Making Machine Learning Available to All: History of BigML


Machine Learning as a Service (MLaaS) is a big thing in the cloud market. Forbes predicted that the global machine learning market would grow from $7.3 billion to $30.6 billion by 2024. To fuel this growth, data scientists and ML engineers are tasked with developing more models to keep up with the increasingly dynamic business needs of customers and shareholders.

One does not need to have deep knowledge of ML techniques to make the most of what makes BigML stand out in the market.

Founded in 2011 by serial entrepreneur Francisco J. Martin, BigML is an ML service that offers an easy-to-use interface to import data and get predictions from it. With BigML, you can build machine learning and deep learning models without the need for coding.

Speaking exclusively to Analytics India Magazine, Atakan Cetinsoy, Vice President, Predictive Applications at BigML stated, “Our mission was to make ML easy and beautiful for everyone. This is in contrast to the complexity introduced by traditional ML tools, be they open source libraries of commercial tools. Our team’s foresight of the critical role ML can play for businesses is proven, as the last decade has proven.”

A pioneer in the domain

The first version of BigML was released in 2012. The team started with simple decision trees and added many more supervised and unsupervised learning models to the platform with each release.

Decision trees are a supervised learning method used to build a model that predicts the value of a target variable by learning simple decision rules from the data characteristics.

“Back then, ML was primarily an academic research topic and had no contact with the business lexicon. As a result, most sales meetings in the early 2010s started explaining what ML is to executives. BigML literally pioneered the ML-as-a-Service movement,” said Cetinsoy.

The platform’s user-friendly interface (BigML Dashboard) enables ML without code and requires little to no ML knowledge. “If you have background knowledge that certainly helps, but over the years we’ve seen many engineers and business professionals absorb BigML, starting with free training videos and optional certifications to be able to build intelligent applications on top of the BigML platform.” We also provide detailed documentation as we don’t believe in black box ML approaches,” he added.

Built from scratch

BigML rewrote all of its ML algorithms from scratch instead of merging multiple open source packages to provide a smoother and more consistent end-user experience that is very robust. This approach has higher upfront costs for product development.

“Nonetheless, after a decade and over a million lines of source code, we’re in a very good place in terms of our platform maturity and overall end-user experience,” Cetinsoy said.

As a pioneer of the ML-as-a-Service area BigML’s no/low-code approach to ML removes the barriers to entry for a much larger audience than just graduate students and data scientists, which most other tools target. It offers a REST API that works well with programming languages ​​like Python, Ruby, and Java.

BigML offers many binding options for developers to create custom workflows using their favorite programming language to solve a specific predictive use case. As an API-first company, BigML has developed one of the most comprehensive ML APIs, where not only models but also datasets, evaluations and many more platform artifacts are first-class citizens.

There are many tool providers in the field of ML software. In a way, all ML platforms compete with non-consumer companies that have not yet adopted production ML systems. After that come the internal IT teams trying to build their own ML platforms, which is very costly and risky unless you are a tech giant.

The old guard like IBM, SPSS, and SAS have tried to offer ML capabilities in the cloud for the last few years. However, this can cannibalize their traditional statistical software business and was not an overwhelming transformation given the high cost and confusing product portfolio. That leaves VC-funded startups like DataRobot, RapidMiner, and Dataiku that rely on a mix of open-source libraries under the hood. As a result, their APIs lag behind. Some are struggling with the new macro backdrop given their high burn rates.

“While I can’t provide a detailed comparison table, I can summarize that BigML comes out on top due to its smooth learning curve, free tier access that allows anyone to experience the platform, flexible deployments, and low-cost private deployments,” said Cetinsoy.

Focusing on challenges in the current developer ecosystem, said Cetinsoy: “Developers have to make decisions when it comes to building intelligent prediction applications. These range from pre-trained models exposed via APIs that can target a specific use case or data type (e.g. speech recognition) to open source tools and libraries with fairly steep learning curves (e.g. Tensorflow ).”

The latest version of BigML is Object Recognition, which offers an incredibly easy-to-use object recognition feature. Previously, Image Processing was released, which allows BigML users to treat images as just another data type. Both recently released features are extremely helpful to tackle a wide range of computer vision use cases such as: B. medical image analysis, quality control in manufacturing, number plate recognition in transport and person recognition in security surveillance.


Comments are closed.