Global optimum
A global optimum is a feasible point or solution whose objective value is at least as favorable as that of every other point or solution in the domain.
Grokking AI Algorithms, Second Edition
A global minimum is the lowest value of an objective function over its entire domain.
- How it is used
- Distinguishes the best solution across an optimization search space from a merely local best.
- How deep it goes
- Defines the distinction and works through an intuitive minimum; does not establish a general way to prove an algorithm found one.
- Examples
- Uses restaurant rankings, pepper-spiciness data, the Booth function, and a drone-material fitness function.
- What you can do after
- Evaluate/understand whether a reported best is local, global, or simply the best solution found so far.
Read how Grokking AI Algorithms, Second Edition explains global optimum →
Data Without Labels
A global minimum is a point whose objective value is no greater than that of every point in the domain.
- How it is used
- Frames optimization as minimizing error or loss while maximizing predictive accuracy.
- How deep it goes
- Explains global and local minima through iterative gradient descent, random initialization, and convergence behavior.
- Examples
- Uses a mountain descent analogy, gradient-descent loss minimization, and repeated k-means runs as concrete illustrations.
- What you can do after
- Diagnose/debug with it by recognizing when optimization may have stopped at a local rather than global minimum.
Natural Language Processing in Action, Second Edition
A global minimum is a parameter setting with the lowest objective value over the entire domain under consideration.
- How it is used
- Explains what neural-network optimization seeks when adjusting weights to reduce training error.
- How deep it goes
- Defines the optimum through error surfaces, then distinguishes it from local minima and discusses stochastic search strategies.
- Examples
- Uses skiing, convex bowls, mean squared error, training batches, and neural-network initialization to illustrate optimization paths.
- What you can do after
- Apply stochastic gradient descent by reshuffling examples and tuning the learning rate while recognizing local-minimum limitations.
Read how Natural Language Processing in Action, Second Edition explains global optimum →
Optimization Algorithms: AI techniques for design, planning, and control problems
A feasible point whose objective-function value is at least as high as that of every candidate in the domain.
- How it is used
- Frames optimization as the search for the best feasible objective value across a candidate space.
- How deep it goes
- Defines global and local minima mathematically, then stops short of prescribing a guaranteed solving algorithm.
- Examples
- Uses a feasible landscape containing global, strong local, and weak local minima, alongside local and random search behavior.
- What you can do after
- Evaluate/understand whether a candidate is globally best, locally best, merely feasible, or near-optimal.
Advanced Algorithms and Data Structures
A minimum is a point or value at which a function is no greater than the function values in the relevant neighborhood or domain.
- How it is used
- Tunes the number of Bloom filter hash functions to reduce the false-positive ratio.
- How deep it goes
- Derives an exact stationary-point formula using logarithms and derivatives, but skips the detailed second-derivative verification.
- Examples
- Uses the formula k = ln(2)·m/n and discusses rounding k to an integer.
- What you can do after
- Evaluate the optimal Bloom filter hash-function count from fixed element and storage capacities.
Read how Advanced Algorithms and Data Structures explains global optimum →
Related concepts
- locally optimal solutioncontrasts with · A locally optimal solution is a feasible point whose objective value is no worse than those of sufficiently nearby points, though a better solution may exist elsewhere.
- dynamic programmingproduced by · Dynamic programming is an algorithmic technique that solves problems by decomposing them into overlapping subproblems and storing or reusing their solutions.
- maximin methodused by · The maximin method selects the alternative whose worst possible payoff is greatest among all alternatives.
- search spacepart of · A search space is the set of possible states, choices, configurations, or candidate solutions that may be considered when exploring or solving a problem.
- successorequivalent to · A successor is the node with the smallest key greater than a specified node's key in an ordered tree.
- optimal solutionkind of · An optimal solution is a feasible solution that achieves the best objective value among the alternatives considered.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.