AI Exam Study Notes

All essential concepts and key figures from the AI curriculum

Artificial Intelligence (AI) Definition

Definition

AI is an area of computer science that aims to make machines do intelligent things, that is, learn and solve problems, similar to the natural intelligence of humans and animals.

In AI, an intelligent agent receives information from the environment, performs computations to decide what action to take to achieve the goal, and takes actions autonomously.

AI can improve its performance with learning.

Machine Learning (ML) Definition

Definition

ML is a set of mathematical algorithms that can automatically analyze data and make decisions or predict results for given data.

"A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E."
- Tom Mitchell (1997)

The term "Machine Learning" was coined by Arthur Samuel in 1959 at IBM.

Types of Artificial Intelligence

Important

Narrow AI (Weak AI): Used to solve specific problems. Almost all AI applications today are narrow AI.

General AI (Strong AI): Able to solve general problems, similar to humans in learning, thinking, and inventing.

Super AI (Superintelligence): AI after the singularity point when it surpasses human intelligence.

Types of Machine Learning

Important

Supervised Learning: Models trained with labeled data. Divided into:

  • Classification
  • Regression

Unsupervised Learning: Models fed with unlabeled data. Divided into:

  • Clustering
  • Association

Semi-supervised Learning: Uses both labeled and unlabeled data.

Reinforcement Learning: Learns through trial and error to find actions that yield maximum cumulative reward.

Turing Test

Important

Proposed by Alan Turing in October 1950 in paper "Computing Machinery and Intelligence".

Three-person game called imitation game where interrogator tries to determine which player is computer and which is human.

Interrogator limited to using responses to written questions.

To date, no computer has passed the Turing test.

Ray Kurzweil predicts AI will pass Turing test in 2029.

Neural Networks (NN)

Algorithm

Developed in 1950s-1970s based on human brains mimicking biological neural networks.

Usually has three layers:

  • Input Layer
  • Hidden Layer
  • Output Layer

Requires training with large amount of data.

After training, can predict results for unseen data.

Deep Learning (DL)

Algorithm

Developed from 2010s-present.

Special type of neural network with more than one hidden layer.

Made possible with increased computing power (GPUs) and improved algorithms.

DL is a subset of ML.

Outperforms many other algorithms on large datasets.

Support Vector Machine (SVM)

Algorithm

Best-known supervised learning algorithm.

Used for both classification and regression problems.

Developed at AT&T Bell Laboratories by Vladimir Naumovich Vapnik in 1990s.

One of the most robust prediction methods.

Advantages:

  • Good accuracy and faster prediction than Naïve Bayes
  • Uses less memory
  • Works well with clear margin separation and high-dimensional space

Disadvantages:

  • Not suitable for large datasets
  • High training time
  • Poor performance with overlapping classes

Naive Bayes

Algorithm

Supervised learning algorithm based on Bayes' Theorem, used mainly for classification.

Called "naive" because assumes all features are independent - rarely true but works surprisingly well.

Types of Naive Bayes:

  • Gaussian Naive Bayes - for continuous numeric features
  • Multinomial Naive Bayes - for counts (word frequency)
  • Bernoulli Naive Bayes - for binary features

Advantages: Simple and fast, works well with high-dimensional data, requires small training data.

Disadvantages: Independence assumption unrealistic, poor performance with strongly correlated features.

Decision Trees

Algorithm

Nonparametric, supervised learning method for classification and regression.

Represents a set of if-then rules learned from data.

Predicts by following path from root node to leaf.

Deeper trees capture more detail but risk overfitting.

Terminology:

  • Root node: Start of the tree
  • Internal node: Question on a feature
  • Leaf node: Terminal node with prediction
  • Branch/edge: Outcome of question
  • Depth: Number of edges from root to leaf

Random Forest

Algorithm

Ensemble learning algorithm using multiple decision trees.

Works for both classification and regression.

Improves performance by combining many trees.

Reduces overfitting compared to single tree.

How it works:

  • Creates decision trees randomly
  • Each tree trained on random subset of samples
  • Each split considers random subset of features
  • Final prediction: majority vote (classification) or average (regression)

K-Nearest Neighbors (K-NN)

Algorithm

Supervised learning algorithm for classification and regression.

Classification based on majority vote among K nearest neighbors.

How it works:

  • Pick value for K (number of neighbors)
  • Find K closest data points using distance metric (Euclidean)
  • Predict label: majority class (classification) or average (regression)

Key Notes:

  • Instance-based learning (no training phase)
  • Sensitive to: value of K, distance metric, feature scaling
  • Larger K → smoother decision boundaries

LDA vs PCA

Comparison

Linear Discriminant Analysis (LDA):

  • Supervised
  • Maximizes class separation
  • Creates discriminants (new axes)

Principal Component Analysis (PCA):

  • Unsupervised
  • Maximizes variance
  • Creates components (new axes)

Both commonly used for dimension reduction.

Key People in AI

People

Alan Turing: British mathematician, developed bombe code-breaking machine, proposed Turing test.

Ray Kurzweil: Author, inventor, futurist, predicted AI passing Turing test in 2029 and singularity in 2045.

Stephen Hawking: Theoretical physicist, warned about AI risks.

Gray Scott: American futurist, quote about AI surpassing humans by 2035.

Steve Polyak: American neurologist, quote about natural stupidity vs AI.

Arthur Samuel: Coined term "Machine Learning" at IBM in 1959.

Tom Mitchell: Provided famous ML definition in 1997.

Vladimir Vapnik: Developed SVM at AT&T Bell Labs.

Demis Hassabis: Founded DeepMind (AlphaGo, AlphaFold).

Elon Musk: Founded OpenAI and Neuralink.

Simon Knowles & Nigel Toon: Founded Graphcore (IPU).

Robert Griesemer, Rob Pike, Ken Thompson: Developed Go language at Google.

Yangqing Jia: Developed Caffe framework at UC Berkeley.

Model Evaluation Terms

Important

Accuracy: Percentage of correct predictions out of all predictions.

Precision: Of all items predicted as positive, how many were actually positive?

Recall: Of all actual positive cases, how many did the model catch?

F1 Score: Harmonic mean of Precision and Recall. Balances false positives and false negatives.

AUC-ROC: Measures model's ability to separate positive and negative classes.

Cloud AI vs Edge AI

Comparison

Cloud AI:

  • Runs on large remote servers
  • Advantages: No expensive hardware, upload large datasets, utilize cloud computing power
  • Disadvantages: More bandwidth, higher latency, security issues

Edge AI:

  • Runs on local machines
  • Advantages: Low latency, works without Internet, real-time, secure
  • Disadvantages: Need own hardware, limited computation power

AI Hardware Tools

Important

GPUs (Graphics Processing Units):

  • Significantly speed up calculations with parallel processing
  • Example: NVIDIA GeForce RTX 1080/2080 Series
  • Best for: Rapid prototyping, dense AI training

FPGAs (Field Programmable Gate Arrays):

  • Manufacturers: Altera, Xilinx
  • Intel acquired Altera in 2015
  • Best for: Real-time, fixed algorithms, mission-critical workloads

IPUs (Intelligence Processing Units):

  • Graphcore's massively parallel processors
  • Run faster than GPUs, cost less
  • Best for: Sparse data, dynamic graphs

AI Development Tools & Frameworks

Important

Programming Languages:

  • Python (most popular for AI)
  • C/C++ (fast, used in OpenCV, YOLO)
  • Java (Android apps, Weka, Deeplearning4j)
  • R (statistical analysis)
  • Julia (high-performance computing)
  • Go (Google-developed, simple and efficient)

Key Frameworks:

  • Scikit-Learn (most widely used for ML)
  • TensorFlow (Google, most widely used for AI)
  • PyTorch (Facebook, gaining popularity)
  • Keras (standard interface for TensorFlow)
  • Caffe/Caffe2 (deep learning framework)
  • PaddlePaddle (Baidu's deep learning framework)

Important AI Quotes

Important
"There is no reason and no way that a human mind can keep up with an artificial intelligence machine by 2035."
- Gray Scott (American futurist)
"Success in creating AI would be the biggest event in human history. Unfortunately, it might also be the last, unless we learn how to avoid the risks."
- Stephen Hawking (British theoretical physicist)
"Before we work on artificial intelligence why don't we do something about natural stupidity?"
- Steve Polyak (American Neurologist)