Practice exam part 2

10 Questions

Multi-layer perceptron. There is a specific operation that a multi-layer perceptron cannot perform. Which?

The multi-layer perceptron is a universal function approximator: it can perform any operation

Grid Coding. Select the wrong statement. Entorhinal cortex grid cells

In a single individual differ in their grid orientation

Tolman.

Proved, in 1948, that rats maintain a cognitive map of their surroundings even without explicit reward

Encoding models. The space in which encoding models cannot be formulated:

There is no space in which encoding models cannot be formulated

Biological Realism. Find the wrong answer. The basic perceptron model fails to mimic biological neurons in:

its fundamental input-output architecture

Grid fit procedure. Find the wrong answer. Caveats of grid fitting:

Cannot be used for least-squares estimation, and has to be used for maximum likelihood estimation

Gradient Descent. Find the wrong answer.

Gradient Descent cannot be used for least-squares estimation, and has to be used for maximum likelihood estimation

Reinforcement learning. The learning rate in a Rescorla-Wagner model:

Determines the speed at which learning tends towards its asymptote

Temporal Difference Learning TD solves this problem of the Rescorla-Wagner model:

RW only predicts immediate rewards: no higher-order conditioning

Signal detection theory. Imagine a simple signal-in-noise SDT experiment. In the first half, I reward hits but not correct rejections, and in the second half of the experiment, I reward correct rejections but not hits. What will the impact on my participants’ behavior be?

They will adjust their criterion, and be more conservative in the second half while they were more liberal in the first.

Study Notes

Multi-Layer Perceptron

A multi-layer perceptron cannot perform a specific operation, which is not specified.

Grid Coding

Entorhinal cortex grid cells are involved in Tolman's encoding models.

Encoding Models

There is a space in which encoding models cannot be formulated, which is not biological realism.

Perceptron Model

The basic perceptron model fails to mimic biological neurons in terms of their ability to handle non-linearly separable patterns.

Grid Fit Procedure

One of the caveats of grid fitting is that it can be overly reliant on the grid's parameters.

Gradient Descent

One of the incorrect statements about gradient descent is not specified.

Reinforcement Learning

In a Rescorla-Wagner model, the learning rate determines how quickly the model learns from rewards and punishments.

Temporal Difference Learning (TD)

TD solves the problem of the Rescorla-Wagner model, which is that it only learns from the immediate consequences of an action.

Signal Detection Theory (SDT)

If, in a simple signal-in-noise SDT experiment, hits are rewarded but not correct rejections in the first half, and correct rejections are rewarded but not hits in the second half, the participants' behavior will change to prioritize the newly rewarded behavior in each half.