Dempster–Shafer theory

Dempster–Shafer theory (DST), also known as the theory of belief functions or evidence theory, offers a generalized framework for reasoning under uncertainty. It stands apart from traditional probability theory by allowing for the explicit representation of ignorance. Introduced by Arthur P.

Dempster in the context of statistical inference, the theory was significantly developed by Glenn Shafer into a comprehensive mathematical theory of evidence. DST models epistemic uncertainty, which is uncertainty stemming from a lack of knowledge, by assigning belief to sets of hypotheses rather than just individual propositions. This allows for a more nuanced representation of our state of knowledge, acknowledging that we might be unsure about which specific outcome will occur, but still have some degree of belief in a broader set of possibilities.

The genesis of DST can be traced to Dempster's 1967 work on statistical spaces, where he proposed a method for dealing with statistical inference when observations are imprecise. Glenn Shafer, in his 1976 book 'A Mathematical Theory of Evidence,' significantly expanded and popularized the theory, framing it as a direct alternative and extension to Bayesian probability. Shafer's work introduced the concept of 'belief functions' and 'basic probability assignments' (BPAs), which form the core of the theory.

Since their foundational contributions, DST has undergone considerable development, with researchers proposing various rules for combining evidence, particularly to address issues arising from conflicting sources. Notable extensions include the Transferable Belief Model (TBM) and the theory of hints, each offering different perspectives and solutions to the challenges of evidential reasoning.

The primary significance of Dempster–Shafer theory lies in its robust ability to combine evidence from disparate and potentially conflicting sources, a common scenario in complex decision-making environments. Unlike probability theory, which requires a complete probability distribution over all possible outcomes, DST can operate with partial information and explicitly represent ignorance. This makes it particularly valuable in fields such as artificial intelligence, where systems must make decisions based on uncertain sensor data or incomplete knowledge bases.

It provides a more flexible and expressive way to model belief and uncertainty, enabling more sophisticated reasoning in expert systems, diagnostic tools, and information fusion applications. Its capacity to handle conflicting evidence without forcing premature conclusions is a key advantage.

DST operates by defining a 'frame of discernment,' which is the set of all possible mutually exclusive propositions. Belief is then assigned to subsets of this frame using a 'basic probability assignment' (BPA), denoted by 'm'. The BPA assigns a numerical value, m(A), to each subset A, representing the degree of belief that is committed exactly to A and not to any proper subset of A.

The sum of all m(A) over all subsets A must equal 1. A 'belief function' B(A) is then derived, representing the total belief committed to A and its subsets. When new evidence arrives, DST employs a 'rule of combination' to merge the new BPA with existing ones.

The most common is Dempster's rule, which aggregates evidence, discounts conflicting information, and normalizes the result to yield a new BPA. This process allows for iterative refinement of beliefs as more evidence becomes available.

The theoretical elegance of DST has led to its application in a wide array of domains. In engineering, it's used for fault diagnosis and reliability analysis. In environmental science, it aids in risk assessment and decision support systems.

The theory is also relevant in fields like medical diagnosis, where integrating diverse clinical data is crucial. Furthermore, DST has influenced the development of advanced AI systems, particularly in areas requiring robust reasoning under uncertainty, such as autonomous navigation and intelligent agents. Ongoing research continues to explore more sophisticated combination rules, particularly for handling highly conflicting evidence, and to integrate DST with other uncertainty management frameworks, aiming to further enhance its applicability and power.