## The Emergence of Rules in Cell–Assemblies of FLIF Neurons

Citations: | 6 - 5 self |

### BibTeX

@MISC{Belavkin_theemergence,

author = {Roman V. Belavkin and Christian R. Huyck},

title = {The Emergence of Rules in Cell–Assemblies of FLIF Neurons},

year = {}

}

### OpenURL

### Abstract

Abstract. There are many examples of intelligent and learning systems that are based either on the connectionist or the symbolic approach. Although the latter can be successfully combined with statistical learning to create a hybrid system, it is not so clear how symbolic processing can emerge from a connectionst system. Human mind is a living proof that such a transition must be possible. Inspired by biological cognition, our project explores the ways symbolic processing can emerge in a system of neural cell–assemblies (CAs). Here, we present the meta–process that regulates learning of associations between the CAs. The process is compared with the stochastic learning theory, and its outcome is a set of optimal rules. The paper concludes by an example of a working system and the discussion of it biological plausibility. 1

### Citations

6115 |
The mathematical theory of communication
- Shannon, Weaver
- 1949
(Show Context)
Citation Context ...on algorithms implemented earlier in cognitive architectures and stochastic symbolic systems [3, 4]. Theoretical foundations of this theory are based on the variational problems of information theory =-=[19, 20]-=-, a generalisation of which is outlined below. 3.1 Optimisation with information constraints Rational action selection is related to the theories of choice and optimisation. Fundamental in the theory ... |

1592 |
Neural networks and physical systems with emergent collect computational abilities
- Hopfield
- 1982
(Show Context)
Citation Context ...emblies The system is based on networks of sparsely connected cells. The topology of the network is pre–defined by some random pattern, and it can be highly recurrent similar to the Hopfield networks =-=[10]-=-. Unlike the Hopfield nets, however, the links are unidirectional making our model more biologically plausible. The non–linearity of the cells in the network leads to a complex dynamics similar to tha... |

1424 |
Unified Theories of Cognition
- Newell
- 1990
(Show Context)
Citation Context ... the behaviour. The former facilitated the solution of a great variety of engineering problems (e.g. signal processing, pattern recognition), while the latter have revolutionised cognitive psychology =-=[18]-=-. Despite the successes, the process of unification of neural and symbolic cognitive systems have been slow even though human cognition — the main subject of both approaches — is a clear example that ... |

1309 | Reinforcement learning: A survey
- Kaelbling, Littman, et al.
- 1996
(Show Context)
Citation Context ...he activation of the system accodring to the utility of its experience. This method has many similarities with the reiforcement learning algorithms, where randomisation is used to control exploration =-=[13, 22]-=-, and with the adaptive networks where the reward signals were used to train artificial neurons [21, 2]. Here we have demonstrated how such an process can be implented in a sparcely connected system o... |

1167 | Information Theory and Statistics
- Kullback
- 1968
(Show Context)
Citation Context ...l probabilities respectively, then I(p, q) is the Shannon information. The important properties of information divergence is that it is convex, non–negative and its minimum is achieved for p = q (see =-=[15]-=-). The maximum of I(p, q), which can be infinite, is achieved for p → δωω ′, which are the probability measures concentrated entirely on single elements of Ω (here δωω ′ is the Kronecker symbol). Thus... |

1050 |
A quantitative description of membrane current and its application to conduction and excitation in nerve
- Hodgkin, Huxley
- 1952
(Show Context)
Citation Context ...ailed presentation can be found in [11]. 2.1 Fatiguing Leaky Integrate and Fire neurons Biological neurons are complex systems, and the levels of details varied significantly even in the early models =-=[17, 9]-=-. Our system uses spiking, fatiguing, leaky, integrate and fire (FLIF) units (artificial cells) [16]. Our model is a compromise between computational efficiency and biological plausibility reflecting ... |

1047 |
Theory of games and economic behavior
- Neumann, Morgenstern
- 1944
(Show Context)
Citation Context ...e relation on P is induced by the expected utility (p, u) = R u(ω)p(ω)dω, so that for any p, Ω q ∈ P , measure p is preferred if (p, u) ≥ (q, u), and it is the classical Bayesian estimation procedure =-=[24, 23]-=-. More generally, the problem of optimisation under uncertainty can be viewed as maximisation in the conjugate space. Indeed, given a Banach space U, the conjugate space V is the totality of all linea... |

974 |
The Organization of Behavior
- Hebb
- 1949
(Show Context)
Citation Context ...sides of the same coin. Although a single neuron can classify a large number of patterns, it is believed that the groups of connected cells called cell–assemblies (CAs) form the basis human cognition =-=[8]-=-. However, recent advances in modelling human–level cognition were made mostly using symbolic cognitive architectures, such as SOAR [18] and ACT–R [1]. The success of the latter can be explained large... |

730 |
A logical calculus of ideas immanent in nervous activity
- McCulloch, Pitts
- 1943
(Show Context)
Citation Context ...ailed presentation can be found in [11]. 2.1 Fatiguing Leaky Integrate and Fire neurons Biological neurons are complex systems, and the levels of details varied significantly even in the early models =-=[17, 9]-=-. Our system uses spiking, fatiguing, leaky, integrate and fire (FLIF) units (artificial cells) [16]. Our model is a compromise between computational efficiency and biological plausibility reflecting ... |

513 | An integrated theory of mind
- Anderson, Bothell, et al.
- 2004
(Show Context)
Citation Context ...l–assemblies (CAs) form the basis human cognition [8]. However, recent advances in modelling human–level cognition were made mostly using symbolic cognitive architectures, such as SOAR [18] and ACT–R =-=[1]-=-. The success of the latter can be explained largely by using the hybrid approach, where symbols are applied selectively based on statistical associations and other sub–symbolic computations. This wor... |

226 |
Statistical Decision Functions
- Wald
- 1950
(Show Context)
Citation Context ...e relation on P is induced by the expected utility (p, u) = R u(ω)p(ω)dω, so that for any p, Ω q ∈ P , measure p is preferred if (p, u) ≥ (q, u), and it is the classical Bayesian estimation procedure =-=[24, 23]-=-. More generally, the problem of optimisation under uncertainty can be viewed as maximisation in the conjugate space. Indeed, given a Banach space U, the conjugate space V is the totality of all linea... |

218 |
Pulsed Neural Networks
- Maass, Bishop
- 1999
(Show Context)
Citation Context ...ons are complex systems, and the levels of details varied significantly even in the early models [17, 9]. Our system uses spiking, fatiguing, leaky, integrate and fire (FLIF) units (artificial cells) =-=[16]-=-. Our model is a compromise between computational efficiency and biological plausibility reflecting properties that are, in our opinion, particularly important for the emerging dynamics. The ‘integrat... |

191 | Toward a Modern Theory of Adaptive Network
- SUTTON, BARTO
- 1981
(Show Context)
Citation Context ...s with the reiforcement learning algorithms, where randomisation is used to control exploration [13, 22], and with the adaptive networks where the reward signals were used to train artificial neurons =-=[21, 2]-=-. Here we have demonstrated how such an process can be implented in a sparcely connected system of FLIF neurons, where the formation of cell– assemblies can be employed for a symbolic–like processing.... |

119 |
Cortical substrates for exploratory decisions in humans. Nature
- Daw, O’Doherty, et al.
- 2006
(Show Context)
Citation Context ...s to optimisation with information constraints. Finally, recent studies in neuroscience of exploratory behaviour suggest that the method proposed may have some biological plausibility. In particular, =-=[6]-=- failed to identify conclusively any specific area of the brain correlated with the exploration function, and the model based on the Gibbs distribution was proposed as the most plausible. Some researc... |

72 |
Pattern recognizing stochastic learning automata
- Barto, Anandan
- 1985
(Show Context)
Citation Context ...s with the reiforcement learning algorithms, where randomisation is used to control exploration [13, 22], and with the adaptive networks where the reward signals were used to train artificial neurons =-=[21, 2]-=-. Here we have demonstrated how such an process can be implented in a sparcely connected system of FLIF neurons, where the formation of cell– assemblies can be employed for a symbolic–like processing.... |

30 | Engines of the brain: The computational instruction set of human cognition
- Granger
- 2006
(Show Context)
Citation Context ...l based on the Gibbs distribution was proposed as the most plausible. Some researchers have speculated about the role of tonically active cholinergic neurons in the basal ganglia and striatal complex =-=[7]-=-. These neurons account for a small proportion of the connections that are quite uniform and nontopographic. It was suggested that these neurons may play the role of a stochastic noise. Interstingly, ... |

19 | The use of entropy for analysis and control of cognitive models
- Belavkin, Ritter
(Show Context)
Citation Context ...en shown earlier how the entropy feedback from the posterior probability can be used to control β in the ACT–R architecture, which significantly improves cognitive models of human and animal learning =-=[3, 5]-=-. A similar stochastic control has been used to implement optimal learning and adaptation of agents in stochastic environments [4]. In the next section, we present how such a stochastic process was im... |

17 | Tracing recurrent activity in cognitive elements (TRACE): A model of temporal dynamics in a cell assembly
- Kaplan, Sonntag, et al.
- 1991
(Show Context)
Citation Context ...the other cells in the system. Such reverberating groups of cells are often referred to according to Hebb [8] as the cell–assemblies (CAs). An important property of CAs’ dynamics is their persistence =-=[14]-=-: Once ignited, the activity within the cells in a CA may be sufficient to support itself. Many variables can contribute to this effect. In particular, the fatigue and recovery rate parameters in our ... |

15 | Counting with neurons, rule application with nets of fatiguing leaky integrate and fire neurons
- Huyck, Belavkin
- 2006
(Show Context)
Citation Context ... processing and learning in a connectionist system. Previously, the authors have demonstrated how states in a CA– based system can be controlled and used to perform a typical symbolic task (counting) =-=[12]-=-. This work has developed into a much more ambitious project called CABOT, where the same principles are applied in a system integrating elements of vision, categorisation, natural language processing... |

12 | On emotion, learning and uncertainty: A cognitive modelling approach
- Belavkin
- 2003
(Show Context)
Citation Context ...he meta–process for learning the connectivity between the modules is based on the stochastic action–selection algorithms implemented earlier in cognitive architectures and stochastic symbolic systems =-=[3, 4]-=-. Theoretical foundations of this theory are based on the variational problems of information theory [19, 20], a generalisation of which is outlined below. 3.1 Optimisation with information constraint... |

9 | Acting irrationally to improve performance in stochastic worlds
- Belavkin
- 2005
(Show Context)
Citation Context ...he meta–process for learning the connectivity between the modules is based on the stochastic action–selection algorithms implemented earlier in cognitive architectures and stochastic symbolic systems =-=[3, 4]-=-. Theoretical foundations of this theory are based on the variational problems of information theory [19, 20], a generalisation of which is outlined below. 3.1 Optimisation with information constraint... |

8 |
Modelling cell assemblies
- Huyck
- 1999
(Show Context)
Citation Context ... learning process will also be considered.in the last section. 2 OVERVIEW OF THE ARCHITECTURE Below is an overview of the neural model and its parameters. A more detailed presentation can be found in =-=[11]-=-. 2.1 Fatiguing Leaky Integrate and Fire neurons Biological neurons are complex systems, and the levels of details varied significantly even in the early models [17, 9]. Our system uses spiking, fatig... |