Preparing Interviews - ML Questions
Inspired From
I am making the solutions into bullet point so I can remember them easily.
*
= Appeared in my personal interview.
Questions TBA
What is regularization, why do we use it, and give some examples of common methods?
- TBA
Explain Principal Component Analysis (PCA)?
- TBA
Why is ReLU better and more often used than Sigmoid in Neural Networks?
- TBA
8) Given stride S and kernel sizes for each layer of a (1-dimensional) CNN, create a function to compute the receptive field of a particular node in the network. This is just finding how many input nodes actually connect through to a neuron in a CNN. [src]
The receptive field are defined portion of space within an inputs that will be used during an operation to generate an output.
Considering a CNN filter of size k, the receptive field of a peculiar layer is only the number of input used by the filter, in this case k, multiplied by the dimension of the input that is not being reduced by the convolutionnal filter a. This results in a receptive field of k*a.
More visually, in the case of an image of size 32x32x3, with a CNN with a filter size of 5x5, the corresponding recpetive field will be the the filter size, 5 multiplied by the depth of the input volume (the RGB colors) which is the color dimensio. This thus gives us a recpetive field of dimension 5x5x3.
9) Implement connected components on an image/matrix. [src]
10) Implement a sparse matrix class in C++. [src]
11) Create a function to compute an integral image, and create another function to get area sums from the integral image.[src]
12) How would you remove outliers when trying to estimate a flat plane from noisy samples? [src]
Random sample consensus (RANSAC) is an iterative method to estimate parameters of a mathematical model from a set of observed data that contains outliers, when outliers are to be accorded no influence on the values of the estimates.
[src]
13) How does CBIR work? [src]
[Answer]
Content-based image retrieval is the concept of using images to gather metadata on their content. Compared to the current image retrieval approach based on the keywords associated to the images, this technique generates its metadata from computer vision techniques to extract the relevant informations that will be used during the querying step. Many approach are possible from feature detection to retrieve keywords to the usage of CNN to extract dense features that will be associated to a known distribution of keywords.
With this last approach, we care less about what is shown on the image but more about the similarity between the metadata generated by a known image and a list of known label and or tags projected into this metadata space.
14) How does image registration work? Sparse vs. dense optical flow and so on. [src]
15) Describe how convolution works. What about if your inputs are grayscale vs RGB imagery? What determines the shape of the next layer?[src]
In a convolutional neural network (CNN), the convolution operation is applied to the input image using a small matrix called a kernel or filter. The kernel slides over the image in small steps, called strides, and performs element-wise multiplications with the corresponding elements of the image and then sums up the results. The output of this operation is called a feature map.
When the input is RGB(or more than 3 channels) the sliding window will be a sliding cube. The shape of the next layer is determined by Kernel size, number of kernels, stride, padding, and dialation.
16) Talk me through how you would create a 3D model of an object from imagery and depth sensor measurements taken at all angles around the object. [src]
There are two popular methods for 3D reconstruction:
SfM is better suited for creating models of large scenes while MVS is better suited for creating models of small objects.
19) Implement non maximal suppression as efficiently as you can. [src]
Non-Maximum Suppression (NMS) is a technique used to eliminate multiple detections of the same object in a given image.
To solve that first sort bounding boxes based on their scores(N LogN). Starting with the box with the highest score, remove boxes whose overlapping metric(IoU) is greater than a certain threshold.(N^2)
To optimize this solution you can use special data structures to query for overlapping boxes such as R-tree or KD-tree. (N LogN)
[src]
21) What is data normalization and why do we need it? [src]
Data normalization is very important preprocessing step, used to rescale values to fit in a specific range to assure better convergence during backpropagation. In general, it boils down to subtracting the mean of each data point and dividing by its standard deviation. If we don’t do this then some of the features (those with high magnitude) will be weighted more in the cost function (if a higher-magnitude feature changes by 1%, then that change is pretty big, but for smaller features it’s quite insignificant). The data normalization makes all features weighted equally.
22) Why do we use convolutions for images rather than just FC layers? [src]
Firstly, convolutions preserve, encode, and actually use the spatial information from the image. If we used only FC layers we would have no relative spatial information. Secondly, Convolutional Neural Networks (CNNs) have a partially built-in translation in-variance, since each convolution kernel acts as it’s own filter/feature detector.
24) Why do we have max-pooling in classification CNNs? [src]
for a role in Computer Vision. Max-pooling in a CNN allows you to reduce computation since your feature maps are smaller after the pooling. You don’t lose too much semantic information since you’re taking the maximum activation. There’s also a theory that max-pooling contributes a bit to giving CNNs more translation in-variance. Check out this great video from Andrew Ng on the benefits of max-pooling.
25) Why do segmentation CNNs typically have an encoder-decoder style / structure? [src]
The encoder CNN can basically be thought of as a feature extraction network, while the decoder uses that information to predict the image segments by “decoding” the features and upscaling to the original image size.
26) What is the significance of Residual Networks? [src]
The main thing that residual connections did was allow for direct feature access from previous layers. This makes information propagation throughout the network much easier. One very interesting paper about this shows how using local skip connections gives the network a type of ensemble multi-path structure, giving features multiple paths to propagate throughout the network.
27) What is batch normalization and why does it work? [src]
Training Deep Neural Networks is complicated by the fact that the distribution of each layer’s inputs changes during training, as the parameters of the previous layers change. The idea is then to normalize the inputs of each layer in such a way that they have a mean output activation of zero and standard deviation of one. This is done for each individual mini-batch at each layer i.e compute the mean and variance of that mini-batch alone, then normalize. This is analogous to how the inputs to networks are standardized. How does this help? We know that normalizing the inputs to a network helps it learn. But a network is just a series of layers, where the output of one layer becomes the input to the next. That means we can think of any layer in a neural network as the first layer of a smaller subsequent network. Thought of as a series of neural networks feeding into each other, we normalize the output of one layer before applying the activation function, and then feed it into the following layer (sub-network).
28) Why would you use many small convolutional kernels such as 3x3 rather than a few large ones? [src]
This is very well explained in the VGGNet paper. There are 2 reasons: First, you can use several smaller kernels rather than few large ones to get the same receptive field and capture more spatial context, but with the smaller kernels you are using less parameters and computations. Secondly, because with smaller kernels you will be using more filters, you’ll be able to use more activation functions and thus have a more discriminative mapping function being learned by your CNN.
29) Why do we need a validation set and test set? What is the difference between them? [src]
When training a model, we divide the available data into three separate sets:
- The training dataset is used for fitting the model’s parameters. However, the accuracy that we achieve on the training set is not reliable for predicting if the model will be accurate on new samples.
- The validation dataset is used to measure how well the model does on examples that weren’t part of the training dataset. The metrics computed on the validation data can be used to tune the hyperparameters of the model. However, every time we evaluate the validation data and we make decisions based on those scores, we are leaking information from the validation data into our model. The more evaluations, the more information is leaked. So we can end up overfitting to the validation data, and once again the validation score won’t be reliable for predicting the behaviour of the model in the real world.
- The test dataset is used to measure how well the model does on previously unseen examples. It should only be used once we have tuned the parameters using the validation set.
So if we omit the test set and only use a validation set, the validation score won’t be a good estimate of the generalization of the model.
30) What is stratified cross-validation and when should we use it? [src]
Cross-validation is a technique for dividing data between training and validation sets. On typical cross-validation this split is done randomly. But in stratified cross-validation, the split preserves the ratio of the categories on both the training and validation datasets.
For example, if we have a dataset with 10% of category A and 90% of category B, and we use stratified cross-validation, we will have the same proportions in training and validation. In contrast, if we use simple cross-validation, in the worst case we may find that there are no samples of category A in the validation set.
Stratified cross-validation may be applied in the following scenarios:
- On a dataset with multiple categories. The smaller the dataset and the more imbalanced the categories, the more important it will be to use stratified cross-validation.
- On a dataset with data of different distributions. For example, in a dataset for autonomous driving, we may have images taken during the day and at night. If we do not ensure that both types are present in training and validation, we will have generalization problems.
31) Why do ensembles typically have higher scores than individual models? [src]
An ensemble is the combination of multiple models to create a single prediction. The key idea for making better predictions is that the models should make different errors. That way the errors of one model will be compensated by the right guesses of the other models and thus the score of the ensemble will be higher.
We need diverse models for creating an ensemble. Diversity can be achieved by:
- Using different ML algorithms. For example, you can combine logistic regression, k-nearest neighbors, and decision trees.
- Using different subsets of the data for training. This is called bagging.
- Giving a different weight to each of the samples of the training set. If this is done iteratively, weighting the samples according to the errors of the ensemble, it’s called boosting.
Many winning solutions to data science competitions are ensembles. However, in real-life machine learning projects, engineers need to find a balance between execution time and accuracy.
32) What is an imbalanced dataset? Can you list some ways to deal with it? [src]
An imbalanced dataset is one that has different proportions of target categories. For example, a dataset with medical images where we have to detect some illness will typically have many more negative samples than positive samples—say, 98% of images are without the illness and 2% of images are with the illness.
There are different options to deal with imbalanced datasets:
- Oversampling or undersampling. Instead of sampling with a uniform distribution from the training dataset, we can use other distributions so the model sees a more balanced dataset.
- Data augmentation. We can add data in the less frequent categories by modifying existing data in a controlled way. In the example dataset, we could flip the images with illnesses, or add noise to copies of the images in such a way that the illness remains visible.
- Using appropriate metrics. In the example dataset, if we had a model that always made negative predictions, it would achieve a precision of 98%. There are other metrics such as precision, recall, and F-score that describe the accuracy of the model better when using an imbalanced dataset.
33) Can you explain the differences between supervised, unsupervised, and reinforcement learning? [src]
In supervised learning, we train a model to learn the relationship between input data and output data. We need to have labeled data to be able to do supervised learning.
With unsupervised learning, we only have unlabeled data. The model learns a representation of the data. Unsupervised learning is frequently used to initialize the parameters of the model when we have a lot of unlabeled data and a small fraction of labeled data. We first train an unsupervised model and, after that, we use the weights of the model to train a supervised model.
In reinforcement learning, the model has some input data and a reward depending on the output of the model. The model learns a policy that maximizes the reward. Reinforcement learning has been applied successfully to strategic games such as Go and even classic Atari video games.
36) What is Precision?
Precision (also called positive predictive value) is the fraction of relevant instances among the retrieved instances
Precision = true positive / (true positive + false positive)
[src]
37) What is Recall?
Recall (also known as sensitivity) is the fraction of relevant instances that have been retrieved over the total amount of relevant instances.
Recall = true positive / (true positive + false negative)
[src]
38) Define F1-score. [src]
It is the weighted average of precision and recall. It considers both false positive and false negative into account. It is used to measure the model’s performance.
F1-Score = 2 * (precision * recall) / (precision + recall)
47) Define LSTM. [src]
Long Short Term Memory – are explicitly designed to address the long term dependency problem, by maintaining a state what to remember and what to forget.
48) List the key components of LSTM. [src]
- Gates (forget, Memory, update & Read)
- tanh(x) (values between -1 to 1)
- Sigmoid(x) (values between 0 to 1)
49) List the variants of RNN. [src]
- LSTM: Long Short Term Memory
- GRU: Gated Recurrent Unit
- End to End Network
- Memory Network
50) What is Autoencoder, name few applications. [src]
Auto encoder is basically used to learn a compressed form of given data. Few applications include
- Data denoising
- Dimensionality reduction
- Image reconstruction
- Image colorization
52) What’s the difference between boosting and bagging?
Boosting and bagging are similar, in that they are both ensembling techniques, where a number of weak learners (classifiers/regressors that are barely better than guessing) combine (through averaging or max vote) to create a strong learner that can make accurate predictions. Bagging means that you take bootstrap samples (with replacement) of your data set and each sample trains a (potentially) weak learner. Boosting, on the other hand, uses all data to train each learner, but instances that were misclassified by the previous learners are given more weight so that subsequent learners give more focus to them during training. [src]
53) Explain how a ROC curve works. [src]
The ROC curve is a graphical representation of the contrast between true positive rates and the false positive rate at various thresholds. It’s often used as a proxy for the trade-off between the sensitivity of the model (true positives) vs the fall-out or the probability it will trigger a false alarm (false positives).
54) What’s the difference between Type I and Type II error? [src]
Type I error is a false positive, while Type II error is a false negative. Briefly stated, Type I error means claiming something has happened when it hasn’t, while Type II error means that you claim nothing is happening when in fact something is.
A clever way to think about this is to think of Type I error as telling a man he is pregnant, while Type II error means you tell a pregnant woman she isn’t carrying a baby.
55) What’s the difference between a generative and discriminative model? [src]
A generative model will learn categories of data while a discriminative model will simply learn the distinction between different categories of data. Discriminative models will generally outperform generative models on classification tasks.
56) Instance-Based Versus Model-Based Learning.
-
Instance-based Learning: The system learns the examples by heart, then generalizes to new cases using a similarity measure.
-
Model-based Learning: Another way to generalize from a set of examples is to build a model of these examples, then use that model to make predictions. This is called model-based learning.
[src]
57) When to use a Label Encoding vs. One Hot Encoding?
This question generally depends on your dataset and the model which you wish to apply. But still, a few points to note before choosing the right encoding technique for your model:
We apply One-Hot Encoding when:
- The categorical feature is not ordinal (like the countries above)
- The number of categorical features is less so one-hot encoding can be effectively applied
We apply Label Encoding when:
- The categorical feature is ordinal (like Jr. kg, Sr. kg, Primary school, high school)
- The number of categories is quite large as one-hot encoding can lead to high memory consumption
58) What is the difference between LDA and PCA for dimensionality reduction?
Both LDA and PCA are linear transformation techniques: LDA is a supervised whereas PCA is unsupervised – PCA ignores class labels.
We can picture PCA as a technique that finds the directions of maximal variance. In contrast to PCA, LDA attempts to find a feature subspace that maximizes class separability.
59) What is t-SNE?
t-Distributed Stochastic Neighbor Embedding (t-SNE) is an unsupervised, non-linear technique primarily used for data exploration and visualizing high-dimensional data. In simpler terms, t-SNE gives you a feel or intuition of how the data is arranged in a high-dimensional space.
60) What is the difference between t-SNE and PCA for dimensionality reduction?
The first thing to note is that PCA was developed in 1933 while t-SNE was developed in 2008. A lot has changed in the world of data science since 1933 mainly in the realm of compute and size of data. Second, PCA is a linear dimension reduction technique that seeks to maximize variance and preserves large pairwise distances. In other words, things that are different end up far apart. This can lead to poor visualization especially when dealing with non-linear manifold structures. Think of a manifold structure as any geometric shape like: cylinder, ball, curve, etc.
t-SNE differs from PCA by preserving only small pairwise distances or local similarities whereas PCA is concerned with preserving large pairwise distances to maximize variance.
66) What is the basic difference between LSTM and Transformers? [src]
LSTMs (Long Short Term Memory) models consist of RNN cells designed to store and manipulate information across time steps more efficiently. In contrast, Transformer models contain a stack of encoder and decoder layers, each consisting of self attention and feed-forward neural network components.
66) What are RCNNs? [src]
Recurrent Convolutional model is a model that is specially designed to make predictions using a sequence of images (more commonly also know as video). These models are used in object detection tasks in computer vision. The RCNN approach combines both region proposal techniques and convolutional neural networks (CNNs) to identify and locate objects within an image.
Questions that I am familiar with
What’s the trade-off between bias and variance?*
- Simple models : few parameters => high bias and low variance
- Complex models: large parameters => low bias and high variance
Why?
- High bias because unable to capture the underlying patterns in the data
- low variance because generalizes badly
What is gradient descent?
- Optimization algorithm used to find the values of parameters (coefficients) of a function (f) that minimizes a cost function (cost).
Why uses it?
- The parameters cannot be calculated analytically (e.g. using linear algebra)
What is cost function?
- A Cost function is a scalar functions which Quantifies the error factor of the Neural Network. Lower the cost function better the Neural network. Eg: MNIST Data set to classify the image, input image is digit 2 and the Neural network wrongly predicts it to be 3
What is learning rate?
- A hyper-parameter that controls how much we are adjusting the weights of our network with respect the loss gradient.
What is momentum?
- Momentum lets the optimization algorithm remembers its last step, and adds some proportion of it to the current step.
- Even if the algorithm is stuck in a flat region, or a small local minimum, it can get out and continue towards the true minimum.
What is the difference between Batch Gradient Descent and Stochastic Gradient Descent?
- Batch gradient descent computes the gradient using the whole dataset. This is great for convex, or relatively smooth error manifolds.
- Stochastic gradient descent (SGD) computes the gradient using a single sample. SGD works well for error manifolds that have lots of local maxima/minima because of noisy update.
Explain over- and under-fitting and how to combat them?*
- Overfitting: Val/Test Set having lower performance than on a Training Set => Tried to capture too much pattern in Training set
- Underfitting: => Train Set have a bit lower performance than Val/Test Set => Failed to capture enough pattern in Training set
How to combat them?
- Cross-validation
- Regularization
- L1 (Lasso) or L2 (Ridge) regularization: Add penalty terms
- Dropout: dropping out of some of the units in a neural network (variant model trained in every step)
- Early stopping: stop when it almost overfit
- Data augmentation: More data
- …
How do you combat the curse of dimensionality?
What is Curse of Dimensionality?
- analyzing and organizing data in high-dimensional spaces is difficult
How to combat it?
- Feature Selection(manual or via statistical methods) : Select important dimensionalities
- Principal Component Analysis (PCA): Project into lower dimensionalities
- …
What is vanishing gradient?
- Back propagation becomes less and less useful in passing information to the lower layers due to Chain rule (number become extremely small).