This California-based AI Startup, Predibase, is Taking Declarative Machine Learning to the Next Level with a Low Coding Base (an Alternative to AutoML)

While it is anticipated that $35.8 billion has been spent globally on AI technologies in 2019, approximately 80% of businesses have seen their AI initiatives stagnate because of poor data quality and lack of trust in AI systems, according to an Alegion analysis. Forty-four percent of “data executives” in the U.S.-based organizations who responded to a recent study claimed that their positions weren’t clearly defined and that they weren’t hired enough or were too segregated to work effectively. Disorganization within the organizations hinders data science teams’ abilities to deliver timely AI and analytics initiatives. The two issues that respondents stated most worried most were the impact of a revenue loss or damage to brand reputation brought on by malfunctioning AI systems and a propensity toward flashy expenditures with quick returns. In the end, they are issues with organizations. However, according to Piero Molino, co-founder of the AI development platform Predibase, insufficient tools frequently make them worse. 

Predibase is based on open source technologies such as Ludwig, a collection of machine learning tools, and Horovod, a framework for training AI models. Both were first created by Uber, which handed over project control to the Linux Foundation several years ago. Predibase’s AI pipeline definition capabilities claim to scale petabytes of data across thousands of machines with just a few lines of code. According to the organization, utilizing the platform, a user can develop a text-analyzing AI system that specifies the input and output data in just six lines of code. Predibase enables them to include options in the configuration file that provide a more granular degree of control if they wish to iterate and modify that system. For model training, Predibase interacts with data sources like Snowflake, Google BigQuery, and Amazon S3. Depending on the use case, users can train models via the platform or programmatically and then host, serve, or deploy those models into regional production settings.

'Declarative machine learning systems provide the best of flexibility and simplicity to enable the fastest-way to operationalize state-of-the-art models. Users focus on specifying the “what”, and the system figures out the “how”.'

Predibase claims itself to be an example of how innovation is being used to streamline the ML life cycle. Predibase aims to help data science teams define the inputs and outputs they want for their ML model. In other words, they produce configuration files that Predibase then determines how to use. Data science teams can still make as many modular adjustments as they’d want to accommodate new or changing consumer requirements. Predibase’s proposal, in summary, is to reduce the complexity of the ML life cycle, which is the main obstacle to the success of data science projects.

It’s simple to iterate and enhance models because everything is accessible via a configuration parameter. One may continue to have complete control and flexibility, which is one of the benefits. Therefore, one may configure every aspect of the models through setup, including selecting from various model topologies, training settings, and data preparation options. With merely a new configuration, make modifications. It can also be expanded. 

Recently, the company raised $16.25 million in the Series A funding round. Predibase’s beta product, which is now invite-only, will be made available to a larger market using the extra funding from the Series A. Additionally, it will be used to expand Predibase’s 21-person team by adding machine learning engineers and developing a go-to-market organization. 

References:

• https://predibase.com/
• https://www.crunchbase.com/organization/predibase
• https://thenewstack.io/predibase-takes-declarative-approach-to-automl/?utm_content=buffer28048&utm_medium=social&utm_source=twitter.com&utm_campaign=buffer
• https://techcrunch.com/2022/05/10/predibase-exits-stealth-with-a-low-code-platform-for-building-ai-models/