Skip to content
Last Dictionary
Plane
Philosophy & MindMathematics & StatisticsReinforcement Learning
contested

free energy principle

nounAlso: FEP, variational free energy

In plain wordsA sweeping theory by the neuroscientist Karl Friston that living things persist by keeping themselves in expected states — in effect, by constantly minimizing surprise about what they sense.

  1. The proposal, due to Karl Friston, that any self-organizing system that maintains itself in a bounded set of states, statistically separated from its environment by a boundary called a Markov blanket, can be described as minimizing variational free energy: an information-theoretic quantity that upper-bounds surprise, the negative log probability of its sensory states under an internal generative model. Minimizing free energy by updating internal states amounts to approximate Bayesian perception (predictive processingpredictive processing, n.: The idea that the brain is a prediction machine: it constantly guesses what its senses will report next and learns mainly from its mistakes, so what we perceive is largely its best guess.); minimizing it by acting on the world is active inferenceactive inference, n.: A theory that living things, and machines built on the same idea, act in order to make the world match their predictions, choosing actions expected to keep them in preferred states and reduce their uncertainty.. Variational free energy is the negative of the evidence lower boundevidence lower bound, n.: A quantity that is always less than or equal to how well a model explains its data, but much easier to calculate; pushing it upward is how models such as VAEs learn. maximized in variational autoencodersvariational autoencoder, n.: A neural network that learns to squeeze data into a compact, slightly fuzzy code and rebuild it, arranged so that decoding random codes produces new, plausible examples.. Supporters see a unifying theory of life and mind; critics argue that, as a principle, it is unfalsifiable and that its scope and key terms are unclear.

F=Eq(s)[ln⁡q(s)−ln⁡p(o,s)]=DKL[q(s) ∥ p(s∣o)]−ln⁡p(o)  ≥  −ln⁡p(o)F = \mathbb{E}_{q(s)}\left[\ln q(s) - \ln p(o, s)\right] = D_{\mathrm{KL}}\left[q(s) \,\|\, p(s \mid o)\right] - \ln p(o) \;\geq\; -\ln p(o)

Origin

Introduced by Karl Friston, James Kilner and Lee Harrison in 'A Free Energy Principle for the Brain' (Journal of Physiology–Paris, 2006), drawing on free-energy methods in Geoffrey Hinton and Peter Dayan's machine-learning work of the 1990s; summarized in Friston's 'The Free-Energy Principle: A Unified Brain Theory?' (Nature Reviews Neuroscience, 2010).

Usage note

Friston and colleagues present it as a principle, like the principle of least action in physics, rather than as a testable theory; process theories built on it, such as predictive coding, make testable predictions. In the formula, o denotes observations, s hidden states and q an approximate posterior.

Related

See also
  • active inferenceactive inference, n.: A theory that living things, and machines built on the same idea, act in order to make the world match their predictions, choosing actions expected to keep them in preferred states and reduce their uncertainty.
  • predictive processingpredictive processing, n.: The idea that the brain is a prediction machine: it constantly guesses what its senses will report next and learns mainly from its mistakes, so what we perceive is largely its best guess.
  • Kullback–Leibler divergenceKullback–Leibler divergence, n.: A score for the cost of believing the wrong odds: on average, how much more surprised you will be by real outcomes than someone who knew the true probabilities. It is zero only when your odds are exactly right.
  • evidence lower boundevidence lower bound, n.: A quantity that is always less than or equal to how well a model explains its data, but much easier to calculate; pushing it upward is how models such as VAEs learn.
  • Bayesian inferenceBayesian inference, n.: Reasoning that treats unknowns as uncertain beliefs and updates them with evidence using Bayes' theorem. The result is not one answer but a range of answers, each with a probability.
  • artificial lifeartificial life, n.: The study of life by building lifelike systems — in computer simulations, robots or chemistry — to learn how living things organize themselves, reproduce and evolve, and what other kinds of life could exist.

Sources

  1. Friston, K., Kilner, J. & Harrison, L. (2006). A Free Energy Principle for the Brain. Journal of Physiology–Paris 100(1–3), 70–87.
  2. Friston, K. (2010). The Free-Energy Principle: A Unified Brain Theory? Nature Reviews Neuroscience 11(2), 127–138.

Mentioned in

predictive processingpredictive processing, n.: The idea that the brain is a prediction machine: it constantly guesses what its senses will report next and learns mainly from its mistakes, so what we perceive is largely its best guess., active inferenceactive inference, n.: A theory that living things, and machines built on the same idea, act in order to make the world match their predictions, choosing actions expected to keep them in preferred states and reduce their uncertainty., artificial lifeartificial life, n.: The study of life by building lifelike systems — in computer simulations, robots or chemistry — to learn how living things organize themselves, reproduce and evolve, and what other kinds of life could exist., Karl FristonKarl Friston, prop. n.: A British neuroscientist who proposed that brains, and perhaps all living systems, work by constantly predicting their surroundings and acting to reduce surprise, an idea some see as a blueprint for AI.