Minds and Computers : An Introduction to the Philosophy of Artificial Intelligence
Download 1.05 Mb. Pdf ko'rish
|
document (2)
- Bu sahifa navigatsiya:
- Exercise 19.1
Figure 19.1
Computing AND. threshold value of C is such that it will only fire if both A and B fire. If A alone fires, then the activation value of C will be 3 which is below the threshold value of 5. Similarly if B alone fires. If, however, they both fire, then the weighted sum transfer function tells us that the acti- vation value of C will be the sum of the values of the a fferent con- nection weights, which in this case will be 6. This activation value of 6 is higher than the threshold value of 5 assigned to C, so it will fire in accordance with its threshold activation function. This network serves as a logic gate. It computes the binary logical truth function of conjunction. The output unit fires i ff both A and B fire. Exercise 19.1 (a) How might we modify the network depicted in Figure 19.1 such that it computes the binary logical function of disjunction – i.e. so that the output node fires i ff either A or B (or both) fire. (b) Design a network with two input nodes and one output node such that the output node will fire if either of the input nodes fire but not if they both fire. If you succeeded in completing Exercise 19.1(b) then you have designed a network which computes the binary logical truth function of exclusive disjunction. This is not a straightforward exercise since we need to do two things that are new to us. One is to assign inhibitory weights; the other is to add a hidden unit to the network between the input and output units. The solution is depicted in Figure 19.2. If we disregard node D in Figure 19.2 for a moment, then we have the solution to Exercise 19.1(a). All we needed to do was lower the threshold value of C to a value below either of the a fferent connec- tion weights. (Alternatively we could have raised both connection weights to a value above the threshold.) In order to prevent C from firing when both A and B fire, however, we need to add node D. Node D will fire i ff both A and B fire and will inhibit the activation of C to prevent it from firing. If A alone fires, then the activation value of D will be below the threshold value and it will not fire, but the acti- vation value of C will be above its threshold and it will fire. Similarly if B alone fires. If A and B both fire, however, then D will fire, as its activation will be above its threshold value. The activation value of C will be the sum of its a fferent connection weights (3 3 5 1) 190 which is below its threshold value so, by virtue of being inhibited by D, it will not fire. To recap, C will fire if either A fires or B fires but not if they both fire, quod erat demonstrandum. 19.3 SYNTHESISING SPEECH In this section we’re going to design an artificial neural network to function as an English speech synthesiser. This is a nice example of the kind of contextually sensitive processing tasks at which connec- tionist networks excel. English orthography is not phonemic – it does not admit of a regular mapping onto English phonemes. Unlike Japanese, for instance, which is such that the pronunciation of any given grapheme is contextually invariant, the pronunciation of a given English grapheme is dependent on its orthographic context. In other words, English is not ‘pronounced as it is written’, to speak loosely. Rather, the sound that a given letter stands for is contextually dependent – one and the same letter can stand for di fferent sounds and the same sound can be represented by various letters, depending on the spelling context. Consequently, designing a speech synthesiser for English involves, inter alia, implementing the contextually sensitive processing task of converting orthographic representations to phonemic representa- tions. We’re going to begin constructing an artificial neural network for implementing this conversion of orthographic input to phonemic output. 191 Download 1.05 Mb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling