Artificial Intelligence Perception, Communication, and Ethics Syllabus

Perception, Communication, and Ethics syllabus — full chapter and topic list

Expand any chapter to see its topics and sub-topics. This is the whole examinable outline for Perception, Communication, and Ethics in Artificial Intelligence, not a summary of it.

1. 1

   Natural Language Processing

   4 topics

   • Language Models
   • Text Classification and Information Retrieval
   • Parsing and Syntactic Analysis
   • Machine Translation and Large Language Models
2. 2

   Computer Vision

   4 topics

   • Image Formation and Preprocessing
   • Feature Detection
   • Object Recognition and Detection
   • Image Segmentation and Scene Understanding
3. 3

   Robotics

   4 topics

   • Robot Hardware and Sensors
   • Robotic Perception and Localization
   • Motion Planning
   • Robotic Control and Manipulation
4. 4

   Philosophy, Ethics, and Safety of AI

   4 topics

   • Weak and Strong AI
   • Ethics of AI
   • AI Safety and Alignment
   • The Future of AI

Perception, Communication, and Ethics flashcards for Artificial Intelligence

24 of 50 cards from the Perception, Communication, and Ethics deck — real questions with worked answers.

1. In an $n$-gram language model, what does the Markov assumption state about the probability of the next word?

   It assumes the probability of a word depends only on the previous $n-1$ words, so $P(w_i \mid w_1,\dots,w_{i-1}) \approx P(w_i \mid w_{i-n+1},\dots,w_{i-1})$.

2. Define perplexity as an intrinsic evaluation metric for language models.

   Perplexity is the inverse probability of a test set normalized by the number of words: $PP(W) = P(w_1,\dots,w_N)^{-\frac{1}{N}} = \sqrt[N]{\frac{1}{\prod_{i=1}^{N} P(w_i \mid w_1,\dots,w_{i-1})}}$. Lower perplexity indicates a better model.

3. Why is smoothing (e.g., Laplace/add-one) needed in n-gram language models?

   To assign non-zero probability to unseen n-grams, avoiding zero probabilities that would make the whole sequence probability zero. Add-one smoothing: $P(w_i \mid w_{i-1}) = \frac{C(w_{i-1}, w_i) + 1}{C(w_{i-1}) + V}$, where $V$ is the vocabulary size.

4. In text classification, state the Naive Bayes decision rule and its core independence assumption.

   Choose the class $\hat{c} = \arg\max_{c} P(c) \prod_{i} P(w_i \mid c)$. It assumes features (words) are conditionally independent given the class.

5. Define term frequency-inverse document frequency (TF-IDF) and give its formula.

   TF-IDF weights a term by how often it appears in a document versus how rare it is across the corpus: $\text{tfidf}(t,d) = \text{tf}(t,d) \times \log\frac{N}{\text{df}(t)}$, where $N$ is the number of documents and $\text{df}(t)$ is the number containing term $t$.

6. In information retrieval, define precision and recall.

   Precision $= \frac{\text{relevant retrieved}}{\text{total retrieved}}$ (fraction of retrieved documents that are relevant); Recall $= \frac{\text{relevant retrieved}}{\text{total relevant}}$ (fraction of all relevant documents that were retrieved).

7. Give the formula for the F1 score used to evaluate a classifier or IR system.

   $F_1 = 2 \cdot \frac{\text{precision} \cdot \text{recall}}{\text{precision} + \text{recall}}$, the harmonic mean of precision and recall.

8. In the vector space model of IR, how is similarity between a query and a document typically measured?

   By cosine similarity: $\cos(\theta) = \frac{\vec{q} \cdot \vec{d}}{\lVert \vec{q} \rVert \, \lVert \vec{d} \rVert}$, the cosine of the angle between the query and document TF-IDF vectors.

9. Distinguish constituency parsing from dependency parsing.

   Constituency (phrase-structure) parsing groups words into nested phrases (NP, VP) via a context-free grammar; dependency parsing represents syntax as directed head-dependent relations between individual words, producing a dependency tree.

10. What algorithm is classically used for parsing with a context-free grammar in Chomsky Normal Form, and what is its time complexity?

    The CYK (Cocke-Younger-Kasami) algorithm, a bottom-up dynamic-programming parser with time complexity $O(n^{3} \cdot |G|)$ for a sentence of length $n$.

11. What is a Probabilistic Context-Free Grammar (PCFG) and how is the probability of a parse tree computed?

    A PCFG assigns each production rule a probability, with rules for the same left-hand side summing to 1. The probability of a parse tree is the product of the probabilities of all rules used to build it.

12. In dependency grammar, what does it mean for a parse to be 'projective'?

    A dependency parse is projective if no dependency edges cross when drawn above the sentence in linear order; equivalently, each head and its descendants form a contiguous span.

13. In the Transformer architecture, give the scaled dot-product attention formula.

    $\text{Attention}(Q,K,V) = \text{softmax}\!\left(\frac{Q K^{\top}}{\sqrt{d_k}}\right) V$, where $Q$, $K$, $V$ are the query, key, and value matrices and $d_k$ is the key dimension.

14. What is the purpose of the $\sqrt{d_k}$ scaling factor in Transformer attention?

    It counteracts the growth of dot-product magnitudes as dimension $d_k$ increases, preventing the softmax from entering regions with extremely small gradients.

15. What is the BLEU score and what does it measure in machine translation?

    BLEU (Bilingual Evaluation Understudy) measures translation quality by modified n-gram precision against reference translations, combined with a brevity penalty; scores range from 0 to 1 (often reported 0-100), higher being better.

16. Contrast encoder-only, decoder-only, and encoder-decoder Transformer architectures with an example of each.

    Encoder-only (e.g., BERT) produces bidirectional representations for understanding tasks; decoder-only (e.g., GPT) does autoregressive generation with causal masking; encoder-decoder (e.g., T5, original Transformer) maps an input sequence to an output sequence, ideal for translation.

17. What distinguishes 'in-context learning' (few-shot prompting) in large language models from fine-tuning?

    In-context learning conditions the model on examples within the prompt at inference time without updating weights; fine-tuning updates the model's parameters through additional training on task data.

18. State the thin-lens equation used in image formation.

    $\frac{1}{f} = \frac{1}{v} + \frac{1}{u}$, where $f$ is the focal length, $u$ the object distance, and $v$ the image distance from the lens.

19. In the pinhole camera model, how does a 3D point project onto the image plane?

    Perspective projection: a 3D point $(X, Y, Z)$ maps to image coordinates $x = f\frac{X}{Z}$, $y = f\frac{Y}{Z}$, where $f$ is the focal length. Objects farther away (larger $Z$) appear smaller.

20. What is a Gaussian filter used for in image preprocessing, and what does its standard deviation $\sigma$ control?

    It smooths/blurs an image to reduce noise by convolving with a Gaussian kernel $G(x,y) = \frac{1}{2\pi\sigma^{2}} e^{-\frac{x^{2}+y^{2}}{2\sigma^{2}}}$. Larger $\sigma$ produces stronger blurring over a wider neighborhood.

21. What is histogram equalization and what problem does it address?

    An image-enhancement technique that redistributes pixel intensities to produce a more uniform histogram, improving global contrast in images that are too dark, too bright, or low-contrast.

22. What quantity does the Sobel operator compute, and how is gradient magnitude obtained?

    The Sobel operator estimates the image intensity gradient using two $3\times3$ kernels for horizontal ($G_x$) and vertical ($G_y$) derivatives; the edge magnitude is $G = \sqrt{G_x^{2} + G_y^{2}}$ and direction is $\theta = \arctan\!\left(\frac{G_y}{G_x}\right)$.

23. List the main stages of the Canny edge detector.

    1) Gaussian smoothing, 2) gradient computation (magnitude and direction), 3) non-maximum suppression to thin edges, 4) double thresholding, and 5) edge tracking by hysteresis to link edges.

24. What does the Harris corner detector respond to, and what image property defines a corner?

    It detects corners as points where intensity changes significantly in all directions. Using the structure tensor's eigenvalues $\lambda_1, \lambda_2$, a corner has both large; the response is $R = \det(M) - k(\text{trace}\,M)^{2}$.

See more Perception, Communication, and Ethics flashcards →

Planning Perception, Communication, and Ethics for Artificial Intelligence

Perception, Communication, and Ethics (Artificial Intelligence) FAQ