Scribd
Upload
20 views

ML Lecture 2

Uploaded by

azhar34658
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
20 views

ML Lecture 2

Uploaded by

azhar34658
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 19

CS – Machine Learning

Muhammad Hamza
Lecturer, CS
GDC KTS, Haripur
Overview

◉ Batch (Offline) Machine Learning


◉ Online Machine Learning
◉ Instance Based Learning
◉ Model Based Learning

2
Batch (Offline) vs Online Machine Learning

3
1. Batch/Offline Machine Learning
Definition:
The model is trained on the entire dataset or large
batches of data at once. This means that the model does
not learn continuously from new data but rather in a
series of steps or updates. The model is only updated
when the entire batch of data is available.

4
1. Batch/Offline Machine Learning
Disadvantages:
• Lots of data
• Hardware Limitation
• Availability

5
2. Online Machine Learning

Definition:
Online Machine Learning is a method of machine learning
where the model learns incrementally, one data point at a
time, as new data becomes available. Unlike batch
learning, where the model is trained on the entire dataset
at once, online learning updates the model continuously
with each new piece of data.

6
2. Online Machine Learning

Key Features:
1. Incremental Learning: The model is updated
continuously as new data arrives, rather than waiting
for a large batch of data. This makes online learning
ideal for applications where data is generated in real-
time, such as sensor readings, user interactions, or
streaming data.
2. Real-Time Adaptation: Online learning allows models
to adapt quickly to changes in the data distribution. If
the underlying patterns in the data shift over time, the
model can adjust without needing to be retrained
from scratch. 7
2. Online Machine Learning

Key Features:
3. Lower Latency: Since the model is updated
incrementally, it can be deployed and updated with
minimal delay, making it suitable for time-sensitive
applications.
4. Efficiency: Online learning typically requires less
memory and computational power compared to batch
learning because it processes one data point at a time,
rather than the entire dataset.

8
Applications of Online Machine Learning:

1. Recommendation Systems: Where user preferences


change over time and the system needs to adapt
quickly to provide relevant recommendations.
2. Stock Market Predictions: Where the model needs to
update its predictions continuously based on new
market data.
3. Fraud Detection: Where the model needs to adapt to
new types of fraud as they are detected.

9
Disadvantages of Online Machine Learning:

1. Risk of Instability: Since the model updates frequently,


it may become unstable or overly sensitive to noise in
the data if not managed properly.
2. Complexity in Implementation: Implementing online
learning can be more complex due to the need for
continuous monitoring and updating.
3. Potential for Overfitting: The model might overfit to
recent data points if not properly regularized, leading
to poor generalization.

10
Instance Based vs Model Based Learning

11
Instance Based Learning

• Works by memorizing things.


• Instance-based learning, also known as memory-based
learning, is a type of machine learning where the
model memorizes the training instances rather than
constructing an explicit general model.
• The algorithm makes predictions by comparing new
instances with the instances stored in memory (the
training data) and uses a similarity measure to
determine the output.

12
Instance Based Learning
Key Characteristics
1. No Explicit Model: Instead of generating a model based on the
training data, instance-based algorithms keep all or most of the
training data and use it directly to make predictions.
2. Lazy Learning: Instance-based learning is often referred to as
"lazy learning" because it defers processing until a query is
made. There’s no significant training phase; the algorithm only
generalizes when it needs to make a prediction.
3. Similarity Measure: To make a prediction, the algorithm finds
the most similar stored instances to the new input instance. The
similarity is typically measured using distance metrics like
Euclidean distance.

13
Instance Based Learning
Examples
1. k-Nearest Neighbors (k-NN): One of the most
common instance-based learning algorithms, which
classifies a new instance based on the majority class
among its nearest neighbors.
2. Locally Weighted Learning: Where predictions are
made by fitting a model locally using instances that are
close to the query point.
3. Case-Based Reasoning: A method where a new
problem is solved by finding a similar past case and
reusing it.

14
Instance Based Learning

Advantages:
• Flexibility: Can adapt quickly to new data without retraining.
• Simplicity: Easy to implement and understand, particularly k-NN.
Disadvantages:
• Storage: Requires storing all training data, which can be
inefficient in terms of memory.
• Speed: Making predictions can be slow for large datasets since
the algorithm needs to compare the new instance with many (or
all) stored instances.
• Noise Sensitivity: Prone to overfitting and affected by noisy data
since it memorizes all instances, including outliers.

15
Model based Learning

Model-based learning is a way of learning where the system creates


and uses an internal model to understand how things work in the
environment. This model helps the system predict what will happen
next and plan its actions accordingly.

16
Advantages of Model based Learning

Better Planning: The system can think ahead by predicting future


outcomes, leading to smarter decisions.
Fewer Trials Needed: Because it can simulate different scenarios, it
doesn’t need to try as many things in the real world, saving time
and resources.
Adaptability: If the environment changes, the system can update its
model and quickly adjust its actions.

17
Disadvantages of Model based
Learning
Complex to Build: Creating an accurate model can be hard and
requires a lot of effort.
Depends on Accuracy: If the model is wrong or incomplete, the
system may make bad decisions.
High Computational Cost: Running simulations and updating the
model can be computationally expensive, especially in complex
environments.

18
Thank You

19

You might also like