CS380 - Final

Using the moving average history window of length 5, what is the final average of the sequence 0 50 60 70 80 10 _____

(50 + 60 + 70 + 80 + 10) / 5 = 54 Only grab the last 5 number, nucka

When rand() is called it gives back a number in what range: [________________, ______________________]

0, 32767... 0x7FFF

If a search algorithm is complete, then what is guaranteed and under what condition is it guaranteed?

A path if it exists

Jeff Orkin used what algorithm to search through possible sequences of action that would achieve a goal state:

A*

Name 4 search algorithms that are smart?

A* Best-First Floyd Warshall JPS

Name 4 search algorithms that guarantee an optimal path:

A* (heuristic pending) Dijkstra Floyd Warshall JPS

IF Agent A has node visibility array 0111010, Agent B has 1001000, and Agent C has 0001010. If we want to attack Agent A and not be seen by Agents B and C, which nodes could we stand on? A. 2nd node B. 3rd node C. 4th node D. 5th node E. 7th node

A. 2nd Node B. 3rd Node

Terrain Analysis Project: Visible to player map had pink squares. (a) What was their significance? and (b) what 2 properties gave them that significance?

A. Ambush Spots B. Adjacent to and Visible to a red tile

Pathfinding in Company of Heroes used (pick all that apply) A. Grid search space B. Hierarchical pathfinding C. Nav mesh D. Crush map E. Impass map F. A* G. IDA*

A. Grid Search Space B. Hierarchical Pathfinding E. Impass Map F. A*

With gesture recognition a new gesture must first be processed to minimize the effect of which aspects? (circle all that apply) A. size B. curvature C. position D. Weight E. speed (number of sample)

A. Size C. Position

Summarize Smart Objects

Agent gets behavior from object

Which search space(s) can be used with the Floyd-Warshall algorithm

All of the above

What info does a Company of Heroes precise map contain?

Allowed unit size

Describe an influence map (data structure and algorithm)

Array containing unit strengths

Describe the data structure and content of a flow field

Array of a vector field containing steering vectors

Is it better to: A. Perform a pathfinding search simultaneously for all request every frame. B. Perform a pathfinding search on one path request at a time and make the rest wait n a queue

B.

IN spatial terrain analysis, a visibility layer will weight grid cells by: A. How many agents are close to a given grid cell. B. How exposed a given cell is to all other grid cells. C. How close a grid cell is to static obstacles. D. How exposed the grid cell is to agent line-of-sight.

B. How exposed a given cell is to other grid cells

Circle all algorithms that are useful for gesture recognition: A. genetic algorithms B. neural networks C. nearest neighbor D. decision trees E. n-gram F. weakness modification learning

B. Neural Networks C. Nearest Neighborts

Explain how n -gram recombination work (green eggs and ham exercise);

Based on last element in SERIES, PREDICT next N-Gram, attach ACCORDING to previous sequence

What are flocking creatures called

Boids

Circle the easiest symbol group to detect as input in a mouse-based gesture recognition system: A. W M T J 3 B. U P V O 2 C. N W S E 1 D. L C Z S 0 E. D S O C 5 F. Q W U L 9

C. Because reasons...? Do whatever you want

An influence map should be calculated: A. From scratch each frame so that it stays accurate and up-to-date. B. Only when the AI needs to consult it, to avoid extra work. C. By continuously updating it only when forces move. D. Once every couple frames since forces tend to move slowly.

C. By continously updating it only when forces move.

Circle all that apply to indirect adaptation: A. Riskier than direct adaptation once shipped B. Generalizes from training data. C. Constructs a model of how the game is being played D. Tunes and adapts parameters that directly control the AI.

C. Constructs a model of how the game is being played.

An influence map can help you find: A. Badly damaged units B. Choke points C. The front of the battle D. Ambush spots

C. Front of the battle

Which kind of N-Gram would be most appropriate for the sequence 131131832131413_ A. Uni-Gram B. Bi-Gram C. Tri-Gram D. Quad-Gram E. N-Gram would not not be appropriate

C. TRI-GRAM

In spatial terrain analysis an area search layer will show: A. Potential hiding places for enemies. B. Promising areas to search for enemies. C. Which grid cells have been recently visited. D. Grid cells that do not contain enemy agents.

C. Which grid cells have been recently visited

The player made a gesture consisting of [5, 18, 9, 12]. Using the nearest neighbor technique along with root mean square error, which gesture in the database matches best? A. [1,18,9,16] B. [2,15,5,12] C. [8,15,12,12] D. [10,18,9,14] E. [5,18,9,18]

C. [8, 15, 12, 12] You do things... or just remember to put down C

Why is the Problem of Other Mind relevant to deciding if machines can ever have human-level consciousness?

Cannot prove that other minds are real.

The pathfinding project used a _____________ spline that was smooth in the ____________________.

Catmull-Rom C1, first derivative smooth baby

Name 3 other rules/steering behaviors to enhance flocking

Changing Goals Obstacle Avoidance Wandering

List the 3 flocking rules

Cohesion, Alignment, Separation

What is the Technological Singularity?

Computers become as smart as humans, and exponentially get more intelligent. Da borg.

Which search space typically results in the fewest nodes?

Corner

Name the father of flocking

Craig Reynolds

IF Agent A has node visibility array 0111010, Agent B has 1001000, and Agent C has 0001010. Circle all nodes that are safe from these agents. A. 2nd node B. 3rd node C. 4th node D. 5th node E. 7th node

D. 5th Node E. 7th Node

What is the abstract strategy/algorithm in Weakness Modification Learning that could be applied to any game?

Determine when a weakness occurs, alter behavior slightly so that history does not repeat

What are the three agent vision check in order from least to most expensive?

Distance, FOV, Raycast

Agent hearing can most efficiently be modeled with event-driven messages that are sent base on _______ and ______

Distance, Hierarchical Zones

How can you fake learning (without using indirect or direct adaptation)?

Dumb down AI and increase over time

Name 3 major benefits of the State Machine Language (SML) approach

Easy to program Easy to debug Easy to read

For a grid search space that allows for diagonals, determine if each heuristic is admissible/inadmissible and rank them from best to worst for most accurate underestimating distance in this context (1 being best, 5 being worst) Euclidean distance: Octile distance: Zero distance: Manhattan distance: Chebychev distance:

Euclidean distance: Admissable, 2 Octile distance: Admissable, 1 Zero distance: Admissable, 4 Manhattan distance: Inadmissible, 5 Chebychev distance: Admissible, 3

What positive effect does an inadmissible heuristic cost have on the A* search

Faster Search

What is the general strategy of Rabin's Filtered Randomness algorithm?

Filter out patterns using rules, if one is broken re-roll

What are genetic algorithms good at?

Finding optima in poorly understood spaces

Explain how to calculate a natural random offset for bullets aimed at the center of a target:

Gaussian Randomness, Central Limit Theorem

Why is one search algorithm (not search space) fast then another?

Good cost heuristic

Which Search space typically results in the most nodes

Grids

What game and person popularized the use of behavior trees?

Halo2; Damian Isla

Write a ONE LINE equation to generate Gaussian randomness in the range[-3, 3]with roughly a NORMAL distribution (which has a standard deviation of 1). Your solution should make use of the random function RandReal() that returns a uniform random number in the range [0,1]:

He's following you About 30 feet back. He gets down on all fours and breaks into a sprint. He's gaining on you. DICK TRACY.

The top level hierarchical map in Company of Heroes most resembled which classic search space representation?

Hex grid space

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Avoid allocating memory for nodes during search:

LARGE

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Check for possible straight-line path before running A*:

LARGE

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Choice of Search space representation:

LARGE

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Goal Bounding:

LARGE

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Hierarchical pathfinding:

LARGE

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Overestimate heuristic:

LARGE

Given that Floyd-Warshall is faster than A*, what is the primary tradeoff to get that speed?

Memory

In SML, what single mechanism supports event-driven behavior?

Messaging

Why is Behavior tree better than FSM?

More modular and scalable

Name 1 disadvantage that SML has compared to a traditional scripting language

Must be compiled

Name a problem with using neural networks:

Must be trained, and can be over-trained

Which search space is best suited for representing the entire walkable surface area?

Navmesh

Which search space(s) represent the walk-able surface are?

Navmesh, Grid, Circle-Based

In SML, what is the key benefit of the global message response feature?

No code duplication. Violently consume Dick Tracey.

In planning what did Jeff Orkin use as his heuristic?

Number of unmet pre-conditions

What is the cost/time complexity of Manager coordination with n agents: Using a blackboard architecture with n agents: Coordination using decentralized agent communication with n agents:

O(n) O(k) O(n^2)

Describe the general Player Modeling strategy / algorithm:

Observe and record player traits on a range of 0-1, and use it to adjust game.

If A* uses an admissible heuristic, then what is guaranteed to be found if it exists?

Optimal Path

Name two learning algorithms used in the game Black & White:

Perceptrons Decision Tree

At runtime, how does the Floyd-Warshall generate paths faster than A*

Pre-computes path

Based on Jeff Orkin's GDC talk, actions are composed of __________ and ___________

Pre-conditions and Effects

How can 1D Perlin Noise be used for AI?

Provides a smooth and continuous set of random numbers.

Generally there are 2 types of movement. __________ and __________ movement

Reactive, Planned

Why is the Mind Body problem relevant to deciding if machines can ever have human-level consciousness?

Robots are physical and therefore cannot have a spiritual component.

Describe 2 key benefits of subsumption architecture

Robustness; Goal preservation

Name two declarative AI architectures

Rule-based and Utility-based

Look at Page 5 of the test and do A*

SH!T

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. A* bidirectional pathfinding:

SMALL

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Avoid considering parent node when exploring neighboring nodes:

SMALL

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Check if destination is unreachable using some other method before running A*:

SMALL

Given this optimization or choices, judge whether the benefit is LARGE or SMALL relative to the average search cost. Cache failed start-end pairs:

SMALL?

Jeff Orkin planning is based on what academic planning system?

STRIPS

Name 2 major problems with using a state machine for game AI

Scalability and Maintenance No enforced structure

Conceptually agents have a ________, _______, ______ cycle that they loop through every frame.

Sense, Think, Act

Name 2 advantages of a limited, focused scripting system like a trigger system?

Simple Easy for designers

In flocking, group behavior emerges from what?

Simple Individual Rules

Describe a blackboard architecture:

Single location to store common information between agents to allow for open communication and lower memory footprint.

Agent must avoid flip-flopping decisions. Agents should stick with a decision until _____________ or ______________

Situation Changes Enough, or Enough time has passed

Name a problem with using genetic algorithms:

Slow and not guaranteed to find an optimal solution

An influence map is likely to use which equation to determine influence (where d stands for distance)?

Strength / (2 * D)

In general, what are neural networks good at?

Taking noisy data and processing it to produce a match

Why is Planning a declarative architecture?

Tells A.I. the goal, lets A.I. decide how to get to goal.

Formation can be said to have 2-level steering. What are the 2 level?

Template Individuals going to spot

If the heuristic cost in A* is absolutely accurate what does it exactly represent?

The cost to goal

Why might Searle's Chinese Room argument be wrong?

The system as a whole knows Chinese

In player modeling AttackOnTwoFronts, UsesSmokeGrenades, and CanDoTrickyJumps are examples of ________________

Traits

What was "Rabin's Radical belief on Randomness in Game"?

True random is too random

Searle's Chinese Room is a thought experiment to argue against what?

Turing test and strong A.I.

Use this sequence for the following question to determine the probability of the next move being a Kick: Punch, Punch, Punch, Kick, Punch, Uppercut, Punch, Punch, Kick, Punch, Punch, Kick, Punch, Punch: Uni-Gram: _________ Bi-Gram ___________ Tri-Gram __________ Quad-Gram ___________

Uni-Gram: 3/14 Bi-Gram: 3/9... 1/3 Tri-Gram: 3/4 Quad-Gram: 1/1

What quality or property is a good tactical indicator in static terrain analysis?

Visibility

Why is indirect adaptation preferred over direct adaptation? Circle all that are true: A. Indirect is more robust and stable B. Indirect is usually computationally faster. C. Indirect is easier to fix when behaving badly. D. Indirect results in more novel learning

What ever you feel like mang... Probably A? Maybe D? not B... Maybe C?

If a delayed message is scoped to the state and the state never changes from the time the message was sent, will the message be delivered to a global message response if it isn't handled in the current state?

Yes

In player modeling, list all possible values that AttacksOnTwoFronts could have:

[0,1]

Using the moving average equation with an alpha of 0.2, what is the final average of the sequence 0 50 60 70 80 10 _____

a * (newObs) + (1 - a) (avg) 0.2 * (10) + (0.8) * (0 + 50 + 60 + 70 + 80) / 5 = 43.6

In player modeling, UsesSmokeGrenades is stored as what data type?

a float

The cost in a Dijkstra search is the ______________ cost from the __________ node to the ______________ node.

accumulated... start... current

Subsumption has layers of code that run: ___________. The priority is with the _____________ layers. The ___________ layers make request to the ___________ layers.

concurrently; lower; higher, lower

The cost in a Best-First search is the ___________ cost from the _____________ node to the ____________ node

estimated... current... end

Calculation for octile distance:

f(x) = min(diffX, diffY) * sqrt(2) + max(diffX, diffY) - min(diffX, diffY)

PERLIN NOISE STUFF

feed Dick Tracy to bears

Name 2 Problems with using a complex scripting language

hard to debug; time commitment

If the heuristic is admissible in A* then it ________________ the true cost.

never over-estimates

Name 2 Benefits of using a complex scripting language

no recompiling code; extensible to players