Noemon is developing a biologically plausible, local learning paradigm that replaces backpropagation, enabling synapse‑level weight updates during inference. This approach lets AI models continuously adapt on‑device, reducing training data and compute costs while lowering memory, bandwidth, and power demands for both edge and cloud deployments.
Funding
Funding not disclosed
Founders
Product
Problem
Current AI systems rely on backpropagation, which drives high training costs, large data requirements, brittle long‑horizon performance, and prevents models from learning continuously during inference.
Solution
Noemon is developing a local, plasticity‑based, distributed learning paradigm that replaces backpropagation. By enabling synapse‑level updates during inference, the approach allows models to adapt from experience, reducing the need for massive retraining cycles. This learning rule also reshapes hardware demands, allowing more efficient memory usage, lower bandwidth, and reduced power consumption for both training and deployment. The result is AI that can scale more economically, operate robustly over long horizons, and be personalized or adapted on‑device without centralized retraining.
Target Audience
Primary customers are AI research labs, enterprise AI teams, and robotics or edge‑device manufacturers seeking scalable, adaptable models with lower training and inference costs.
Features
- Biologically plausible, local learning algorithms that update weights during inference
- Continuous, on‑device adaptation enabling personalized and long‑horizon behavior
- Reduced training data and compute requirements through experience‑driven updates
- Hardware‑aware design that lowers memory bandwidth, power draw, and interconnect complexity
- Compatibility with existing AI accelerator architectures while opening new efficiency frontiers
- Support for scalable deployment across robotics, edge devices, and large‑scale cloud models