CBR 2

CASE INDEXING:

**Case Components in CBR:**

A case in Case-Based Reasoning (CBR) typically consists of three main components: the
problem or scene, the solution, and the result. The problem or scene describes the problem to
be solved and the states of the world during the case. The solution details how the problem is
addressed under specific circumstances, and the result records the success of the solution's
implementation and the post-solution situation. While the problem and solution are essential, the
result is optional but valuable for learning from both successes and failures.

**Principles of Case Indexing:**

Case indexing is crucial for efficient retrieval of relevant cases. Three principles guide case
indexing:
1. **Relevance to the Domain:** Indices should be specific to the concrete domain, ensuring
effective division of the case base and faster retrieval. Unlike general database indices, CBR
system indices must be predictive and concrete, simplifying access and retrieval based on the
nature of the domain.

2. **Predictive and Abstract Indices:** While indices should be predictive and concrete for
current use, they also need to be abstract enough to accommodate future applications of the
case base. Striking a balance between predictability and abstraction ensures versatility in the
future use of the case base.

3. **Recognition and Concreteness:** Indices should be concrete enough to be recognizable in

the future. Overly abstract indices risk blurring differences among cases, hindering effective
recognition and retrieval. A balance is needed to maintain concreteness without sacrificing the
ability to distinguish between cases.
Efficient case indexing enhances the utility of CBR systems by facilitating quick and accurate
retrieval of cases relevant to new problems, supporting the system's learning and
problem-solving capabilities.

CASE RETREIVAL

**Case Retrieval Process:**

Case retrieval in Case-Based Reasoning (CBR) is the process of finding the best match in the
case base for the current problem, utilizing knowledge gained from past solved cases. The
retrieval process involves three phases: feature identification, tentative matching, and final
selection.

- **Feature Identification:**
- Identify relevant features based on problem analysis.
- Extract features directly from the problem description, elicit features post-understanding, or
obtain features from users through human-machine interaction.
- Features play a crucial role in directing the search for similar cases.

- **Tentative Matching:**
- Find a group of candidate cases from the case base using identified features.
- Rely on indices and perform partial matches, constructing a partial order based on
similarities.
- Similarities can be computed through syntax structures or a more in-depth analysis.
- Methods include nearest neighbor, induction, knowledge-guided induction, and template
retrieval.

- **Final Selection:**
- Choose one or several cases with the highest similarity to the target case from the tentative
matches.
- Domain knowledge plays a key role in explanations provided by knowledge engineers or
computed based on the knowledge model.
- Evaluate explanations and arrange candidates based on criteria, selecting the most relevant
or rational match.

**Case Retrieval Algorithms:**

- Different organizational forms exist for case bases, including serial or parallel, flat or layered,
and small or large granularity indices.
- The mostly-used methods are based on the inverted index, enabling serial or parallel
searches. Kitano's application system handles over 25,000 cases.

**Common Retrieval Methods:**

1. **Nearest Neighbor:**
- Assesses similarity based on a weighted sum of features.
- Effective for smaller case bases due to linear time complexity.
- Key challenge lies in determining feature weights.

2. **Induction:**
- Determines features that best discriminate cases.
 - Generates a decision tree for effective case organization.
- Useful when a single case feature is required, dependent upon others.

3. **Template Retrieval:**
- Similar to SQL-like queries, returning cases within specified parameters.
- Limits the search space to a relevant section of the case base.
- Applied before other techniques like nearest neighbor.

These retrieval methods play a crucial role in leveraging past cases to influence the quality of
current problem-solving in CBR systems.

SIMILARATY RELATIONS IN CBR:

**5.7 Similarity Relations in CBR:**

**Introduction:**
Case retrieval in Case-Based Reasoning (CBR) relies on defining and assessing the similarity
among cases. The success of case retrieval is pivotal for the effective application of CBR,
making similarity a core concept in this reasoning paradigm.

**5.7.1 Semantic Similarity:**

- Analogical cases exhibit semantic similarity, where positive, negative, and uncertain analogies
can be identified.
- Conditions for analogical cases include essential property and causality not being part of
negative analogy.
- Two types of similarity are defined: surface similarity (non-essential aspects) and structural
similarity (affecting the adaptation process).
- Both types play distinct roles in analogy processes, aiding tentative analogy, individual
identifications, analogical search, and mapping construction.

**5.7.2 Structural Similarity:**

- Structures are isomorphic if a mapping between them preserves structural consistency.
- Isomorphism significantly influences the validity of analogical reasoning.
- Structural similarities are crucial for similarity-based retrievals, allowing for analogy even in
cases with dissimilar surface similarities.
- Isomorphism and similarities among substructures help focus on local similarities in dissimilar
cases.

**5.7.3 Goal’s Features:**

- Goal features enhance semantic similarity and structural consistency, directing the
problem-solving process.
- Cases with goal relevance gain higher priority for consideration.
- Keane's analogical retrieval model emphasizes analyzing goal features (structural index) in
base cases but warns against overemphasizing them to avoid filtering out potential useful
cases.

**5.7.4 Individual Similarity:**

- Emphasis on restraints from classification information about individuals.
- Individuals with similarities belong to the same classification, aiding concept clustering.
- Analogy among individuals helps identify primary roles in problem-solving, with partial
solutions guiding the search for whole solutions.

In summary, these similarity relations, including semantic, structural, goal-based, and individual
similarities, form the foundation for successful case retrieval in CBR, influencing the
effectiveness of applying past cases to solve current problems.

CASE REUSE:

**5.8 Case Reuse:**

**Introduction:**
Case reuse in Case-Based Reasoning (CBR) involves leveraging past cases to solve current
problems. It can be categorized into two types based on the information reused: result reuse
and method reuse.

**Result Reuse:**
- Directly applies adaptation to the solution stored in cases.
- Identifies prominent differences between base and target cases and applies adaptation rules.
- Reuses solutions stored in cases.

**Method Reuse:**
- Reuses algorithms or rules employed in problem-solving from base cases.
- Produces a new solution for the target case.
- Adaptation methods depend on concrete problems.

**Case Revise:**
- Occurs when the proposed solution is deemed unsatisfactory.
- Evaluation of the proposed solution is necessary before initiating the revise process.
- Evaluation can be based on feedback after solution application or consultation with domain
experts.
- Revision aims to output a better solution based on the current problem description and a
suggested solution obtained in the reuse process.

**Revision Methods:**
1. **Substitution:**
- Structural adaptation technique involving component substitutions.
- Includes reinstantiation, parameter adjustment, local search, query, specialized search, and
case-based substitution.

2. **Transformation:**
- Involves common-sense transformation and model-guided repair.
- Common-sense transformation replaces, deletes, or increases components based on basic
understanding.
- Model-guided repair uses a causal model to guide adaptation, often employed in diagnosis
systems.
3. **Special-purpose Adaptation and Repair:**
- Domain-dependent method used in structural adaptations.
- Utilizes heuristics to index potential useful cases.
- Heuristics give evaluations on approximate solutions and are controlled by rule-based
production systems.

4. **Derivational Replay:**
- Focuses on how solutions are derived.
- Derives a solution in a new situation based on the method of deriving an old solution or
solution piece.
- A case-based revise method compared with case-based substitution methods.

In summary, case reuse involves leveraging past solutions or methods for current
problem-solving. Case revision processes evaluate and improve proposed solutions, employing
various methods such as substitution, transformation, special-purpose adaptation, and
derivational replay to enhance the adaptation process.

CASE RETAINTION:

**5.9 Case Retention:**

**Introduction:**
Case retention in Case-Based Reasoning (CBR) is akin to case retrieval but involves storing
new cases in the case base. It encompasses both the storage and retrieval aspects of
information. This process contributes to learning and knowledge acquisition, determining what
information to retain, and how to organize, index, and manage the case base.

**Storage and Retrieval:**

- Incorporating a new case involves the index mechanism, assigning indices for future retrieval,
and insertion algorithms to place the case appropriately in the case base.
- Search in case retention mirrors that in case retrieval, where insertion algorithms find a spot for
the new case, and retrieval algorithms return the most similar cases.

**Learning and Knowledge Acquisition:**

- Case retention is part of learning and knowledge acquisition, focusing on what information to
retain and how to integrate it into the case base.
- Considerations for retention include feature descriptions, the problem solution, and
explanations of success.

**Indexing and Adjustment:**

- New cases receive additional indices for efficient retrieval.
- Adjustments to index contents or case base structure may be necessary to enhance retrieval,
such as changes in index intensity or feature weights.

**Challenges with Growing Case Base:**

- As the case base expands, it poses challenges like increased memory usage and retrieval
time.
- Effective organization and management of the case base are essential to mitigate resource
wastage.

**Four-R Process in CBR:**

- Retrieval, reuse, revise, and retention constitute the Four-R process in a CBR cycle,
emphasizing the central stages in CBR reasoning.

INFLUENCE BASED LEARNING:

**5.10.1 Learning Tasks of IBL:**

**Instance Representation:**
- Instances in Instance-Based Learning (IBL) are described as feature-value pairs in an
n-dimensional instance space.
- Each instance has n features, with one corresponding to the category feature and the rest
termed predictor features.
- IBL algorithms focus on learning concepts involving one category feature, leading to disjoint
categories.

**Output and Concept Description:**

- IBL outputs a concept description, functioning as a mapping from instances to categories.
- The classification function processes similarity results and past performance records to provide
classifications for instances.
- A concept description updater maintains performance records and decides which instances to
include in the concept description.

**Classification Process:**
- Classification relies on the assumption that similar instances share similar classifications.
- In the absence of prior knowledge, IBL assumes equal contributions from all features, requiring
normalization of feature value domains.

**No Explicit Abstractions:**

- Unlike many supervised learning algorithms, IBL doesn't employ explicit abstractions like
decision trees.
- IBL focuses on calculations of similarities during classification, avoiding storing abstract
representations.

**Performance Assessment:**
- Generalization capability assesses IBL's ability to describe concepts and its learning power.
- Classification accuracy, learning rate, incorporation costs (updating concept descriptions), and
storage requirements are key performance metrics.

**Storage Requirement:**
- IBL algorithms' storage requirement is measured by the size of concept descriptions, indicating
the number of saved instances used for classification decisions.

FORCAST SYSTEM FOR CENTRAL FISHING GROUND:

**5.11 Forecast System for Central Fishing Ground:**

**Background and Decision-Making:**
- Human decision-making is influenced by past experiences, but limitations in memory hinder
recalling appropriate cases, especially with a vast number of cases.
- Combining human decision-making strengths with computer memory advantages is the focus
of case-based decision aiding.

**Utilizing Technology in Marine Resources Management:**

- Global attention is directed towards utilizing modern science and technology for developing
marine resources and ensuring sustainable sea fishery.
- Technologies such as computer systems, remote sensing, automation, and geographical
information systems play a crucial role in macroscopic decision-making and microcosmic
resource use in marine fishery.

**Research and Development Under 863 Program:**

- The 863 Program supports research on remote sensing information and resource assessment
systems for sea fishery.
- System integration and case illustration are emphasized to enhance resource utilization and
fishery development in the marine exclusive economic zone.

**Demonstration Area:**
- The China East Sea (north latitude 25 to 34 degrees, east longitude 130) serves as a
demonstration area for fishing ground prediction.
- Focus on technologies like data mining, Case-Based Learning (CBL), expert systems, and
remote sensing analysis for fishery, supported by geographical information system technology.

**Learning and Decision-Support System:**

- Integration of technologies, including data mining, CBL, expert systems, and geographical
information systems, forms a Learning and Decision-support System for fishing ground analysis
and prediction.
- Expert knowledge and experiences complement technological inputs, strengthening the
analytical and predictive capabilities for effective fisheries management.
CBR 1 TOPIC WISE

INTRODUCTION

**Case-Based Reasoning (CBR):**

CBR is an Artificial Intelligence approach that tackles new problems by leveraging knowledge
from previously solved, similar problems. Cases, representing contextualized experiences, form
the basis for reasoning in CBR.

**Characteristics of a Case:**
1. *Implicit Concrete Knowledge:* Cases contain implicit knowledge related to specific contexts,
typically focusing on how to perform certain tasks.
2. *Diverse Forms and Granularities:* Cases come in various forms and granularities,
incorporating different time slices and the effects of solution application.
3. *Useful Experiences:* Cases encode experiences beneficial for reasoning, aiding in goal
attainment or predicting potential failures.
**CBR in Artificial Intelligence:**
CBR deviates from traditional "cause-effect" chain-based reasoning by implicitly storing
knowledge in cases rather than explicit rules. It excels in adapting past successful solutions to
solve analogous problems.

**Realism in CBR:**
CBR aligns with the realities of regulation and repetition in the world. Similar actions under
comparable conditions often yield parallel results, reflecting the idea that past experiences can
provide insights into the future.

**Contributions of CBR to AI:**

1. *Knowledge Acquisition:* CBR streamlines the challenging process of knowledge acquisition
by avoiding the need for an explicit domain model. It relies on cases, reducing the knowledge
acquisition bottleneck.
2. *Knowledge Maintenance:* Unlike traditional systems requiring complex knowledge updates,
CBR systems learn by acquiring new cases, simplifying maintenance and conflict resolution.
3. *Efficient Problem Solving:* CBR enhances problem-solving efficiency by reusing previous
solutions, offering a more efficient approach compared to starting from scratch.
4. *Quality Improvement in Problem Solving:* The records of previous failures aid in avoiding
similar failures, enhancing the overall quality of problem-solving outcomes.
5. *User Acceptance:* CBR systems, grounded in historical facts, provide more convincing
conclusions, leading to improved user acceptance.

**CBR Research and Applications:**

The Key Laboratory of the Institute of Intelligent Information Processing has conducted
extensive CBR research. Examples include the development of a memory network model,
CBR-based design systems for internal-combustion engine oil products, weather forecast
systems, and prediction and deployment systems for dams.

In summary, CBR's reliance on experiential knowledge stored in cases has proven valuable in
overcoming challenges associated with knowledge acquisition, maintenance, and
problem-solving efficiency in the field of Artificial Intelligence.

BASIC NOTATION:

**Analogical Problem Solving:**

Analogical problem solving involves finding solutions to new problems by drawing analogies
with recalled solutions from similar old problems. This process incorporates reasoning to
generate new knowledge and adjust solutions based on past experiences.

**Learning Skills in Computation Models:**

Beyond recording solutions to similar problems, a computation model needs learning skills to
adjust solutions based on past experiences. This adaptive learning allows for the generation of
new knowledge through reasoning.

**Formalization of Analogical Problem Solving:**

Analogical problem solving is formalized by considering a solution \(B\) to problem \(A\). Given a
new problem \(A'\) similar to \(A\), the objective is to determine its solution \(B'\). The relationship
\(β\) represents causality between \(B\) and \(A\), while \(α\) signifies the similarity between
source domain \(A\) and target domain \(A'\). The dependence \(β'\) between \(B'\) and \(A'\) is
computed based on \(β\) and \(α\).

**Definitions in Analogy Learning:**

1. *Causality (β):* Represents the dependence between the solution \(B\) and the problem \(A\).
2. *Similarity (α):* Measures the similarity between the source domain \(A\) and the target
domain \(A'\).
3. *Dependence (\(β'\)):* Computed based on \(β\) and \(α\), representing the relationship
between the solution \(B'\) and the new problem \(A'\).

Analogical problem solving, as a form of learning based on knowledge or experience, enables

the computation model to adapt solutions from similar old problems to address new challenges.

PROCESS MODEL:

**Case-Based Reasoning (CBR): An Overview**

CBR, a form of analogical reasoning, retrieves similar cases from memory when faced with a
new problem. The process involves refining retrieved cases to find the best match and
ultimately generating a solution for the new problem.

**Stages of CBR:**

1. **Case Retrieval:**
- Retrieval of similar cases from memory based on certain properties of the target case.
- Preliminary exploration of potential base cases involves local analogical mapping.

2. **Analogical Mapping:**
- Arrangement of potential base cases in a partial order according to similarity.
- Selection of the most similar base case for a one-to-one mapping with the target case.

3. **Analogical Transformation:**
- Construction of a complete or partial solution for the new problem using the mapping and the
base case's solution.
- Reuse of information from the base case, categorized into result reuse and method reuse.

4. **Explanation:**
- Explanation of any failure in the transformation process.
- Cause-and-effect analysis report for successful cases.

5. **Case Revision:**
- Adaptation of a proposed solution to prevent factors leading to failures.
- Evaluation of the proposed solution before initiating the revise process.

6. **Analogy Verification:**
- Validation of the analogy mapping between base cases and the target case.
7. **Case Retention:**
- Addition of the newly-generated solution to the case base for future use.
- Knowledge acquisition through retention involves choosing relevant information and refining
base cases.

8. **Considerations in CBR:**
- **Case Representation:** Efficient case representation is crucial for system performance.
- **Analysis Model:** Used to analyze the target case and elicit information for finding the best
match.
- **Case Retrieval:** Finding potentially useful cases and determining the best match.

**Conclusion:**
CBR, categorized into problem-solving and interpretive types, offers a practical approach to
reasoning based on past experiences. The success of CBR systems relies on effective case
representation, analysis models, and accurate case retrieval. The iterative nature of CBR,
involving explanation, revision, and retention, facilitates continuous learning and adaptation for
improved problem-solving efficiency and quality.

CASE REPRESENTATION:

**Memory Mechanism for Knowledge: An Overview**

Understanding the brain's memory mechanism for knowledge remains an open problem, leading
to various knowledge representation methods in knowledge-based systems. Traditional
methods like production rules, semantic networks, frames, and object-oriented representation
may not be suitable for learning systems, especially in analogically learning systems.

**Psychological Models of Memory:**

Research in physiology and psychology has explored memory extensively. Models such as
episodic memory, semantic memory, associative memory, and Schank's dynamic memory
theory offer conceptual frameworks for understanding memory.

**Semantic Memory Unit (SMU):**

Semantic Memory Units are concepts, modes, and themes involved in knowledge study,
analysis, understanding, and memory. These units serve as centers for linking concrete
knowledge and problems in an organized manner. Factors chosen as SMUs depend on the
characteristics of the knowledge.

**Memory Network:**
The Memory Network summarizes the interrelations among Semantic Memory Units (SMUs). It
involves various slots:
- SMU_NAME slot: Describes the conceptual SMU.
- Constraint slots (CON): Provide restraints on the SMU.
- Taxonomy slots (TAX): Define relationships in the taxonomy.
- Causality slots (CAU): Define cause-and-effect connections.
- Similarity slots (SIM): Define similar SMUs.
- Partonomy slots (PAR): Define whole-part relations.
- Case slots (CAS): Define relevant cases.
- Theory slots (THY): Define theories about the SMU.

The network ensures easy retrieval of knowledge, incorporates semantic relations among
nodes, and is complex yet reflective of intricate inner relations among different types of
knowledge.

**Learning and Forgetfulness in Memory Network:**

The construction of the Memory Network is a complex learning process. It accounts for
forgetfulness, with memory intensity representing the strength of memory over time. The
network is a long-term memory, while memory in slots is short-term.

**Comparison with Semantic Network:**

While the Memory Network shares similarities with the semantic network, it surpasses it by:
- Storing and using theories and cases in different representation methods.
- Storing special knowledge by adding restraints to nodes.
- Organizing similar knowledge using intension nodes.
- Having an agent as a memory unit capable of independent task completion.

**Reasoning in Memory Network:**

Reasoning processes, such as inheritance reasoning, restraints satisfiability, and forward- and
backward reasoning, can be applied based on the memory network. Case-Based Reasoning
(CBR) methods can be adopted for knowledge stored in Case slots (CAS).

In summary, the Memory Network provides a comprehensive and flexible model for knowledge
representation, retrieval, and learning, incorporating various types of information and supporting
diverse reasoning processes.