predictive coding
A scheme, originally for compressing sound files, in which a system minimizes the precision-weighted mismatch (prediction error) between what its model predicts and what it samples; applied to the brain as belief updating.
As defined in 87 | Karl Friston on Brains, Predictions, and Free Energy
Other definitions
A class of brain/AI models in which the system continually predicts inputs and updates based on prediction errors, assuming smooth statistical functions.
Forget Network Layers—Cortical Columns Think Like Graphs
Where it comes up · 12
- The Brain’s Learning Algorithm Isn’t Backpropagation6 claims
- 235 | Andy Clark on the Extended and Predictive Mind2 claims
- 87 | Karl Friston on Brains, Predictions, and Free Energy2 claims
- Forget Network Layers—Cortical Columns Think Like Graphs2 claims
- 227 - Knowing The Mind1 claim
- 247 - Constructing Minds1 claim
- Ask Me Anything 131 claim
- 359 - Getting Used To It1 claim
and 4 more
Most important claims · 10 of 20Search all claims for “predictive coding” →
- A recent paradigm shift centers on precision: much psychiatric phenomenology and possibly its neurochemistry can be explained by a failure to encode precision (the error bars / signal-to-noise of sensory evidence and beliefs); since nearly every pharmacological treatment targets modulatory neurotransmitters that change sensitivity to input rather than encode the beliefs themselves, broken precision estimation is the natural therapeutic target.87 | Karl Friston on Brains, Predictions, and Free Energy
- Neurologically, we hold internal predictive models of our perceptual and behavioral states; when stimuli conform to predictions their salience is inhibited and they fade, while prediction errors—surprises—are what we notice, and this applies to both positive and negative stimuli.359 - Getting Used To It
- As neuroscience and artificial intelligence continue to inform each other, predictive coding stands as a compelling bridge between the remarkable learning capabilities of biological brains and the next generation of neural network architectures.The Brain’s Learning Algorithm Isn’t Backpropagation
- Predictive coding originated from mid-twentieth century research proposing that the brain's fundamental objective is to predict incoming sensory information, which enhances survival by allowing organisms to anticipate threats and interpret noisy observations.The Brain’s Learning Algorithm Isn’t Backpropagation
- Well-predicted sensory inputs cause less neuronal activity than other inputs, which is a signature that led to predictive coding; attention reverses this dampening effect, so giving new evidence its proper due requires deliberate effort to attend against the grain of established expectations.235 | Andy Clark on the Extended and Predictive Mind
- The distinction between voluntary and involuntary action is real both phenomenologically and neurologically—voluntary action involves an efferent copy of motor expectation whose violations startle us, and can be suppressed when instructed—and this distinction remains ethically important even in a world without free will, just as we would worry about a robot that forms plans to murder without attributing free will to it.227 - Knowing The Mind
- Predictive coding—originally devised in the 1950s to compress sound files—is the brain minimizing prediction error: the weighted sum of squared mismatches between what the generative model predicts and what is sampled, equivalent to free energy under simple assumptions; it explains how the brain updates beliefs but says nothing about action.87 | Karl Friston on Brains, Predictions, and Free Energy
- Information flows bidirectionally through the predictive coding hierarchy with top-down connections carrying predictions from higher levels to lower levels, while bottom-up connections carry prediction errors (differences between predictions and actual activity).The Brain’s Learning Algorithm Isn’t Backpropagation
- Predictive perception works like motion-compressed video and JPEGs: the brain predicts the current sensory flow and only the residual prediction errors, which carry whatever sensory information is currently unexplained, need further processing, so if you predict correctly using what you know about the world you need very little bandwidth to refine the picture.235 | Andy Clark on the Extended and Predictive Mind
- Predictive coding's local autonomy makes the algorithm extremely parallelizable and in certain settings more efficient than back propagation, and theoretical considerations suggest that resulting updates may lead to better solutions than back propagation.The Brain’s Learning Algorithm Isn’t Backpropagation