FractalBrain
Fundamentally NewAGI Model
What makes FractalBrain Unique?
Lifelong Learning: FractalBrain grows its parametric model, at both training and inference time, allowing it to assimilate knowledge at inference time, permanently. In contrast, knowledge assimilated by Neural Network at inference time, aka "in-context-learning" is only temporary.
Local Learning: FractalBrain pioneers a sparse, Hebbian-type, local learning approach, enabling orders of magnitude improvements in power and data efficiency over Neural Networks that rely on global optimization methods. This also allows FractalBrain to handle context windows of unlimited size.
Causal Learning: FractalBrain learns a temporal, deterministic model of the world, from actively sensed key environmental observations. This results in knowledge represented as explainable, causal chains, rather than the opaque Neural Network vectors, effectively elliminating the model halucinations.
Reinforcement Learning: FractalBrain RL is natively fused within its deep connectome, allowing for model-based, hierarchical reinforcement learning and options-based strategic planning. This is in stark contrast to Neural Network based Deep RL, which in reality uses shallow and myopic, action-based planning.
Who are we?
We are a team of industry-hardened PhDs and engineers from top institutions including DeepMind, Samsung, IBM, CERN, DESY and Cambridge University. Fractal Brain is the result of over a decade of our R&D efforts, and has only been possible thanks to our world-class expertise in AI/ML, Fractal Theory, Theoretical Physics and Quantum Computing. We have offices in the Bay Area and London and our engineering teams operate remotely. The best way to contact us is at [email protected]
