Based in California ActiveloopA startup that provides a dedicated database to streamline AI projects announced today that it has raised $11 million in Series A funding from Streamlined Ventures, Y Combinator, Samsung Next (the startup accelerator arm of Samsung Group) and several other other investors.
While there are multiple data platforms, Activeloop, founded by Princeton dropout Davit Buniatyan, has carved out a distinct segment with a system to deal with considered one of the most important challenges facing corporations today: leveraging unstructured multimodal data to coach AI -models. The company claims that this technology, called “Deep Lake,” enables teams to construct AI applications at a price as much as 75% lower than market offerings, while increasing engineering team productivity by as much as five times.
The work is very important as more corporations look for methods to leverage their complex data sets for AI applications aimed toward different use cases. According to McKinsey research, generative AI has the potential to generate $2.6 trillion to $4.4 trillion in global business profits annually, with significant impact across dozens of areas, including providing support interactions with customers creative content for marketing and sales and the creation of software code based on natural language prompts.
What does Activeloop Deep Lake help with?
Today, training high-performance foundational AI models requires coping with petabytes of unstructured data, spanning modalities akin to text, audio, and video. The task typically requires teams to discover relevant data sets from disorganized silos and repeatedly leverage them with different storage and retrieval technologies – something that requires a whole lot of standard programming and integration from engineers and may increase the associated fee of the project.
Activeloop targets this inconsistent approach with the standardization of Deep Lake, which stores complex data – akin to images, videos, and annotations, amongst others – in the shape of machine learning (ML)-native mathematical representations (tensors) and facilitates streaming these tensors into the SQL-like Tensor Query Language, a browser-internal visualization engine or deep learning frameworks akin to PyTorch and TensorFlow.
This gives developers a platform for every thing from filtering and searching multimodal data, to tracking and comparing their versions over time, to streaming for training models targeting different use cases.
In a conversation with VentureBeat, Buniatyan says that Deep Lake has all the advantages of a vanilla data lake (akin to ingesting multimodal data from silos), but excels in converting every thing into tensor format, which deep learning algorithms use expect as input.
The tensors are stored neatly in cloud-based object storage or local storage like AWS S3 after which seamlessly streamed from the cloud to graphics processing units (GPUs) for training – passing barely enough data for computation in order that they will be fully utilized. Previous approaches that handled large data sets required copying the information in batches, leaving GPUs idle.
Buniatyan said he began working on Activeloop and this technology in 2018 when he was faced with the challenge of storing and preprocessing hundreds of high-resolution brain scans of mice on the Princeton Neuroscience Lab. Since then, the corporate has developed core database functionality in two important categories: open source and proprietary.
“The open source aspect includes, amongst other things, the record format, version control and a wide selection of APIs designed for streaming and querying. On the opposite hand, the proprietary segment includes advanced visualization tools, knowledge retrieval and a robust streaming engine, which together enhance the general functionality and appeal of their product,” he told VentureBeat.
While the CEO didn’t share the precise number of shoppers Activeloop works with, he noted that the open source project has been downloaded greater than one million times up to now, propelling the corporate's presence within the enterprise segment. Currently, the enterprise-focused offering has a usage-based pricing model and is utilized by Fortune 500 corporations in highly regulated industries akin to biopharma, life sciences, medical device, automotive and legal.
A customer, Bayer Radiologyused Deep Lake to unify multiple data modalities right into a single storage solution, optimize data preprocessing time, and enable a brand new “Chat with X-Rays” feature that permits data scientists to question scans in natural language.
“Activeloop’s knowledge retrieval feature is optimized to assist data teams construct solutions at as much as 75% lower cost than anything available on the market, while significantly increasing retrieval accuracy, which is very important within the industries Activeloop serves.” , added the founder.
Plan to grow
With this round of funding, Activeloop plans to expand its enterprise offering and attract more customers to its AI database in order that they can easily organize complex unstructured data and access knowledge.
The company also plans to make use of the funds to expand its engineering team.
“A key development within the pipeline is an upcoming release of Deep Lake v4 with – faster concurrent I/O, the fastest streaming data loader for training models, fully reproducible data lineage and external data source integrations,” Buniatyan noted, claiming that there are various customers on this area, but “no direct competitors”.
Ultimately, he hopes the technology will save corporations from spending tens of millions on in-house data organization and retrieval solutions and keep engineers from doing a whole lot of manual work and standard programming, making them more productive.
