CogSci1 Top 10 List Chapter Summary
xex Home

The following Top Ten List of Big Ideas summarizes Chapter 6, "Connectionism", of Stan Franklin's [[Artificial Minds]]. The the words are generally taken straight from the text.

  1. Connectionism Connectionism is also known as artificial neural networks. Connectionist models are often used to model cognitive functions at a level of abstraction belew the symbol level.

  2. Why Connectionism? "The central idea of connectionism is that cognition can be modeled as simultaneous interaction of many highly interconnected neuronlife units.

  3. The Neuron Neurons consist of three parts which are the dendrite, the cell body, and the axon. There is also one other very important aspect to neurons which isn't a part of it that is the synapses. The dendrite tree is responsible for collecting inputs from other neurons which passes the inputs, as voltages, to the cell body. From the cell body the input it taken sent to synapses that connect the axon of one neuron to the dendrite of another. In a real nervous system neurons are either inhibitory or excitatory to all subsequent neurons to which they connect.

  4. ANN or Symbolic Models? Connectionists believe that ANN are far superior to symbolic models because of the lack of a central executive, because control is distributed in a AAN, because of the automatic presence of default assignments, and because ANNs degrade more gracefully than symbolic models.

  5. ANN Learning There are five types of ANN learning that are discussed in Artificial Mind which are hard-wiring (weights and connections are specified by a human designer), error correction (responses compared to target output and weights adjusted), reinforcement (numeric score over sequence of trials and weights adjusted), stochastic (Random or Hebbian weight change accepted if cost decreases or by some probablity), and self organizaion (weights modified in response to input).

  6. Logic Gates ANNs can be represented by logic gates. There are five basic logic gates which are the AND gate, the NOR, gate, the NAND, gate, the NOT gate, the XOR gate, and the OR gate. The AND gate gives a high output when all inputs are high, the OR gate gives a high output when one or more inputs is high, the NAND gate is high when not all of the inputs are high, the NOT gate inverses its input, the NOR gate will give a high output only when no inputs are high, and the XOR gate will give a high output when one and only one input is high.

  7. Representation "Once a problem is described using appropriate representations, the problem is almost solved." Representations are very important because humans typically operate using them and not rules.

  8. Local Represenation Local Represetation employs one unit to represent one object, one concept, or one hypothesis.

  9. Distributed Representation Distributated Representation is when one unit may participate in the representation of several items and each item is represented by a pattern of activity over several different untits.

  10. Featural Representation Featural Representation is when items are represented distributively by patterns of activity over sets of units, and the individual units locally represent features of the given item.