Functions, Limits & Continuity — Concept Check
Questions on Functions, Limits & Continuity. A limit asks "what value is this function heading toward?" and continuity means "no sudden jumps" — the smoothness that lets gradient-based learning work at all.
Questions in this quiz
8 questions on Functions, Limits & Continuity — Concept Check. Work through them here, then start the quiz to answer them and see your score.
1.A neural network training process is not improving. The loss remains constant despite small changes in input. What could be the issue?
- AThe activation function is not continuous.
- BThe dataset is too small.
- CThe learning rate is too high.
- DThe model has too many layers.
2.A function f(x) is defined as sin(x)/x. As x approaches 0, what value does f(x) approach?
- AInfinity
- BUndefined
- C1
- D0
3.During backpropagation, a small change in weights leads to a large change in loss. What might be causing this?
- AThe model is overfitting.
- BThe dataset is too diverse.
- CThe activation function is discontinuous.
- DThe learning rate is too low.
4.A step function is used as an activation function in a neural network. What is a potential problem with this choice?
- AIt requires normalization.
- BIt is not continuous, causing unpredictable changes.
- CIt is computationally expensive.
- DIt is too smooth, reducing model capacity.
5.You observe that a model using ReLU activation functions is not learning well. What could be a reason?
- AReLU is too smooth, reducing expressiveness.
- BReLU requires large datasets.
- CReLU is computationally expensive.
- DReLU has a kink at zero, causing issues.
6.A function is continuous if:
- AIt is defined for all real numbers.
- BIt is linear.
- CIt is differentiable.
- DIt has no gaps, jumps, or holes.
7.A model's output changes erratically with small input changes. What might be happening?
- AThe model is too complex.
- BThe learning rate is too low.
- CThe function is not continuous.
- DThe dataset is too small.
8.Why are smooth activation functions preferred in neural networks?
- AThey are easier to implement.
- BThey work with all types of data.
- CThey ensure small input changes lead to small output changes.
- DThey require less computational power.