Numenta develops the Numenta Platform for Intelligent Computing (NuPIC), which accelerates the performance of large AI models on commodity CPUs while significantly reducing power consumption. Their Thousand Brains Project focuses on creating a new type of artificial intelligence based on neocortical principles, enabling more efficient and scalable AI applications.
Funding
Funding not disclosed
Founders
Product
Problem
Large AI models, particularly Transformer networks like BERT and GPT, demand significant computational resources, leading to high infrastructure costs and energy consumption, especially when deployed for real-time applications or long sequence tasks. Existing solutions often require expensive GPUs, creating barriers to entry and hindering scalability for many organizations.
Solution
Numenta provides an AI platform technology, NuPIC, that accelerates the performance of large AI models on commodity CPUs, significantly reducing power consumption and infrastructure costs. By applying neuroscience-based techniques and optimizing for CPU architectures, NuPIC enables efficient and scalable deployment of Transformer models for various applications, including natural language processing and AI-powered games. The platform supports both proprietary and pre-trained models, allowing developers to leverage existing CPU infrastructure and minimize total cost of ownership while maintaining data privacy and security.
Target Audience
Numenta targets AI developers, game studios, and enterprises seeking cost-effective and scalable solutions for deploying large AI models on existing CPU infrastructure, particularly for real-time NLP applications and long sequence tasks.
Features
- Optimized inference server for running large language models (LLMs) on CPUs
- Support for Intel Advanced Matrix Extensions (AMX) for enhanced performance
- Integration with Intel's OpenVINO toolkit
- Model library with a selection of models suited for natural language processing tasks
- Training module for fine-tuning models to specific requirements
- Ability to deploy on-premise or via cloud in CPU-based infrastructures
- Compatibility with BERT and GPT models