Memory Prediction Framework
When I was doing research for our final year project Cyclopia, a system which is built to model the image understanding capability of humans, I came across this wonderful concept of intelligence introduced by Jeff Hawkins in his book called “On Intelligence”.
This theory says that physical arrangement of the brain cortex issue is uniform which means that there is only single principle defining the memory and the neocortex. This means 
that the same algorithm is in operation for learning models for audition, vision, somatosensory perception and language. It also goes on to say that the brain’s intelligent comes from its ability to predict future events based on the past data.
Hawkins outlines the book as follows:
“The book starts with some background on why previous attempts at understanding intelligence and building intelligent machines have failed. I then introduce and develop the core idea of the theory, what I call the memory-prediction framework. In chapter 6 I detail how the physical brain implements the memory-prediction model—in other words, how the brain actually works. I then discuss social and other implications of the theory, which for many readers might be the most thought-provoking section. The book ends with a discussion of intelligent machines—how we can build them and what the future will be like.”
The central concept of the Memory-Prediction Framework is that bottom-up inputs are matched in a hierarchy of recognition, and evoke a series of top-down expectations. These expectations interact with the bottom-up signals to both analyze those inputs and generate predictions of subsequent expected inputs. Each hierarchy level remembers frequently observed temporal sequences of input patterns and generates labels or ‘names’ for these sequences. When an input sequence matches a memorized sequence at a given layer of the hierarchy, a label or ‘name’ is propagated up the hierarchy – thus eliminating details at higher levels and enabling them to learn higher-order sequences. This process produces increased invariance at higher levels. Higher levels predict future input by matching partial sequences and projecting their expectations to the lower levels. However, when a mismatch between input and memorized/predicted sequences occurs, a more complete representation propagates upwards. This causes alternative ‘interpretations’ to be activated at higher levels, which in turn generates other predictions at lower levels.
The main points of Memory prediction framework can be summarized as follows:
- The neocortex is constructing a model for the spatial and temporal patterns that it is exposed to. The goal of this model construction is the prediction of the next pattern on the input.
- The cortex is constructed by replicating a basic computational unit known as the canonical cortical circuit. From a computational point of view , this canonical circuit can be treated as a node that is replicated several times.
- The cortex is organized as hierarchy. This means that the nodes are connected in a tree shaped hierarchy.
- The function of the cortex is to model the world that it is exposed to. This model is built using as spatial and temporal hierarchy by memorizing patterns and sequences at every node of the hierarchy. This model is then used to make predictions about the input.
- The neocortex builds its model of the world in an unsupervised manner.
- Each node in the hierarchy stores a large number of patterns and sequences. The pattern recognition method employed by the cortex is largely based on storing lots of patterns.
- The output of a node is in terms of the sequences of patterns it has learned.
- Information is passed up and down in the hierarchy to recognize and disambiguate information and propagated forward in time to predict the next input pattern.
Memory Prediction Framework (MPF) as expressed in On Intelligence is a biological theory. After working on the foundation established by Hawkins, Dileep George who is a cofounder of Numenta along with Jeff Hawkins, developed the algorithmic and mathematical counterparts of the memory prediction framework which is known as the Hierarchical Temporal Memory (HTM) which is very useful when it comes to practical implementation of the MPF. I will be going into detailed explanation about HTM in my future posts.
Related links:
