Perpetual ML

Perpetual ML is a 100x faster, scalable, end-to-end, all-in-one ML Suite designed for modern data warehouses. It empowers businesses to unlock the best insights and actions from their data in minutes, not days. The suite is built as a low-code / no-code native app and emphasizes speed, explainability, and portability across data warehouses.

Overview

Perpetual ML provides a cohesive platform that accelerates model training, enables continual learning, and offers robust monitoring and decision-making capabilities. Key capabilities include fast initial training, continual learning, confident predictions, geographic-aware modeling, and model monitoring, all while supporting a range of ML tasks from tabular to text classification and ranking.

Key Features

• 100x faster initial training thanks to a built-in generalization algorithm that minimizes hyperparameter optimization
• Continual learning to keep models up to date by resuming from where you left off, without restarting from scratch
• Conformal Prediction for better confidence intervals and more reliable decisions
• Geography-aware learning for natural decision boundaries in geographic data
• Model monitoring to detect distribution shift and maintain performance without needing external tools
• Wide ML task support: tabular classification, regression, time series, learning-to-rank, and text classification using embeddings
• Portability: not locked to a single vendor; currently developed for Snowflake with plans for Databricks and other warehouses
• Effortless parallelism for superior computational performance and resource efficiency
• No need for specialized hardware (no GPU/TPU required); leverage existing hardware
• Free-form access to try a fast, end-to-end ML experience with a no-hassle setup

How It Works

1. Deploy within your modern data warehouse environment (Snowflake now; other warehouses planned).
2. Utilize built-in generalization for rapid initial training and automatic parameter tuning improvements.
3. Leverage continual learning to keep models current as new data arrives.
4. Monitor models for distribution shifts and confidence in predictions using integrated evaluation metrics.
5. Scale across tasks and data types with a portable, low-code/no-code interface.

How to Get Started

• Contact for a free trial to see Perpetual ML in action.
• Explore features, pricing, and blog resources on the Perpetual ML site.

Safety and Usage Considerations

• Designed for enterprise data workflows with a focus on speed, reliability, and governance across data warehouses.

Core Features

• 100x faster initial training via built-in generalization algorithm (reduces hyperparameter tuning)
• Continual learning to resume training without starting over
• Conformal Prediction for improved confidence intervals
• Geography-aware modeling for natural decision boundaries in geographic data
• Integrated model monitoring for distribution shift detection
• Broad ML task support: tabular classification/regression, time series, ranking, text classification using embeddings
• Portable architecture, currently Snowflake-ready with future Databricks and other warehouses support
• Effortless parallelism for higher performance and resource efficiency
• No specialized hardware required; runs on existing hardware
• End-to-end ML suite with low-code / no-code interface