![](https://d15shllkswkct0.cloudfront.net/wp-content/blogs.dir/1/files/2020/11/machine-learning-banner-artificial-intelligence_107791-611.jpg)
![](https://d15shllkswkct0.cloudfront.net/wp-content/blogs.dir/1/files/2020/11/machine-learning-banner-artificial-intelligence_107791-611.jpg)
An artificial intelligence startup called Brainome Inc. is launching today, hoping to change the way organizations approach machine learning with a new concept it calls “measure-before-build.”
The idea is pretty simple. What Brainome’s Daimensions tool does is measure the information content in the data that’s used to train machine learning models against the target type, before the models are built and trained.
The company says this approach is far more efficient and cost-effective than traditional data preparation methods. Typically, most practitioners will use large data sets and throw massive amounts of computing power at them, but that results in extremely complex and opaque models, Brainome says.
Brainome co-founder and Chief Executive Bertrand Irissou argues that this is the wrong approach, saying that designing a new machine learning model is really no different to starting a new engineering or science project.
“Before building a car, plane, bridge, or computer chip, you must measure before you design and build,” Irrisou said. “Today’s data scientists and machine learning experts are forced to rely on what is essentially guesswork instead of having access to any advanced type of measurement. Brainome takes a completely new angle by providing much-needed tools based on a novel, systematic measurement-based approach.”
Daimensions enables teams to know if their training set has enough data to learn rules and avoid overfitting. That’s when a model makes predictions that are too closely fixed to what was in the original training data, and thus won’t generalize well when new data is processed.
It also helps teams to find and design the data features that matter for each specific model they’re trying to build and iterate data preparation and model design more quickly without training, Brainome says. In turn, this makes it faster and easier to train and execute, using more compact and efficient models.
By introducing “measurement” as a structured discipline to machine learning, Brainome says companies will be better able to predict the speed of their projects, the associated costs and the likelihood of success.
“It was only a matter of time before someone came along and offered metrics that show if an AI or machine learning endeavor is likely to succeed,” said Constellation Research Inc. analyst Holger Mueller.”We now see the beginning of this new software category with an offering from Brainome. As with all brand-new categories, some caution is advised as early versions of software need to prove themselves first.”
Brainome says its tool has already been put to use by early adopters in fields such as financial, health and advertising technologies.
“Brainome has been a breath of fresh air in helping us model a problem in the healthcare domain,” said Eric Davis, vice president of AI Language Tech Labs at SK Telecom. “Our previous approach was time-consuming, full of guesswork, and took over a week to iterate from feature extraction to experimentation to results. Brainome took a lot of the guesswork out of data quantity needs and feature importance, allowing us to reduce our experimentation cycle from a week to mere hours.”
THANK YOU