Question

list and discuss ten shortcomings of artificial neural networks

Answer 1

Shortcomings of artificial neural networks:

1) Black box: The calculation of result is hidden. It is not known how data is processed.

2) Optimizing parameters: Optimizing the network can be challenging, as neural network requires  too many parameters.

3) Computationally expensive: Artificial neural networks are also more computationally expensive than traditional algorithms. It can take several weeks to train completely from scratch and requires high processing time for large neural networks.

4) Operation: The neural network needs training to operate.

5) Unpredictable behavior : Neural networks behaves a similar way the human brain does. They learn  by examples as they cannot be programmed to perform a specific task. Careful selection of examples is important as  network finds solution by itself and its operation can be unpredictable. This reduces trust in the network.

6) Determination of appropriate network structure: Determination of appropriate network structure is done through experience and trial and error as there is no specific rule defined for determining the structure.

7) The duration of the network is unknown: The network is reduced to a certain value(which is not optimal) of the error on the sample means that the training(period is unknown) has been completed.

8) Hardware dependence: According to the structure of artificial neural networks, they are dependent of equipment like, processors with parallel processing power.

9) Representing the problem to the network is difficult: As artificial neural networks works with numerical information, problems need to be translated into numerical value. Representation of the problem to artificial neural networks directly influences the performance of the network .

10) Large amount of data required for training: In general ,Neural networks usually require large amount of data for learning.