Questions for Emergent Simulation All questions are due upon completion of all the Labs In the following assignments the question numbers correspond to these in the text book for easier identification of the subject area. Please use them as a guide in finding proper exercises to prepare your answers. Since these questions are strongly related to the text book material, please read the corresponding section of the text to verify your understanding of the question and possible explanation based on the Emergent models. Try to keep your answer short (one paragraph for each question or in case of subquestions (a), (b), (c) one paragraph for each one). Please use illustrations from the simulator sparingly – only if you have to illustrate your answer. Since many of these questions are correlated with the lab exercises, work on them systematically during the semester, so you have less work at the end when all the questions are due. Students must work independently and present their own results answering questions according to their own understanding of the obtained results; however collaboration between students in the lab preparatory work (setting tolls, finding information about the problems and related data or discussing the underlying theory) is strongly encouraged. Finally, through these questions I can verify that you did a decent work during the lab, so they will influence your grade from the lab as well (in addition to a separate grade from the questions). Question 2.3 (a) How does the response of the unit change when you change g_bar_l? Why? (b) How does this differ from changes to g_bar_e? (c) Use the same technique you used in the previous question to compute the exact amount of leak current necessary to put the membrane potential exactly at threshold when the g_bar_e value is at the default of .4 (show your math). Question 2.4 What can you conclude about the relationship between the resting potential and the leak reversal potential? Question 2.6 Describe and explain the effects on the spike rate of decreasing g_bar_e to .38, and of increasing it to .42. Question 3.1 How did the lack of bias weights affect the hidden unit activities, and their relation to the number of active units in the input patterns? Question 3.3 (a) What happens generally to the hidden activations with the reduction in leak value? (b) How does this affect the cluster plot of hidden unit activities? (d) If the goal of the network is to have the same hidden representation for each version of the same digit, and different representations for different digits, how does changing the units' excitability (via the leak current) affect the success of the network, and why? Question 3.13 Why do you think kWTA can use a fast update rate where unit-based inhibition cannot? Question 5.1 Explain why the obtained pattern of strong and weak weights resulted from the CPCA Hebbian learning algorithm. Question 5.3 Explain why the delta rule weights solve the problem, but the Hebbian ones do not (don't forget to include the bias weights bias.wt in your analysis of the delta rule case). Question 5.6 How fast does GeneRec learn this EASY task described on p. 171 compared to the Hebbian rule? Be sure to run several times in both, to get a good sample. Question 6.1 Report the summary statistics from the batch text log (Batch_1_Textlog for your batch run). Does this indicate that your earlier observations were generally applicable? Question 6.2 Explain the results in terms of the weight patterns, the unique pattern statistic, and the general effects of Hebbian learning in representing the correlational structure of the input. Question 6.4 Interpret the cluster plot you obtained (especially the clusters with events at zero distance) in terms of the correspondence between hidden states and the current node versus the current letter. Remember that current node and current letter information is reflected in the letter and number before the arrow. Question 8.2 Which different properties of edges are encoded differently by different hidden units? There are four main ones, with one very obvious one being orientation - different hidden units encode edges of different orientations (e.g., horizontal, vertical, diagonal). Describe three more such properties or dimensions. Question 8.4 Explain the significance of the level of conjunctive representations and spatial invariance observed in the V2 receptive fields, in terms of the overall computation performed by the network. Question 8.6 Based on this latest display, do V4 units appear to code for entire objects, or just parts of different objects? Explain. Question 8.7 For those units that were significantly active, based on the number of different locations for which the unit was active (i.e., the area of colored pixels in the display), would you say that these units exhibited at least some degree of spatial invariance? Explain. Question 8.9 By what mechanism does the spatial cue influence the subsequent processing of the target in the valid and invalid cases? Question 9.2 (a) Report the average testing statistic (avg_tst_se) for a batch run of 5 simulated subjects. (b) How do these results compare to the human data presented in figure 9.4? (c) Looking at the training graph log, roughly how many epochs does the network take to reach its maximum error on the AB list after the introduction of the AC list? Question 9.4 Report the average testing error (avg_tst_se) for the batch run, and the number of epochs it takes the network to reach its maximum error on the AB list after the introduction of the AC list. Question 9.8 Report the number of times the network responded "a" instead of "b" for the "b" test trials. Question 9.12 Explain how the AC unit accurately predicts future reward, and at what point it does so (note that the external reward is visible as the activation state of the AC unit on the first plus phase of the recall trial). Question 9.13 Describe what happens to the network's internal representations and output (gaze, reach) responses over the delay and choice trials. You should observe the network making the error. Question 10.1 (a) Do you think the initial phonological activation is caused by the "direct" input via orthography or the "indirect" input via semantics? (b) Check for any cases where this initial phonological pattern is subsequently altered when the later input arrives, and describe what you find. Question 10.6 (a) Is there evidence in the model for a difference between concrete and abstract words in the number of semantic errors made? (b) Explain why this occurs in terms of the nature of the semantic representations in the model for these two types of words (recall that concrete words have richer semantics with more overall units). Question 10.9 (a) Do the most active units code for the appropriate inflectional phonological pattern? (b) Describe the steps you took to reach this answer. Question 10.11 (a) Report the cluster plot and cosine matrix results. (b) Comment on how well this matches your intuitive semantics from having read this textbook yourself. Question 10.12 Think of another example of a word that has different senses (that is well represented in this textbook), and perform an experiment similar to the one we just performed to manipulate these different senses. Document and discuss your results. Question 11.1 (a) At which layers in the network are the differences greatest? (b) Can you explain this in terms of the error signals as they propagate through the network? Question 11.2 (a) Describe what happens in the network during the conflict color naming condition, paying particular attention to the activations of the hidden units. (b) Explain how this leads to the observed slowing of reaction time (settling). Question 11.4 Explain why PFC lesions do not affect learning in the IDS task in the network (focus on the advantages of the PFC, and why the demands of the task do not require these advantages). Question 11.6 Explain why the dorsolateral (dimensional) lesion has no effect on the intradimensional reversal. Question 11.8 Explain why the absence of the dimension-level (dorsolateral) PFC units impairs extradimensional shift performance in this way.