mirror of
https://github.com/harvard-edge/cs249r_book.git
synced 2026-05-05 17:18:48 -05:00
61 lines
1.7 KiB
Plaintext
61 lines
1.7 KiB
Plaintext
# On-Device Learning
|
|
|
|
::: {.callout-note collapse="true"}
|
|
## Learning Objectives
|
|
|
|
* coming soon.
|
|
|
|
:::
|
|
|
|
## Introduction
|
|
|
|
Explanation: This section sets the stage for the reader, explaining why on-device learning is a critical aspect of embedded AI systems.
|
|
|
|
- Importance in Embedded AI
|
|
- Why is On-device Learning Needed
|
|
|
|
## Advantages and Limitations
|
|
|
|
Explanation: Understanding the pros and cons of on-device learning helps to identify the scenarios where it is most effective and the challenges that need to be addressed.
|
|
|
|
- Benefits
|
|
- Constraints
|
|
|
|
## Continuous Learning
|
|
|
|
Explanation: Continuous learning is essential for embedded systems to adapt to new data and situations without requiring frequent updates from a central server.
|
|
|
|
- Incremental Algorithms
|
|
- Adaptability
|
|
|
|
## Federated Machine Learning
|
|
|
|
Explanation: Federated learning allows multiple devices to collaborate in model training without sharing raw data, which is highly relevant for embedded systems concerned with data privacy.
|
|
|
|
- Architecture
|
|
- Optimization
|
|
|
|
## Transfer Learning
|
|
|
|
Explanation: Transfer learning enables a pre-trained model to adapt to new tasks with less data, which is beneficial for embedded systems where data might be scarce.
|
|
|
|
- Use Cases
|
|
- Benefits
|
|
|
|
## Data Augmentation
|
|
|
|
Explanation: Data augmentation can enrich the training set, improving model performance, which is particularly useful when data is limited in embedded systems.
|
|
|
|
- Techniques
|
|
- Role in On-Device Learning
|
|
|
|
## Security Concerns
|
|
|
|
Explanation: Security is a significant concern for any system that performs learning on-device, as it may expose vulnerabilities.
|
|
|
|
- Risks
|
|
- Mitigation
|
|
|
|
## Conclusion
|
|
|
|
- Key Takeaways |