Neural nets and word recognition
- There were two leading theories of word recognition:
- Peterson performed scans to decide between these two theories of
- The COG theory came out ahead.
1. Viewed words.
2. Listened to words.
3. Spoke words.
4. Generate appropriate verbs.
The PET scan images above arise by subtraction. The average baseline
is subtracted by the average.
Genuine neural nets
- Evidence suggested to interconnections depicted:
Ch 8: Overview
- Introduce single unit networks and Boolean functions
- Introduce (Donald) Hebbian learning and the perceptron convergence
- Explain the limits of learning in single unit networks
- Introduce learning algorithms for multilayer networks
- Bottom-up; Top-down – Research methodology
- Connectionism developed with the idea that one should model
computations in a way inspired by the way the brain works.
Features of connectionist networks
- Exploit parallel processing
- Can be used to model multiple satisfaction of soft constraints
- Do not feature explicit rules
- Exhibit graceful degradation
- Intended as models of information-processing and the algorithmic
- Capable of learning