NSHM KNOWLEDEGE CAMPUS
DURGAPUR-GOI(College Code: 273)
CONTINUOUS ASSIGNMENT -1
PRESENTED BY
Student Name SHIVAM KUMAR RAI
University Roll No. 27300122006
University Registration No. 222730110109
Branch BACHELOR OF TECNOLOGY IN COMPUTER SCIENCE & ENGINNERING
Year 4TH
Semester 7TH
Paper Name MACHINE LEARNING
Paper Code PEC-CS701E
Conventional Learning &
Instance-Based Learning
Explore two fundamental machine learning approaches:
Conventional Learning and Instance-Based Learning. We will
examine their core principles, practical applications, and the trade-
offs between generalization and memorisation that define their
utility in AI and predictive modelling.
What is Conventional (Model-Based)
Learning?
Builds Generalised Model
Constructs a comprehensive model from training data, capturing underlying patterns
and relationships.
Algorithm-Driven
Employs statistical algorithms, such as linear regression and neural networks, to define
the model.
Efficient Prediction
Once trained, the model rapidly predicts outcomes for new, unseen data, showcasing its
efficiency.
Real-world Application
An exemplary use is predicting house prices based on features like size, location, and
number of rooms.
What is Instance-Based
Learning?
• Also known as memory-based or lazy learning, as it defers
processing until a query is made.
• Stores all training instances directly, without creating an explicit,
generalized model beforehand.
• Predicts by comparing new data points to stored examples,
identifying similarities.
• A prime example is the K-Nearest Neighbors (KNN) algorithm,
which classifies data based on its closest neighbours.
Advantages of Conventional
Learning
Faster Prediction Enhanced Generalisation
Once the model is trained, Provides superior performance on
predictions on new data are unseen data, as it learns underlying
exceptionally quick. patterns.
Insightful Analysis Efficient Storage
Offers transparency into data Models are typically compact,
relationships and identifies crucial requiring less storage than raw
feature importance. training data.
Advantages of Instance-Based Learning
Simplicity
Requires no complex model creation or deep expert knowledge
for implementation.
Adaptability
Flexible and quickly adjusts to new or small datasets, making it
highly versatile.
Pattern Recognition
Excels at capturing intricate, localised patterns without imposing
strict assumptions.
Anomaly Detection
Particularly effective in applications like pattern recognition and
identifying anomalies within data.
Disadvantages of Conventional Learning
Data Dependency Expertise Required
Requires vast datasets for effective training to Needs significant expert knowledge in statistical
prevent underfitting or bias. modelling for optimal model design and tuning.
Assumption Sensitivity Retraining Overhead
Model performance can suffer if underlying Frequent retraining is necessary to adapt to changes
assumptions are violated by the data distribution. in data patterns, known as model drift, which can be
resource-intensive.
Disadvantages of Instance-Based Learning
• High Storage Requirements: Stores all training
data, leading to substantial memory consumption,
especially for large datasets.
• Computationally Intensive: Predictions are slow
due to the need to compare each new instance
against every stored example.
• Poor Generalization: Struggles to generalize
effectively to unseen data distributions, as it relies on
direct similarity.
• Scalability Issues: Not suitable for very large
datasets due to the memory limits and computational
burden.
Summary & Comparison
Conventional Learning Instance-Based Learning
• Eager: Builds a model before • Lazy: Defers learning until
prediction. needed.
• Generalises: Through • Memorises: Stores raw data
abstract models. instances.
• Efficient: Rapid prediction • Flexible: Adapts well to new
once trained. data.
The choice between these paradigms hinges on dataset size, problem
complexity, and available resources. Both approaches are invaluable
and complement each other across various AI applications, offering
distinct benefits depending on the scenario.
