech giants have introduced competing autoML tool kits.
Amazon, the world’s biggest e-tailer that also moonlights as the world’s largest provider of computing services, launched Amazon SageMaker Autopilot in 2019. Users simply need to specify a column in their data set that they want to predict. While the model produces the results after a simple click, model quality can also be improved. The program shows the leaderboard of best-ranked solutions, and users can choose the best one or even modify parameters to favour speed of execution vs accuracy of results.
In 2019, IBM launched the Watson Studio AutoAI product, a graphical tool that walks users through all stages of the ML process, from data preparation to model ranking and selection to hyper parameter tuning and deployment. Powered by the company’s mighty Watson, this solution auto generates code in Python, Scala, Java, or even JavaScript languages to integrate a model with the user’s own custom solution.
Meanwhile, Microsoft has made it easy for users to build ML models without code, at scale, and with speed. Microsoft AutoML is available from within a data visualisation solution.Its Power BI product offers autoML algorithms to perform text analytics and image recognition and custom models utilising other algorithms with or without integration of Azure Machine Learning.
When we look at the interest in AutoML, we observe an increasing trend since the beginning of 2017. As it becomes popular, the AutoML market is expected to reach $14,512 million by 2030, advancing at a CAGR of 43.7 per cent. The interest in autoML will continue to grow. Perhaps not surprisingly, because of its ability to perform data pre-processing, ETL (extract, transform, load) tasks, and transformation, AutoML rose to popularity in 2020.
Besides, AutoML can be used by businesses that use time-series forecasting by analysing data and a series of observed values ordered through time to predict future events. The laborious and complicated process is carried out by data engineers for staffing, network quality analysis, demand at stock-keeping unit (SKU) level, inventory, log analysis for data centre operations, and business operations for sales. AutoML can automate the entire forecasting process, including feature engineering for discovering predictive signals, hyperparameter tuning, model selection, and more.
DataRobot is a commercial solution for AutoML and one of the unicorns in the AI space. Google AutoML Vision also enables organisations to automatically build and deploy advanced classification ML models to derive insight from images.
When it comes to model tuning, every ML algorithm has its own set of hyperparameters that produces the most accurate model. Hyperparameters are set manually, and when you tune a model, you modify the hyperparameters through a trial-and-error process, which could be time-consuming. Arriving at the best hyperparameters for a given problem is a critical but lengthy process. Despite the endless number of hyperparameter combinations, AutoML can automatically find the best set for a given algorithm or model.
Google’s AutoML-Zero uses simple mathematical concepts as building blocks to generate complete ML algorithms. By employing a sophisticated trial-and-error process, AutoML-Zero identifies the best performing algorithm — from a set of 100 of algorithms, each having unique task capabilities — and retains it for future iteration.
Put simply, AutoML automates some or all steps of the ML process.
Data pre-processing: Improving data quality and converting unstructured, raw data to a structured format with methods like data cleaning, data integration, data transformation, and data reduction.
Feature engineering: AutoML can automate this method to create features that are more compatible with ML algorithms by analysing the input data.
Feature extraction: This process includes combining different features or datasets to generate new features that will enable more accurate results and reduce the size of data being processed.
Feature selection: AutoML can automate the task of selecting only useful features for processing.
Algorithm selection and hyperparameter optimisation: AutoML tools can choose optimal hyperparameters and algorithms without human intervention.
As computing becomes more scalable and accessible, the use of AutoML will only become more prevalent and vital in the world of data science and businesses. However, for better adoption, AI and ML-based solutions need to come with an Explainability, as business users work to speed up the ML process to utilise the real power of AI and ML.
