Chapter 16 – Improve the Emotional Intelligence Deficiencies of Chatbots

1. Restricted Boltzmann Machines are based on directed graphs. (Yes | No)

No. RBM graphs are undirected, unsupervised, and memoryless, and the decision making is based on random calculations.

2. The hidden units of an RBM are generally connected to each other. (Yes | No)

No. The hidden units of an RBM are not generally connected to each other.

3. Random sampling is not used in an RBM. (Yes | No)

No. False. Gibbs random sampling is frequently applied to RBMs.

4. Is there a method to prevent gradients from vanishing in an RNN? (Yes | No)

Yes. When the gradient gets "stuck" around 0, for example, a ReLU function can solve this problem. It will force negative values to become 0 (or a fixed value in a leaky ReLU), and the positive values will not change.

5. LSTM cells never forget. (Yes | No)

No. False. LSTM cells "forget" by skipping connections, adding connections from the past to the present, and other techniques.

6. Word2vector transforms words into indexes along with their labels. (Yes | No)

Yes. Word2vector transforms words into numbers and also keeps track of their labels. 

7. Principal Component Analysis (PCA) transforms data into higher dimensions. (Yes | No)

Yes. The whole point of PCA is to transform data into higher dimensions to find the principal component (highest eigenvalue of a covariance matrix), then the second highest, and down to the lowest values.

8. In a covariance matrix, the eigenvector shows the direction of the vector representing that matrix, and the eigenvalue shows the size of that vector. (Yes | No)

Yes. Eigenvalues indicate how important a feature is, and eigenvectors provide a direction.

9. It is impossible to represent a human mind in a machine. (Yes | No)

No. It is possible.

Chapter 17, Quantum Computers That Think, shows how to build a machine mind. It takes quite some work to include sensors, but the technology is there.

10. A machine cannot learn concepts, which is why classical applied mathematics is enough to make efficient artificial intelligence programs for every field. (Yes | No)

No. Never believe that. Progress is being made and will never stop until mind-machines become mainstream.