Karl Friston
(1959-)
Karl Friston is a British neuroscientist and theoretician at University College London. He is an authority on brain imaging and theoretical neuroscience, especially the use of physics-inspired statistical methods to model neuroimaging data and other complex dynamical systems.
Friston is best known for his "free energy principle," (FEP) a unifying account of self-organisation, which suggests that all biological systems, including the brain, act to minimize uncertainty.
Friston's "free energy" inherits from Feynman's path integral formulation and is an information theoretic measure (as opposed to thermodynamic free energy, as defined by
J. Willard Gibbs, which is the energy available to do work in a system that is below thermal equilibrium (maximal disorder).
Active inference is an application of the FEP to perception, action, and learning, where systems try to make their sensory input predictable. In simpler terms, it's a way of describing how living things maintain their existence by actively reducing uncertainty about their environment. It resembles
Robert Rosen's "anticipatory system" and underwrites "predictive processing" of the sort discussed by
Andy Clark and Jakob Hohwy.
1. Free Energy and Biological Systems:
• Free energy, in this context, is a mathematical quantity that provides a variational bound on surprise, uncertainty, or disorder.
• The free energy principle proposes that biological systems, from cells to brains, minimize variational free energy, effectively acting to maintain a stable internal state despite external changes.
• The free energy principle proposes that biological systems, from cells to brains, minimize this free energy, effectively acting to maintain a stable internal state despite external changes.
• This minimization occurs through perception (updating beliefs about the environment) and action (changing the environment to fit predictions)
2. Bayesian Inference and Active Inference:
• The free energy principle is closely linked to Bayesian inference, where agents
make predictions based on prior knowledge and update these predictions based
on new sensory information.
• Active inference is a key aspect of the principle, where agents actively try to make their sensory input more predictable by changing their actions. For example, a tree growing towards the light minimizes free energy by acting to elude the surprise of finding itself in the shade, given the kind of thing it is.
3. Generative Models:
• The free energy principle invokes the concept of a generative model, which is a model of the world that an agent uses to make predictions.
• By minimizing free energy, agents are essentially optimizing their internal
generative models to better reflect the structure of the environment.
Normal |
Teacher |
Scholar
|
