You are on page 1of 4

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/333842103

Machine Learning and Deep Learning: Thesis Defense related Basic Concepts,
Suggestions, and Potential Questions

Method · June 2019

CITATIONS READS

0 1,663

1 author:

Asifullah Khan
Pakistan Institute of Engineering and Applied Sciences
239 PUBLICATIONS 4,823 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Wind Power Prediction using Machine Learning Techniques View project

Controller Design for a Fractional-Order Nonlinear Glucose-Insulin System View project

All content following this page was uploaded by Asifullah Khan on 18 June 2019.

The user has requested enhancement of the downloaded file.


Machine Learning and Deep Learning: Thesis Defense related
Basic Concepts, Suggestions, and Potential Questions
Asifullah Khan, PIEAS, Pakistan.
Basic Concepts and Requirements:

You Need to Know and provide in your Thesis defense:

Proposed methodology
Related Mathematics
Experimental comparisons with Recent Techniques.
Performance Metrics
Statistical tests. Variation in results?
Parameter Tuning?
Cross validation technique?
Number of classes?
N,l,C ? Sample size, feature size, no. of classes
Generalization of the learner?
 You also need to know how you properly loaded data, clean the data, extracted
features, and finally - performed machine learning.
 And to know about the following 3 things of your ML system.
Data
Learning Methodology
Model
 And three tradeoffs of a learner:
Its capacity
Size of training data
Generalization

Suggestions for an ML/Deep-Learning Presentation:

Provide details of:

1. Dataset
2. Proposed methodology: basic and detailed block-diagrams.
3. Feature extraction, selection. Its visualization.
4. Performance metrics.
5. ANN Architectures
6. Implementations details.
7. Results: ROC and PR curves, tables, Statistics, Learning Curves, etc.
8. Experimental comparisons with at least 3 recently Reported techniques.
9. Show screenshots of your web predictor, if any
10. Show screenshots uploaded GitHub code, its associated Jupiter tutorial, and
your paper.
11. Conclusion and future work.
1
Potential Questions during Thesis Defense:
Basic ML Related Concepts:
Basic steps of an ML system?
Difference b/w ML and Pattern Recognition.
Difference between discriminator and generator?
Representation capacity?
Discriminant Function. LDA. Perception.
VC dimension.
No Free Lunch.
Occam’s Razer.
Probably Approximately Correct (PAC) Learning
PCA vs ICA
Adaptive vs non-adaptive Learning?
Mani-fold learning.
Cross validation; 5-fold, leave-one-out
Batch size and its affect.
Hyper parameters and its tuning.
Structural vs Empirical Risk Minimization.
Preprocessing.
SVM discriminant function and cost function.
Support Vectors? Number of SVs
PCA equation. Is it feature extractor OR selector OR both?

Distributions of train and test data


How to know that data is linearly separable or not?
Data visualization?
How to know that the classifier is learning properly; Learning curves
Difference in Filter and wrapper.
Should Feature selection come first or a sampling technique.
mRMR feature selection.
If two features are correlated then what?
Missing values. How to deal with?
How to start dealing with an ML problem; Designing ML system
ROC of multi classification problem?
Bias Variance dilemma?
How to compute shattering capacity of a classifier?
Fisher score.
lp norm?
Statistical test? Paired t-test, McNemar test, z- test.
How you compute ROC curve? by varying decision threshold?
Keep in mind answers for why you achieved such good results.

N/l ratio of your current problem


If data is less, what to do.
If data is large, what to do.
Clustering indices. Mathematical equation. Intra and inter distances.

2
Euclidean and Mahalanobis distance equation.
Different cost functions equations; MSE, PAD, Entropy loss, Soft max?
Information Gain and its equation.
Slope and interface part in learning.

Perception vs MLP learning


Learning-strategy/optimizer; GD, SGD, ADAM, RMSPROP, etc.
Drop-out, Gradient Descent, Vanishing gradient.
Batch Normalization and its affect.
Data Augmentation. Can GAN perform data Augmentation and how?
Momentum and Nestroves momentum
Convolution vs correlation. Equation.
Activation function. Sigmoid. ReLu, leaky ReLu.
Adversarial learning. Transfer learning.
Reinforcement learning.
Which ANN for time series data and why.
CNN is better for grid type of data.
Different types of pooling layer
1x1 correlation. How we reduce size with it.
How to optimize Billions of prams in a deep NN.

One shot and zero shot learning?


Code related mathematical and conceptual questions.
Under sampling vs over sampling.
Feature subset selection?
Decision tree. Concept. Mutual information? Entropy.
White noise in data?
Exploitation vs Exploration tradeoff?

View publication stats

You might also like