For an input at timestamp t, an RNN expects 1 output. At timestamps t+1 to t+5, there are respective outputs but no inputs i.e. at timestamp one there is 1 input and 1 output but at timestamp two to six there is no input but 1 output. What kind of an RNN is this?

Assume a 3x3 grayscale image has the following intensity values 10, 0, 40, 20, 60, 30, 100, 80, 70. Range normalize this image then apply a 3x3 max pool filter (padding is 0 and stride is 2) on it. What is the smallest output value computed by this max pool filter when it max pools this image?

If an RNN hidden layer has 3 processing neurons, how many weights (whether hidden state output or bias weights) exist to feedforward the hidden state at time t-1 (previous timestamp) to the hidden state at time t (current timestamp)?

For 30 days, a company kept track of its share price throughout the day, along with the first share price for the next day. The objective was to build a model that can use a day's sequence of share prices to predict the next day's first share price. It is assumed that the share prices towards the end (as opposed to the beginning) of the day have a higher influence on the next day's first share price. Input sequences of 5 share prices were recorded for the first 25 days. On the other 5 days, input sequences of only 3 share prices were recorded. The number of share prices recorded in a day differed depending on the demand for the company's shares on that day. Which of the following approaches is most recommended for this scenario?

What is the maximum number of unique gray levels that can be in a 10x10 grayscale image?