Today’s AI models are becoming resources heavy, pollutants, and overeating electricity for training their giant neural networks, leaving a significant CO2 fingerprint and climate impact that just keeps growing with each new model. As described by Timnit Gebru and co-authors in their recent breakthrough paper:
“While the average human is responsible for an estimated 5 tons CO2 per year … a Transformer (big) model training procedure emitted 284 tons of CO2. Training a single BERT base model (without hyperparameter tuning) on GPUs was estimated to require as much energy as a trans-American flight.”
Each human reading this should feel a bit sad for a moment - while we are trying to do so much to reduce emissions, companies are building models that each is wasting as many dozen people will waste over an entire year! There must be a better way.
Cynamics Hidden Pattern Recognition (HPR) is not only a next-gen NDR but revolutionizing the AI models' CO2 emissions as well - collecting only a radically negligible amount of sample packets from the client’s network enabling our models to be compact, super-small size yet very accurate, and be trained rapidly on commodity CPU machines without requiring enormous GPU powerful infrastructure and an unbelievable training length.
Moreover, Cynamics' pretrained approach, sparring new training from scratch for each and every client while continuously learning and evolving thanks to a patented transfer learning approach makes Cynamics HPR even greener, offering a unique sustainable AI approach.
With Cynamics HPR, you can cover your entire network in a few minutes with appliance-less and agentless onboarding, leave nothing behind as a blindspot, predict attacks and threats from the gateways through the main assets to the endpoints long before they hit, and still do your humble part in reducing CO2 emissions and climate impact.
Reach out to us today to protect your entire network while protecting our environment!
Click on the images to download our white papers to gain more insights!
Comments