twin-shema model
The bishema model is a modular learning model
Modular learning model in which new modules are created by subjectively discovering changes in the dynamics of the environment
What is Shema?
Piaget
Self-organizing processes in cognitive development
Its basic module is the Shema
Assimilation and regulation
Assimilation: incorporating information from the outside world into the Shema
Which Shema to incorporate
Regulation: changing the shema to take in information from the outside world
The assimilation and regulation cycle is similar to both steps of the k-means methodnishio.icon
Equilibration and differentiation
Above, processes of assimilation and regulation are in equilibrium
When data that is far from the Shema comes in, a new Shema is created, and this is differentiation.
Dual-schema model
Assume a robot with sensors and motors
Sensor input is a finite-dimensional real-valued vector
Based on sensor inputs, the internal processing system makes decisions and produces action outputs.
The action output is a real-valued vector of several dimensions
at time t
Sensor input, perceptual vector: $ S_t
Action output, action vector: $ A_t
The concept of an object is "how we act upon it and what are the consequences of our actions".
The case ratio of [pragmatism
That is, $ U_t = (A_t, S_t, S_{t+1}) is the primitive form
We call this the experience vector.
Furthermore, consider F such that $ F(S_t, S_{t+1}) = A_t .
Instantaneous motor output is rarely meaningful for agents operating in real space (2006)
The policy function $ a_t = \pi(s_t) is what should be called an "action" (2006), based on the idea that a series of actions acting on an object is an action.
The two expressions are connected when combined with a function expressing intent, as explained later.
Split the shema of Biaget into two parts.
whynishio.icon
Act Shema and Perceived Shema
Perceived Shema: $ \hat{F}=(F,\alpha): A_t = F(S_t, I_{t+1}) + \alpha\delta_t
Act Shema: $ \hat{G}=(G, \beta): I_{t+1} = G(S_t) + \beta\delta_t
where $ I_{t+1} is the perception (Intention) that the subject wants to obtain at time t+1, ideally $ S_{t+1} \simeq I_{t+1}.
The combination of the two selects actions from inputs from the outside world.
The program of an action is a function that determines A from S $ A_t = H(S_t).
:$ A_t = F(S_t, G(S_t))
Whether or not : $ U_t satisfies the function F of the perceptual shema corresponds to whether or not the shema is appropriate as a symbol to represent an object existing in the external world. The process of assimilation can be performed here.
If the model error $ (A_t - F(S_t, S_{t+1}))^2 is small, the target belongs to the perceptual shema with F
Do you take the margin of error there?$ (S_{t+1} - I_{t+1})^2 ではなくて？nishio.icon
Adds a limit to storage capacity
queue
We will store the U_t when we were in that shema state in this queue.
F optimizes the samples in its queue to approximate with minimum error
Regulation Process
Limited memory capacity contributes to plasticity
Summary: Assimilation and Adjustment
By selecting the shema with the smallest model error, we can identify the shema to which the target belongs (assimilation).
The object is recorded in the shema. The function of the shema is optimized to represent the object (adjustment).
differentiation
The model error at time t is calculated from the model error at time t....
confidence variable
If the adjustment of F is not sufficient, the denominator of R will also be larger in the same dimension
An increase in R can be recognized as some change outside the system
Acquisition of action shema through reinforcement learning
Bishema-based reinforcement learning
Hierarchical modular learning machine
Resources on sensory-motor learning are
context of the subject's interaction with the outside world, Piaget's Shema is divided into two parts, the action Shema and the perception Shema -> twin Shema
Bi-Schema is a cumulative modular learning device based on Piaget's theory of the Shema, in which an autonomous robot acquires Shema (concepts) corresponding to models of the environment/object through interaction with the environment (2005).
The idea of symbolic grounding ignores the robot's own subjective world; instead of starting with human symbols, symbols for the robot must be generated through its own memory organizing mechanisms
J. Uexkull Biological Semiotics
Meanings of symbols do not exist objectively, but are formed through cognitive developmental processes and social communication
---
This page is auto-translated from /nishio/双シェマモデル using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.