Kumo uses Snowpark Container Services to enable deep learning in the Snowflake Data Cloud

Announcing its integration of deep learning capabilities into the Snowflake Data Cloud through Snowpark Container Services, Kumo, a deep learning platform for relational data, made the announcement today at Snowflake Summit 2023.

The feature of Snowpark, which enables developers to write code in their choice language and execute that code directly on Snowflake, has recently been expanded by Snowflake\’s Snowpark Container Services. With this Container Services release, businesses may use their Snowflake accounts to execute full-stack apps and third-party software.

With this connectivity, users, according to Snowflake, can make the most of their data by using cutting-edge capabilities, ensuring data security, and doing away with the requirement for data transportation.

Additionally, GPU support is a feature of Snowpark Container Services, which enables data science and machine learning teams to expedite development and close the gap between model deployment and constant data protection and governance across the AI/ML lifecycle.
One of Snowpark Container Services\’ early users, Kumo uses the technology to roll out cutting-edge neural networks for businesses.

With the help of Kumo\’s predictive AI platform, which uses graph neural network (GNN) technology, developers, data scientists, analysts, and company leaders can produce forecasts that are very accurate.

AI and graph neural networks

In the case of traditional machine learning, data must first be extracted from a data lake or warehouse before human feature generation and tweaking. The new connection, which is now accessible in private preview, enables collaborative users to develop predictions, act directly on raw Snowflake data, and save the outcomes as extra Snowflake tables.

According to Vanja Josifovski, co-founder and CEO of Kumo, \”the new integration will run Kumo\’s AI services directly on relational tables over the cloud without the intermediate steps found in traditional machine learning, such as training set generation and feature engineering, by using graph neural network technology.\”

Josifovski emphasised that users don\’t need to export data from their Snowflake environment in order to develop and run a query that delivers predictions, simulating the process of querying historical data for analysis.

Following a recent partnership between Nvidia and Snowflake, the news enables companies to tailor their generative AI models over the cloud to meet their unique organisational needs.

The connection eliminates the need for companies to move data outside of Snowflake\’s Data Cloud environment in order to create generative AI applications utilising their own data.

Enabling cloud-based deep learning for predictive analytics

In order to make graph learning predictions on their company data, users will be able to directly use Kumo\’s predictive AI solution inside Snowflake thanks to Snowpark Container Services, according to Kumo\’s Josifovski.

(Visit us at VB Transform, our networking event for corporate IT decision makers, on July 11 and 12 in San Francisco to hear more about data and AI in the industry.)

\”Where the ML processing runs has long been a puzzle in machine learning and data warehousing. Our businesses enable customers to broaden the usage of machine learning and predictions to everyone who has access to the Data Cloud by altering the paradigm to conduct the ML processing in the Snowflake Data Cloud, Josifovski told VentureBeat. This is carried out in accordance with a single security programme, which is more simpler than doing it in accordance with several security programmes.

Calculations using linear algebra, which are quite compatible with GPU computing, represent a significant part of modern AI systems. Previously, Kumo had to take the data out of the customer\’s account and process it elsewhere in order to use GPUs. With this connection, all data processing—including GPU processing—takes place immediately within the customer\’s Snowflake account.

\”The approach of not needing a training set and feature engineering shortens the AI/ML lifecycle significantly,\” he said. \”We want to free data scientists from tedious and repetitive tasks so they can concentrate on higher-level tasks like choosing the best predictive task, analysing the results, and figuring out how to get the most business value out of the predictions,\” the company claims.

Through this service, the business unveiled a standout feature: deep learning-driven relational data GNNs.

The graph and its associated properties, which are specified by non-key columns of the data, may be learned from by these deep learning-driven GNNs. Once a graph is created, several AI/ML jobs may be trained effectively on it without the need for additional designed characteristics or distinct training sets.

Additionally, Kumo provides an advanced and scalable autoML technique that simplifies the laborious process of hyperparameter tuning.

\”GNNs are incredibly efficient for a variety of predicting tasks, but they are also challenging to build, scale, and optimise. The necessity for graph generation, which requires knowledge of GNNs and the development of optimisation tasks, is eliminated by Kumo\’s AI platform. Kumo has built a predictive query language to define the AI/ML job, according to Josifovski.

Predictive analytics made easier for citizen coders

According to Josifovski, predictive AI/ML now calls for highly qualified individuals with specialised knowledge. The lifecycle includes feature testing, which calls for a sizable infrastructure to enable training and inference (scoring).

He clarified that the new integration\’s goal is to provide consumers with a simplified process regardless of their level of data science expertise.

Then, they can quickly use predictive graph learning in a variety of commercial contexts, including customer acquisition, loyalty, retention, personalisation, and fraud detection. His business claims that it just takes a few hours to do a whole AI-based investigation.

Kumo, according to Josifovski, \”provides control of the training and inference for skilled data scientists, while allowing users to run queries over the relational data without requiring a deep understanding of AI/ML concepts.\” In a manner similar to how data warehouses are already utilised for analytics, the platform enables a broad variety of users to access it.

Kumo also emphasised that the product\’s natural connection with Snowflake makes it easier to set up and use without the need for security and legal privacy evaluations. By doing so, obstacles are lowered and value realisation is accelerated noticeably.

The business is sure that this will quicken trial and implementation of specific predictions, enabling and enhancing procedures like client acquisition, customization, entity resolution, and other predictive jobs.

Josifovski told VentureBeat that \”in enterprises, many teams issue SQL queries over a data warehouse to obtain analytics that professionals then consume to chart future actions.\” Kumo \”will enable users to automatically obtain actionable predictions without requiring professional interpretation.\”

Leave a comment