COGSCI 1B Midterm Fall 2024 Questions and Answers

COGSCI 1B Midterm Fall 2024 Questions and Answers Strong AI - CORRECT ANSWER-AI formulation: machine w/ AGI, just as smart as humans in every way Applied AGI - CORRECT ANSWER-AI formulation: ML allows - machines to master "expert systems", driving cars etc. Why AI boom (early 21st century) - CORRECT ANSWER-New computing power + available large data sets Expert systems - CORRECT ANSWER-Machines that replicate/ improve human experts specialized domains (slide 8 for examples) Past ML applications - CORRECT ANSWER-See slide 10 Decision Trees - CORRECT ANSWER-ML programs asses probs w/ decisions trees How would a decision tree be designed - CORRECT ANSWER-- could ask a team of mortgage loan officers to make a decision tree - or could or could make program that can make it's own decision tree (ML) or ID3 ID3 - CORRECT ANSWER-ML tools that use stats make decision trees ID3, how work - CORRECT ANSWER-1. measure the informativeness of each attribute (correlation)(information gain) 2. Assigns the attribute with highest information gain to the first node on the tree 3. repeat (slides 14-18) CAPTCHA - CORRECT ANSWER-Completely Automated Public Turing Test To Tell Computers and Humans Apart. Wasn;t good enough first, and computers better reCAPTCHA - CORRECT ANSWER-Showed words computers did understand, when enough humans think it is something, it is confirmed as correct/ uploaded to ebook database. Computer-based personality judgements - CORRECT ANSWER-ML personality judgments based on facebook likes are more accurate than friends and better predicted life outcomes Feature engineering - CORRECT ANSWER-In ML, data is labeled with features that the algorithm will work on Feature engineering problem - CORRECT ANSWER-human experts must label, hard to label, can be labeled in many diff ways Representation learning - CORRECT ANSWER-program do their own feature engineering on raw data, include deep learning Selectivity/invariance problem - CORRECT ANSWER-needs to ignore the irrelevant diffs between what the animal looks like, like and angle and posture (slide 7) Supervised learning - CORRECT ANSWER-network receives explicit feedback on how successful it is deep reinforcement learning - CORRECT ANSWER-The network gets a reward signal (feedback) but now told what is wrong AlphaGo - CORRECT ANSWER-Historic win in abstract strategy board game using mixture of supervised/ unsupervised learning Beginning of vis percep - CORRECT ANSWER--Not talked about until the 1980s functional neuroimaging -Ungerleider and Mishkin's two visual system hypothesis Two vis system hypothesis - CORRECT ANSWER--experiments in monkeys - Divide perception into Dorsal Stream: perception of spatial location - Ventral Stream: perception of form Tri Level Hypothesis (1) - CORRECT ANSWER-- implementation level: what is the harware being used, how can be realized physically - raw primary skitched on retina by line segments in terms of light intensity (problem is how do we establish object constancy) Tri Level Hypothesis (2) - CORRECT ANSW