FAQ Coverage Gaps¶

Generated: 2025-12-29

This document identifies concepts from the learning graph that are not covered in the FAQ, prioritized by importance for student understanding.

Summary¶

• Total Concepts: 200
• Covered in FAQ: 146 (73%)
• Not Covered: 54 (27%)

Critical Gaps (High Priority)¶

These concepts have high centrality in the learning graph (many dependencies) or are fundamental to understanding covered topics. Recommendation: Add questions for all 15 critical gaps.

Convolutional Neural Networks (8 concepts)¶

1. VGG Architecture
2. Centrality: High (referenced in transfer learning)
3. Category: Advanced Topics

4. Suggested Question: "What are the key characteristics of VGG architecture and why was it significant?"

5. Inception Architecture

6. Centrality: High (multi-scale feature learning)
7. Category: Advanced Topics

8. Suggested Question: "How does the Inception architecture use multiple filter sizes in parallel?"

9. Depthwise Separable Convolution

10. Centrality: Medium (efficiency technique)
11. Category: Technical Details

12. Suggested Question: "What is a depthwise separable convolution and why does it reduce parameters?"

13. Dilated Convolution

14. Centrality: Medium (receptive field expansion)
15. Category: Technical Details

16. Suggested Question: "What is dilated convolution and when is it useful?"

17. Global Average Pooling

18. Centrality: High (modern CNN standard)
19. Category: Technical Details

20. Suggested Question: "What is global average pooling and why is it preferred over fully connected layers?"

21. Receptive Field

22. Centrality: High (fundamental concept)
23. Category: Core Concepts

24. Suggested Question: "What is the receptive field in a CNN and how does it grow through layers?"

25. Feature Pyramid

26. Centrality: Medium (multi-scale detection)
27. Category: Advanced Topics

28. Suggested Question: "What is a feature pyramid network and why is it useful for object detection?"

29. Spatial Pyramid Pooling

30. Centrality: Low (specialized technique)
31. Category: Advanced Topics
32. Suggested Question: "What is spatial pyramid pooling and how does it handle variable input sizes?"

Optimization (7 concepts)¶

1. Learning Rate Scheduling
2. Centrality: High (critical for training)
3. Category: Best Practices

4. Suggested Question: "What is learning rate scheduling and which schedules are most effective?"

5. Nesterov Momentum

   • Centrality: Medium (SGD variant)
   • Category: Technical Details
   • Suggested Question: "What is Nesterov momentum and how does it differ from standard momentum?"

6. RMSprop

   • Centrality: High (Adam predecessor)
   • Category: Technical Details
   • Suggested Question: "What is RMSprop and how does it adapt learning rates?"

7. Weight Decay

   • Centrality: High (regularization)
   • Category: Technical Details
   • Suggested Question: "What is weight decay and how is it related to L2 regularization?"

8. Gradient Accumulation

   • Centrality: Medium (memory efficiency)
   • Category: Best Practices
   • Suggested Question: "What is gradient accumulation and when should I use it?"

9. Learning Rate Warmup

   • Centrality: Medium (training stability)
   • Category: Best Practices
   • Suggested Question: "What is learning rate warmup and why does it help training?"

10. Gradient Noise

    • Centrality: Low (regularization technique)
    • Category: Advanced Topics
    • Suggested Question: "What is gradient noise and how can it improve generalization?"

Medium Priority Gaps¶

These concepts are moderately important and would enhance FAQ completeness. Recommendation: Add 8-10 questions from this list.

Neural Network Architectures (5 concepts)¶

1. Residual Connections

   • Suggested Question: "What are residual connections and why do they enable training of very deep networks?"

2. Skip Connections

   • Suggested Question: "How do skip connections help prevent vanishing gradients?"

3. Highway Networks

   • Suggested Question: "What are highway networks and how do they relate to ResNets?"

4. Layer Normalization

   • Suggested Question: "What is layer normalization and how does it differ from batch normalization?"

5. Attention Mechanism

   • Suggested Question: "What is an attention mechanism and how does it help neural networks focus on important features?"

Support Vector Machines (5 concepts)¶

1. Kernel Parameters (Gamma, C)

   • Suggested Question: "How do I choose the C and gamma parameters for SVM with RBF kernel?"

2. Nu-SVM

   • Suggested Question: "What is nu-SVM and how does it differ from C-SVM?"

3. One-Class SVM

   • Suggested Question: "What is one-class SVM and when should I use it for anomaly detection?"

4. SMO Algorithm

   • Suggested Question: "What is the SMO (Sequential Minimal Optimization) algorithm for training SVMs?"

5. Support Vector Details

   • Suggested Question: "What exactly are support vectors and why are they important?"

Data Preprocessing (3 concepts)¶

1. Outlier Detection

   • Suggested Question: "How do I detect and handle outliers in my dataset?"

2. Label Encoding vs One-Hot

   • Suggested Question: "When should I use label encoding vs one-hot encoding for categorical variables?"

3. Data Imputation Strategies

   • Suggested Question: "What are the best strategies for imputing missing values?"

Regularization (1 concept)¶

1. Elastic Net
   • Suggested Question: "What is Elastic Net and when should I use it instead of L1 or L2 regularization?"

Clustering (2 concepts)¶

1. Silhouette Score

   • Suggested Question: "What is the silhouette score and how does it help evaluate clustering quality?"

2. Dendrogram

   • Suggested Question: "What is a dendrogram and how is it used in hierarchical clustering?"

Evaluation Metrics (4 concepts)¶

1. Balanced Accuracy

   • Suggested Question: "What is balanced accuracy and when should I use it instead of regular accuracy?"

2. Matthews Correlation Coefficient

   • Suggested Question: "What is the Matthews Correlation Coefficient and why is it good for imbalanced datasets?"

3. Cohen's Kappa

   • Suggested Question: "What is Cohen's Kappa and how does it measure inter-rater agreement?"

4. Mean Average Precision

   • Suggested Question: "What is mean average precision (mAP) and how is it used in object detection?"

Transfer Learning (2 concepts)¶

1. Domain Adaptation

   • Suggested Question: "What is domain adaptation and how does it help transfer learning across different domains?"

2. Model Zoo

   • Suggested Question: "What is a model zoo and where can I find pre-trained models?"

Foundation Concepts (3 concepts)¶

1. Parametric vs Non-Parametric

   • Suggested Question: "What is the difference between parametric and non-parametric models?"

2. Instance-Based Learning

   • Suggested Question: "What is instance-based learning and how does it differ from model-based learning?"

3. Online Learning

   • Suggested Question: "What is online learning and when is it preferred over batch learning?"

Low Priority Gaps¶

These are advanced, specialized, or leaf-node concepts that can be addressed in future updates. Recommendation: Address selectively based on user demand.

Specialized CNN Concepts (5 concepts)¶

1. Object Detection - Domain-specific application
2. Semantic Segmentation - Domain-specific application
3. Instance Segmentation - Domain-specific application
4. Anchor Boxes - Object detection specific
5. Region Proposals - Object detection specific

Advanced Decision Trees (1 concept)¶

1. Cost Complexity Pruning - Specialized pruning technique

Advanced Regularization (1 concept)¶

1. Batch Renormalization - Variant of batch norm

Advanced Neural Networks (7 concepts)¶

1. Gradient Checkpointing - Memory optimization
2. Mixed Precision Training - Hardware optimization
3. Distributed Training - Scale-out technique
4. Model Parallelism - Large model training
5. Data Parallelism - Batch parallelization
6. Knowledge Distillation - Model compression
7. Neural Architecture Search - Automated design

Implementation Recommendations¶

Phase 1: Critical Gaps (Week 1)¶

Add 15 questions covering all critical gaps, focusing on: - CNN architectures (VGG, Inception, receptive field, global average pooling) - Optimization techniques (learning rate scheduling, RMSprop, weight decay, Nesterov momentum)

Expected Impact: Increase coverage from 73% to 81%

Phase 2: Medium Priority (Weeks 2-3)¶

Add 10 questions from medium priority list, focusing on: - Residual/skip connections - SVM kernel tuning - Advanced preprocessing - Additional evaluation metrics

Expected Impact: Increase coverage from 81% to 86%

Phase 3: Selective Low Priority (Week 4+)¶

Add 3-5 questions based on user feedback and most requested topics

Expected Impact: Increase coverage from 86% to 89%

Coverage by Category After Implementation¶

Questions with Highest Student Demand¶

Based on typical student inquiries in machine learning courses:

1. Architecture Choice: "How do I choose between different CNN architectures?"
2. Learning Rate: "What learning rate schedule should I use?"
3. Kernel Tuning: "How do I tune SVM kernel parameters?"
4. Residual Networks: "Why do residual connections help so much?"
5. RMSprop vs Adam: "When should I use RMSprop instead of Adam?"
6. Outliers: "How do I handle outliers in my data?"
7. One-Class SVM: "How do I use SVM for anomaly detection?"
8. Global Average Pooling: "Why use global average pooling instead of flatten?"
9. Weight Decay: "What's the relationship between weight decay and L2 regularization?"
10. Receptive Field: "How do I calculate receptive field in my CNN?"

Conclusion¶

The FAQ covers 73% of concepts with strong coverage of fundamentals (90% foundation concepts) and complete coverage of some areas (100% KNN). The 27% gap consists primarily of:

1. Advanced CNN architectures (8 concepts) - Critical for modern computer vision
2. Optimization techniques (7 concepts) - Critical for effective training
3. Advanced neural network concepts (5 concepts) - Important for deep learning
4. SVM tuning and variants (5 concepts) - Important for practical SVM use

Adding the 15 critical gap questions would raise coverage to 81%, bringing the FAQ to "very good" completeness. The remaining gaps are primarily specialized techniques that can be addressed based on user demand.

Priority Action: Implement Phase 1 (15 questions on critical gaps) within 1 week to achieve 81% coverage.