|
| 1 | +[ |
| 2 | + { |
| 3 | + "question": "Broadly, what is machine learning?", |
| 4 | + "type": "multiple_choice", |
| 5 | + "answers": [ |
| 6 | + { |
| 7 | + "answer": "Machine learning is the field of study that gives computes the ability to learn without being explicitly programmed.", |
| 8 | + "correct": true, |
| 9 | + "feedback": "Correct: You could also define it as 'The science (and art) of programming computers so they learn from data'" |
| 10 | + }, |
| 11 | + { |
| 12 | + "answer": "Machine learning is about writing explicit rules and algorithms to solve specific problems", |
| 13 | + "correct": false, |
| 14 | + "feedback": "Incorrect: Machine learning isn't about pre-written rules, but instead learning from data following general patterns" |
| 15 | + }, |
| 16 | + { |
| 17 | + "answer": "Machine learning is a technique where algorithms are programmed to iteratively refine their code until they achieve optimal performance.", |
| 18 | + "correct": false, |
| 19 | + "feedback": "Incorrect: Generally, ML algorithms adjust their parameters, not their underlying code." |
| 20 | + }, |
| 21 | + { |
| 22 | + "answer": "Machine learning is about storing and retrieving vast amounts of data efficiently, like a database system.", |
| 23 | + "correct": false, |
| 24 | + "feedback": "Incorrect: While ML uses data, it is about finding patterns and making predictions from said data, rather than storing/retrieving it." |
| 25 | + } |
| 26 | + ] |
| 27 | + }, |
| 28 | + { |
| 29 | + "question": "What is the general flow of a machine learning approach to a problem?", |
| 30 | + "type": "multiple_choice", |
| 31 | + "answers": [ |
| 32 | + { |
| 33 | + "answer": "Study the problem -> Write rules -> Evaluate -> Publish", |
| 34 | + "correct": false, |
| 35 | + "feedback": "Incorrect: In machine learning we don't write explicit rules like in a traditional approach. We train an algorithm to learn said rules instead." |
| 36 | + }, |
| 37 | + { |
| 38 | + "answer": "Study the problem -> Solve the problem -> Publish", |
| 39 | + "correct": false, |
| 40 | + "feedback": "Incorrect: Although this is how ML can seem upon a quick glance, in reality 'solve the problem' is a little more involved than just that." |
| 41 | + }, |
| 42 | + { |
| 43 | + "answer": "Study the problem -data-> Train model -> Evaluate -> Analyse errors -> Update data -data-> Train model -> Evaluate -> Publish", |
| 44 | + "correct": true, |
| 45 | + "feedback": "Correct" |
| 46 | + }, |
| 47 | + { |
| 48 | + "answer": "Collect data -> Train model -> Publish", |
| 49 | + "correct": false, |
| 50 | + "feedback": "Incorrect: This skips the critical steps of ML where you evaluate the model and refine/improve it iteratively" |
| 51 | + } |
| 52 | + ] |
| 53 | + }, |
| 54 | + { |
| 55 | + "question": "Suppose you want to create and train a model that is able to determine whether a vehicle is electric, hybrid or gas (using various aspects of the car). What kind of model would you use?", |
| 56 | + "type": "multiple_choice", |
| 57 | + "answers": [ |
| 58 | + { |
| 59 | + "answer": "Regression model", |
| 60 | + "correct": false, |
| 61 | + "feedback": "Incorrect: regression models will predict a continuous numerical value from the new sample's input features. Our desired output here is categorical." |
| 62 | + }, |
| 63 | + { |
| 64 | + "answer": "Classification model", |
| 65 | + "correct": true, |
| 66 | + "feedback": "Correct" |
| 67 | + }, |
| 68 | + { |
| 69 | + "answer": "Convolutional neural network", |
| 70 | + "correct": false, |
| 71 | + "feedback": "Incorrect: Though this might work, its generally better to start with a simpler model before moving on to more complex models like neural networks." |
| 72 | + }, |
| 73 | + { |
| 74 | + "answer": "Time-series forcasting model", |
| 75 | + "correct": false, |
| 76 | + "feedback": "Incorrect: Time-series models predict future values based on past trends. This doesn't apply here, we want to categorise vehicles not predict trends." |
| 77 | + } |
| 78 | + ] |
| 79 | + }, |
| 80 | + { |
| 81 | + "question": "What is meant by overfitting?", |
| 82 | + "type": "multiple_choice", |
| 83 | + "answers": [ |
| 84 | + { |
| 85 | + "answer": "This is when the input data is very poor quality (garbage in, garbage out)", |
| 86 | + "correct": false, |
| 87 | + "feedback": "Incorrect: Not quite, overfitting in this case would be if your model perfectly learned this poor training data and therefore couldn't make good predictions on new samples." |
| 88 | + }, |
| 89 | + { |
| 90 | + "answer": "This is where the model is too simple to make good predictions on new samples.", |
| 91 | + "correct": false, |
| 92 | + "feedback": "Incorrect: This is an example of underfitting." |
| 93 | + }, |
| 94 | + { |
| 95 | + "answer": "This is where you train a model that does well on your training data, but not very well on new samples.", |
| 96 | + "correct": true, |
| 97 | + "feedback": "Correct" |
| 98 | + }, |
| 99 | + { |
| 100 | + "answer": "This is when your model is not a good fit for the dataset.", |
| 101 | + "correct": false, |
| 102 | + "feedback": "Incorrect: Even if your model IS a good fit for the dataset, you can still end up overfitting if the model learns the patterns of the input data too heavily (and thusly performs poorly on new samples)." |
| 103 | + } |
| 104 | + ] |
| 105 | + }, |
| 106 | + { |
| 107 | + "question": "What is meant by a 'cost function'?", |
| 108 | + "type": "multiple_choice", |
| 109 | + "answers": [ |
| 110 | + { |
| 111 | + "answer": "A function that determines the computational complexity of an algorithm", |
| 112 | + "correct": false, |
| 113 | + "feedback": "Incorrect: This describes algorithmic complexity, not a cost function." |
| 114 | + }, |
| 115 | + { |
| 116 | + "answer": "A metric used to assess how well the model performs on some data.", |
| 117 | + "correct": true, |
| 118 | + "feedback": "Correct" |
| 119 | + }, |
| 120 | + { |
| 121 | + "answer": "This is a function banks use to calculate how much money to give out at an ATM.", |
| 122 | + "correct": false, |
| 123 | + "feedback": "Incorrect: A cost function is not referring to expenses, but instead prediction errors in the model." |
| 124 | + }, |
| 125 | + { |
| 126 | + "answer": "This is a function that calculates how much money your failed AI startup has wasted in cloud computing resources", |
| 127 | + "correct": false, |
| 128 | + "feedback": "Incorrect: A cost function is not referring to expenses, but instead prediction errors in the model." |
| 129 | + } |
| 130 | + ] |
| 131 | + }, |
| 132 | + { |
| 133 | + "question": "Generally, when training a linear regression model, what metric do we use as the cost function?", |
| 134 | + "type": "multiple_choice", |
| 135 | + "answers": [ |
| 136 | + { |
| 137 | + "answer": "Root Mean Squared Error (RMSE)", |
| 138 | + "correct": true, |
| 139 | + "feedback": "Correct" |
| 140 | + }, |
| 141 | + { |
| 142 | + "answer": "Cubic cost function", |
| 143 | + "correct": false, |
| 144 | + "feedback": "Incorrect: This is commonly used in accounting." |
| 145 | + }, |
| 146 | + { |
| 147 | + "answer": "Recall", |
| 148 | + "correct": false, |
| 149 | + "feedback": "Incorrect: This is generally used in classification models, whilst linear regression is a regression model." |
| 150 | + }, |
| 151 | + { |
| 152 | + "answer": "Gradient Descent", |
| 153 | + "correct": false, |
| 154 | + "feedback": "Incorrect: This is not a cost function, but instead a method for updating a cost function." |
| 155 | + } |
| 156 | + ] |
| 157 | + }, |
| 158 | + { |
| 159 | + "question": "What is meant by the no free lunch theorem?", |
| 160 | + "type": "multiple_choice", |
| 161 | + "answers": [ |
| 162 | + { |
| 163 | + "answer": "There is no single best algorithm for predictive modelling problems; you cannot blindly take a 'good' algorithm and expect it to perform well.", |
| 164 | + "correct": true, |
| 165 | + "feedback": "Correct" |
| 166 | + }, |
| 167 | + { |
| 168 | + "answer": "Machine learning models can achieve perfect accuracy without any training data.", |
| 169 | + "correct": false, |
| 170 | + "feedback": "Incorrect: The no free lunch theorem states the opposite of this." |
| 171 | + }, |
| 172 | + { |
| 173 | + "answer": "All optimisation algorithms converge to the same solution given enough time and effort", |
| 174 | + "correct": false, |
| 175 | + "feedback": "Incorrect: The theorem applies to performance of models in generalised contexts, not convergence." |
| 176 | + }, |
| 177 | + { |
| 178 | + "answer": "This alludes to how conferences in the machine learning space do not provide catering", |
| 179 | + "correct": false, |
| 180 | + "feedback": "Incorrect: Though it can be pretty hit-or-miss, conferences usually provide catering. The theorem is about algorithmic performance, not food." |
| 181 | + } |
| 182 | + ] |
| 183 | + }, |
| 184 | + { |
| 185 | + "question": "In ML, how can we be sure that our model truly does well in unseen scenarios?", |
| 186 | + "type": "multiple_choice", |
| 187 | + "answers": [ |
| 188 | + { |
| 189 | + "answer": "Use a really large dataset", |
| 190 | + "correct": false, |
| 191 | + "feedback": "Incorrect: One can still end up overfitting their models, even when using incredibly large datasets" |
| 192 | + }, |
| 193 | + { |
| 194 | + "answer": "Inspect predictions and use your superior intellect to ensure they 'look correct'", |
| 195 | + "correct": false, |
| 196 | + "feedback": "Incorrect: Although this is a common strategy, it's generally not acceptable to do this. Human bias makes this rather unreliable." |
| 197 | + }, |
| 198 | + { |
| 199 | + "answer": "Create a train and test dataset from the original (train-test split) and test the trained model on the test set", |
| 200 | + "correct": true, |
| 201 | + "feedback": "Correct: One can extend this further by holding the test set till the very end, instead testing intermediate models on a validation set generated from the training dataset3" |
| 202 | + }, |
| 203 | + { |
| 204 | + "answer": "Use a simple model and then work your way up to more complex models", |
| 205 | + "correct": false, |
| 206 | + "feedback": "Incorrect: model complexity doesn't guarantee model accuracy in unseen scenarios" |
| 207 | + } |
| 208 | + ] |
| 209 | + }, |
| 210 | + { |
| 211 | + "question": "After collecting your data and defining the problem to solve, what is best practice before choosing and training a model on said data?", |
| 212 | + "type": "multiple_choice", |
| 213 | + "answers": [ |
| 214 | + { |
| 215 | + "answer": "Explore, visualise and preprocess your data", |
| 216 | + "correct": true, |
| 217 | + "feedback": "Correct: This step can help you understand your data which can aid model selection. Bonus: Don't look at your test dataset to reduce your own biases." |
| 218 | + }, |
| 219 | + { |
| 220 | + "answer": "Define a cost function to assess the model used", |
| 221 | + "correct": false, |
| 222 | + "feedback": "Incorrect: Your cost function will likely change depending on the model you use, so this isn't a great idea." |
| 223 | + }, |
| 224 | + { |
| 225 | + "answer": "Perform feature engineering", |
| 226 | + "correct": false, |
| 227 | + "feedback": "Incorrect: generally you should explore your data first before deciding what features you might want to have as inputs to the model." |
| 228 | + }, |
| 229 | + { |
| 230 | + "answer": "Train a complex initial model with which to compare future models", |
| 231 | + "correct": false, |
| 232 | + "feedback": "Incorrect: Explore your data and try simple approaches first (perhaps not even an ML approach)" |
| 233 | + } |
| 234 | + ] |
| 235 | + }, |
| 236 | + { |
| 237 | + "question": "Which of the following is an example of an unsupervised model?", |
| 238 | + "type": "multiple_choice", |
| 239 | + "answers": [ |
| 240 | + { |
| 241 | + "answer": "Logistic regression model", |
| 242 | + "correct": false, |
| 243 | + "feedback": "Incorrect: This requires labelled data and so is an example of supervised learning." |
| 244 | + }, |
| 245 | + { |
| 246 | + "answer": "K-means clustering", |
| 247 | + "correct": true, |
| 248 | + "feedback": "Correct: This model groups unlabelled data by similarity, a classic unsupervised learning approach" |
| 249 | + }, |
| 250 | + { |
| 251 | + "answer": "Decision tree", |
| 252 | + "correct": false, |
| 253 | + "feedback": "Incorrect: Generally you would have labelled data points to train this type of model, making it a supervised approach." |
| 254 | + }, |
| 255 | + { |
| 256 | + "answer": "Diffusion model", |
| 257 | + "correct": false, |
| 258 | + "feedback": "Incorrect: This is a generative model used primarily for tasks like image generation. It usually relies on labelled data points and so is supervised." |
| 259 | + } |
| 260 | + ] |
| 261 | + } |
| 262 | +] |
0 commit comments