Information Theory
What can be compressed, what can be transmitted reliably, and why -- built on probability and linear algebra, from Shannon's entropy through channel capacity to today's machine-learning applications.
Free to read now, no account needed: Readiness: what information theory leans on. Create a free account to open the other chapters, take the quizzes and keep your progress.
1 The concept of information π
Create a free account to open this chapter.
- Quantifying surprise ~8 min
- Why probability alone isn't enough (Hartley's measure and its limits) ~9 min
- Self-information ~10 min
- Why logarithms ~9 min
- Choice of base: bits, nats, and other units ~9 min
- Additivity of information for independent events ~10 min
- Information and expectation ~10 min
- A first widget-driven intuition check ~9 min
2 Entropy π
Create a free account to open this chapter.
- Entropy as expected information ~9 min
- Entropy of a discrete random variable ~9 min
- The binary entropy function ~8 min
- Bounds on entropy (0 to log k) ~9 min
- When entropy is maximized ~9 min
- When entropy is minimized ~8 min
- Entropy of common distributions ~10 min
- Entropy under deterministic transformations ~10 min
- Entropy and coding, a first preview ~9 min
3 Joint entropy, conditional entropy, and mutual information π
Create a free account to open this chapter.
- Joint entropy ~11 min
- Conditional entropy ~11 min
- The chain rule for entropy ~10 min
- Conditioning reduces entropy ~12 min
- Entropy of independent variables ~9 min
- Mutual information, defined two ways ~12 min
- Mutual information as shared information ~12 min
- Conditional mutual information ~12 min
- Mutual information and independence ~9 min
- The data-processing inequality ~13 min
4 Relative entropy and cross-entropy π
Create a free account to open this chapter.
- Kullback-Leibler divergence ~11 min
- Gibbs' inequality (non-negativity) ~10 min
- Why KL divergence isn't a metric ~10 min
- Asymmetry and the two directions of KL ~10 min
- Cross-entropy ~10 min
- The entropy/cross-entropy/KL relationship ~10 min
- KL divergence in model comparison ~10 min
5 Source coding foundations π
Create a free account to open this chapter.
- Why compression is possible ~11 min
- Fixed-length vs. variable-length codes ~11 min
- Prefix (instantaneous) codes ~12 min
- The Kraft inequality ~12 min
- The McMillan inequality ~12 min
- Expected code length and redundancy ~11 min
- The entropy lower bound ~12 min
6 Huffman, Shannon-Fano, and arithmetic coding π
Create a free account to open this chapter.
- The Huffman algorithm ~11 min
- Why merging the two least-probable symbols is optimal ~13 min
- Building and reading a Huffman tree ~10 min
- When a distribution has more than one optimal tree ~10 min
- The Shannon-Fano algorithm ~11 min
- Huffman vs. Shannon-Fano ~10 min
- Arithmetic coding: the interval-narrowing idea ~11 min
- Encoding and decoding with arithmetic coding ~12 min
- Comparing the three methods on one source ~12 min
7 Adaptive and dictionary coding π
Create a free account to open this chapter.
- The problem of unknown source statistics ~9 min
- Adaptive Huffman coding and the sibling property ~13 min
- LZ77 ~12 min
- LZSS ~11 min
- LZ78 ~12 min
- LZW ~12 min
- Dictionary coding in practice (zip, gzip) ~10 min
- Compression method trade-offs ~11 min
8 The source coding theorem and the AEP π
Create a free account to open this chapter.
- Shannon's source coding theorem, stated ~10 min
- Typical sequences ~12 min
- The asymptotic equipartition property, two equivalent forms ~14 min
- Probability concentration as n grows ~12 min
- ~2^(nH(X)) typical sequences ~12 min
- The law of large numbers, revisited ~12 min
- Entropy as the fundamental compression limit ~14 min
- Lossless compression in the limit ~12 min
9 Differential entropy and Gaussian information theory π
Create a free account to open this chapter.
- Differential entropy for continuous variables ~11 min
- Why differential entropy can be negative ~10 min
- Differential entropy of the uniform distribution ~10 min
- Differential entropy of the Gaussian ~12 min
- The Gaussian maximizes entropy for fixed variance ~13 min
- Joint and vector Gaussian entropy ~13 min
- Entropy under linear transformations ~12 min
- Continuous mutual information and relative entropy ~13 min
- Why Gaussians appear throughout information theory ~11 min
10 Communication channels and channel capacity π
Create a free account to open this chapter.
- The source-encoder-channel-decoder-destination model ~10 min
- Discrete memoryless channels and transition probabilities ~12 min
- The binary symmetric channel ~12 min
- The binary erasure channel ~12 min
- Errors vs. erasures ~10 min
- Channel capacity as a mutual-information maximization ~14 min
- Capacity for symmetric and weakly symmetric channels ~16 min
- The capacity-achieving input distribution ~14 min
- Capacity vs. entropy ~10 min
- Channel capacity as an engineering budget ~12 min
11 The noisy-channel coding theorem π
Create a free account to open this chapter.
- Reliable communication, defined ~16 min
- Shannon's noisy-channel coding theorem, stated ~16 min
- Rates below capacity ~18 min
- Rates above capacity ~16 min
- Block length and error probability ~16 min
- Why the theorem is nonconstructive ~15 min
12 Error-correcting and linear codes π
Create a free account to open this chapter.
- Motivation: detecting vs. correcting errors ~9 min
- Hamming distance and minimum distance ~13 min
- Repetition and parity codes ~11 min
- Linear codes as a vector-space idea ~12 min
- Generator matrices ~12 min
- Parity-check matrices ~11 min
- Syndrome decoding ~12 min
- Hamming codes ~11 min
- The Hamming, Singleton, Plotkin, and Gilbert-Varshamov bounds ~14 min
- Convolutional codes, a first look ~11 min
- CRC and error detection ~12 min
13 Rate-distortion theory and lossy compression π
Create a free account to open this chapter.
- When lossless isn't the goal ~9 min
- Distortion measures (Hamming, squared error) ~10 min
- The rate-distortion function ~16 min
- Quantization ~13 min
- Vector quantization and transform coding ~12 min
- The discrete cosine transform ~15 min
- Rate vs. quality, quantitatively ~12 min
- JPEG-style compression, conceptually ~14 min
14 Information theory and cryptography π
Create a free account to open this chapter.
- Entropy and secrecy ~12 min
- Perfect secrecy, defined ~10 min
- The one-time pad ~13 min
- Why key length matters ~13 min
- Information leakage ~12 min
- Shannon's secrecy result ~11 min
15 Information theory, statistics, and machine learning π
Create a free account to open this chapter.
- Maximum likelihood and information ~14 min
- The maximum entropy principle ~15 min
- KL divergence as model fitting ~14 min
- Cross-entropy loss ~13 min
- Fisher information ~13 min
- The Cramer-Rao bound ~15 min
- Information gain and decision trees ~14 min
- Mutual information for feature selection ~12 min
- Softmax and temperature ~13 min
- The information bottleneck, a first look ~12 min
16 Advanced topics in information theory π
Create a free account to open this chapter.
- The data processing inequality, formalized ~10 min
- Fano's inequality ~13 min
- Pinsker's inequality ~12 min
- Variational inference and the ELBO, a preview ~11 min
- Network information theory: multiple-access and broadcast channels, a preview ~10 min
- Distributed source coding, a preview ~10 min
- Quantum information, a first look ~12 min
17 Capstone: the information transmission lab π
Create a free account to open this chapter.
- Designing a complete communication system ~14 min
- Trade-offs across source coding, channel coding, and rate ~15 min
- The capstone lab widget ~18 min