Ensemble AI Inc. is looking to tackle headaches around data quality and help companies build more powerful artificial intelligence models after closing on a $3.3 million seed funding round.
Today’s round was led by Salesforce Ventures, with Amplo, M13 and Motivate also participating. They’re backing Ensemble because the startup has created a pioneering approach to data representation in order to enhance the performance of AI models, without pumping them with vast amounts of extra data or creating more complicated model architectures.
What the startup is doing is using machine learning techniques to enhance AI models, by helping them uncover hidden relationships between their datasets. The company explains that if AI is going to be able to solve real-world problems, it needs access to more and better-quality data. Many companies struggle with limited and sparse or one-dimensional datasets, and that prevents their AI models from generating meaningful or useful results.
Data scientists spend hours trying to fix their data to overcome this, and some progress has been made with more sophisticated AI model architectures, but such endeavors require vast resources and technical expertise that not every company has.
To solve these issues, Ensemble has created a novel embedding model it calls Dark Matter, which uses an “objective function” to create richer representations of data for predictive tasks. Dark Matter, the company says, can understand the complex, nonlinear relationships within datasets through a lightweight data transformation. It distills the complexity of these relationships into a simple “data representation,” so engineers can build better quality AI models that can tackle much harder problems.
Ensemble co-founder and Chief Executive Alex Reneau explained that Dark Matter slots in between the feature engineering and model training and inference processes within data pipelines.
“We’re able to enable customers to maximize their own data that they’re working with, even when it’s limited, sparse or highly complex, allowing them to train effective models with less comprehensive information,” he said. “This foundational technology frees up data scientists to focus on experimentation and also makes ML viable for problems previously unable to be modeled, unlocking new capabilities for our customers.”
The startup believes Dark Matter is a superior solution to synthetic data, which is often used by AI developers to compensate for low-quality or sparse datasets. It explains that though Dark Matter does create new variables, the mechanics are fundamentally different.
Because synthetic data recreates existing distributions from Gaussian noise, it means that no new information is actually created. The synthetic data merely mirrors the statistical properties of the existing data, so there’s no meaningful impact on predictive accuracy, the company explained.
On the other hand, Dark Matter learns how to create new embeddings with fundamentally different statistical properties and distributions that result in measurable improved predictive accuracy.
Salesforce Ventures’ Caroline Fiegel told VentureBeat that Ensemble offers a promising solution that can potentially accelerate the adoption of AI. She explained that many organizations are struggling to deploy AI models in production given issues with poor data quality and the potential use of personally identifiable information.
“When you peel that back and really start to understand why, it’s because the data is disparate. It’s kind of low-quality,” she said. “It’s riddled with PII.”
Ensemble says Dark Matter has already been put to use by a number of early adopters in areas such as biotechnology, healthcare, personalization and advertising technology, with promising results. For instance, one biotech customer has used its tech to create a model that’s better able to predict virus-host interactions within the gut microbiome, it said.
Looking forward, Ensemble said it will use the funds from today’s round to expand its team and accelerate its product development and go-to-market plan.