Beam Search: Fundamentals and Applications

By Fouad Sabry

What Is Beam Search

In the field of computer science, beam search refers to a heuristic search technique that investigates a graph by extending the node that appears to have the greatest potential among a restricted group. The memory requirements of best-first search can be reduced with the use of an optimization called beam search. The best-first search is a type of graph search that arranges all of the partial solutions (states) in some order determined by a heuristic. However, in beam search, only a certain number of the best partial solutions are maintained as candidates. This number is specified in advance. This means that the algorithm is greedy.

How You Will Benefit

(I) Insights, and validations about the following topics:

Chapter 1: Beam search

Chapter 2: Heuristic (computer science)

Chapter 3: Search algorithm

Chapter 4: Best-first search

Chapter 5: Greedy algorithm

Chapter 6: Breadth-first search

Chapter 7: Tree traversal

Chapter 8: Machine translation

Chapter 9: Neural machine translation

Chapter 10: Raj Reddy

(II) Answering the public top questions about beam search.

(III) Real world examples for the usage of beam search in many fields.

(IV) 17 appendices to explain, briefly, 266 emerging technologies in each industry to have 360-degree full understanding of beam search' technologies.

Who This Book Is For

Professionals, undergraduate and graduate students, enthusiasts, hobbyists, and those who want to go beyond basic knowledge or information for any kind of beam search.

• Language: English
• Publisher: One Billion Knowledgeable
• Release date: Jul 1, 2023

Book preview

Chapter 9: Beam search

In the field of computer science, beam search refers to a heuristic search technique that investigates a network by extending the node that seems to have the most potential within a restricted group. The memory requirements of best-first search may be reduced with the use of an optimization called beam search. The best-first search is a kind of graph search that arranges all of the partial solutions (states) in some order determined by a heuristic. However, in beam search, only a certain number of the best partial solutions are maintained as candidates. This number is specified in advance. Because of this, the algorithm is avaricious.

In 1977, Raj Reddy of Carnegie Mellon University came up with the concept that would later be known as beam search..

Beam search constructs its search tree using breadth-first search rather than depth-first search.

At each successive level of the hierarchy, It creates all of the states that will succeed the present level at this time, arranging them from least expensive to most expensive to perform the heuristic on them.

However, It can only hold a single number that has been chosen in advance, \beta , declare examples that are the greatest at each level (called the beam width).

Only those states will undergo expansion after that.

The wider the beam, the more intense the beam, the fewer states that remain are eliminated.

Having an unboundedly wide beam width, There is no state elimination, and the beam search is exactly the same as the breadth-first search.

Memory requirements for the search are determined by the beam width's upper and lower limits.

Due to the fact that a desired state could undergo trimming,, The completeness of a method, or the promise that it will finish with a solution, is something that beam search gives up, if there is such a thing).

Beam search is not the most effective method (that is,, There is no assurance that it will come up with the most appropriate answer.

The most common use of a beam search is to preserve tractability in huge computer systems that lack an adequate amount of memory to hold the whole search tree.

The beam search was finished when it was combined with the depth-first search, which led to the development of the beam stack search (BULB). The search algorithms that were developed as a result are anytime algorithms like beam search that immediately locate excellent but likely sub-optimal answers, then retrace and continue to find better solutions until they converge on an ideal solution.

In the context of a search inside the local area, we call local beam search a specific algorithm that begins selecting \beta randomly generated states and then, pertaining to each individual level of the search tree, it always considers \beta new states among all the possible successors of the current ones, until it achieves the desired result.

Considering that the local beam search often lands on the local maxima, a common solution is to choose the next \beta states in a random way, with a likelihood that is based on how heuristically the states are evaluated.

Stochastic beam search is the name given to this method of looking.

{End Chapter 9}

{End Chapter 1}

Chapter 3: Heuristic (computer science)

In the fields of computer science and mathematical optimization, heuristic (from Greek εὑρίσκω I find, discover) is a strategy that was developed to help solve problems more rapidly in situations when traditional approaches are insufficient for finding an approximative solution, or when traditional procedures are unable to locate any specific remedy.

This is accomplished via the use of optimality trading, completeness, accuracy, or accuracy in exchange for velocity.

To some extent, It is possible to see it as a short cut.

A heuristic function, which is usually referred to as simply a heuristic, is a function that rates the many possibilities in search algorithms at each branching stage depending on the information that is available in order to choose which branch to follow. For instance, it might be a close approximation of the precise answer.

The purpose of a heuristic is to generate a solution in a fair amount of time that is adequate for resolving the issue that is currently being considered. This solution may not be the best out of all the possible answers to this issue; instead, it could only be a close approximation of the correct answer. However, despite this, it is still important since locating it will not take an insurmountable amount of time.

Heuristics have the potential to create outcomes on their own, or they may be used with optimization algorithms in order to boost such algorithms' overall effectiveness (e.g., they may be used to generate good seed values).

Heuristics are the only viable option for a variety of complex optimization problems that need to be routinely solved in real-world applications as a result of results about NP-hardness in theoretical computer science. This is because NP-hardness makes it extremely difficult to find optimal solutions.

Since heuristics may be used even in circumstances in which there are no known algorithms, they are an essential part of both the area of artificial intelligence and the process of simulating human thought on a computer.

The following is a list of the trade-off factors that may be used to decide whether or not to employ a heuristic for the solution of a certain problem::

Optimality: Does the heuristic ensure that the optimal solution will be discovered in situations when there are several possible solutions to a given problem? Is it really vital to locate the most effective solution?

Completeness refers to whether or not the heuristic can identify all of the possible