Active Learning Interview Questions: Sampling, Label Budgets, and Selection Bias

Active learning asks an expensive labeler for selected examples rather than labeling a random sample. It can use a finite budget more efficiently, but it also makes the labeled training set a product of the model’s own decisions. Interviewers look for candidates who can name that feedback loop and keep an independent measurement system outside it.

The selection policy is part of the experimental treatment; the evaluation set and labeling operations must be protected from that treatment.

Establish the evidence boundary before answering

Official or primary-source facts. Recent active-learning surveys describe label complexity and selection strategies, while practical research on non-response shows that a label request does not guarantee an unbiased returned label. These sources motivate explicit sampling and annotation assumptions rather than a claim that uncertainty sampling is automatically optimal. The linked sources at the end support those facts and should take priority if a framework or toolchain changes.

Candidate reports. This is a technical-topic guide, not a company-process guide. I found no reliable, same-cycle candidate-report set that adds a general rule for this topic, so this article makes no claim about a particular employer’s interview format, scoring, or implementation choice.

Reasoned practice guidance. The scenarios, diagnosis order, and acceptance gates below are engineering inferences. They are intentionally framed as things to test in a stated environment, not as universal behavior.

Build the mental model, not a slogan

Pool-based active learning normally begins with a seed set, a large unlabeled pool, a model score, and a query policy. Uncertainty scoring finds examples near a decision boundary under the current model; diversity scoring avoids spending the full budget on near-duplicates; random sampling supplies coverage and a baseline. A production policy often combines them, but the blend should be evaluated against an equal-cost random baseline.

The label budget has more dimensions than a count. Include annotator minutes, queue latency, disagreement, sensitive-content review, rework, and the opportunity cost of asking a difficult question. A thousand trivial labels and a thousand ambiguous labels are not comparable budget units. Capture the policy, model version, and annotator outcomes for each request.

Selection bias appears because the model preferentially labels points it considers informative. That can be useful for training but makes the acquired set non-representative of the deployment population. Never use it as the sole test set. Reserve a randomly sampled or otherwise target-representative evaluation set before selection; preserve its labels and access controls across rounds.

An interviewer may change one assumption halfway through: distributed workers, a new data slice, a checkpoint resume, a different ABI, or a new attacker capability. Return to the boundary and state which observation would change your answer. That is stronger than defending a memorized rule.

Work a bounded interview exercise

Prompt. A fraud-review team can label 5,000 transactions this month. A model’s uncertainty policy selects mostly new merchants and leaves established merchants underrepresented. The dashboard reports a large improvement on the selected labels. Design a policy and evaluation plan.

1. Begin by split design. Freeze a time-aware, representative evaluation set that is not eligible for selection. Define slice metrics by merchant tenure, amount, geography, and review outcome, and choose an operational metric such as recall at a fixed review capacity. The selected-label score remains useful as process telemetry, but it is not the headline model estimate.

2. Allocate the budget deliberately: a random exploration share for coverage and calibration, an uncertainty share for boundary learning, and a diversity or quota mechanism for underrepresented regions. The exact percentages are an experimental choice. Pre-register a small set of variants, use the same budget for the random baseline, and include labeler throughput and disagreement in the comparison.

3. Make annotation operational. Route ambiguous items to a second review path, record non-responses and reasons, sample completed labels for quality, and avoid letting the model’s confidence become the labeler instruction. If label availability differs by merchant type, treat missing labels as a selection mechanism and report it.

The final answer should name a stop condition. For example: roll back a release if a protected slice crosses its error budget; reject a schema change if compatibility tests fail; stop a performance tune if the profile shows the presumed allocation is not material. A decision with a measured exit condition is more useful than an unbounded exploration.

Debug from evidence in this order

• Independence: protect a deployment-representative evaluation set before query selection begins.
• Budget: log label count, cost, latency, disagreement, and completion by sampling policy and slice.
• Coverage: measure selected-pool representation against the unlabeled and target populations.
• Baseline: compare to random sampling at matched cost and time, with repeat seeds where possible.

This sequence is deliberately conservative. It separates a fact established by a specification or runtime document from a local observation, then asks for the smallest intervention that can disprove the current theory. It also produces artifacts a reviewer can reproduce: an input slice, command line or configuration, before/after measurement, and rollback point.

Turn the explanation into an engineering decision

For Active Learning Interview Questions: Sampling, Label Budgets, and Selection Bias, practice a one-page decision record rather than a stream of concepts. Begin with the user-visible or system-level failure: a bad slice metric, incompatible reader, allocation spike, unsafe data path, or portability assumption. State what must remain true after the change. This is the invariant. Examples include preserving a monetary value’s unit, keeping a request within a latency budget, preventing untrusted data from reaching a sensitive operation, or retaining the behavior promised by an API. An invariant makes it possible to reject an attractive implementation that optimizes the wrong thing.

Next, make the scope testable. Name the build or runtime version, data slice, traffic window, hardware, model checkpoint, compiler target, or security boundary that applies. A concise qualifier is more credible than a universal statement. “Under this ONNX Runtime build on the target CPU,” “for the old writer and new reader compatibility matrix,” or “when this request is reachable without an authorization guard” gives the interviewer a condition they can challenge. If the condition is unknown, say which document, log, experiment, or source inspection would resolve it.

Then compare two or three options against the invariant. For each option, write its mechanism, benefit, failure mode, and proof. A mechanism is not a benefit: a lower quantization range may improve resolution near zero but may saturate a rare activation; a broader CodeQL query may find more paths but can consume triage capacity; a new layout may reduce bytes but create an ABI or alignment cost. The proof should be concrete: a compatibility matrix, fixed-input test, profile, slice evaluation, path trace, or compiler output. It should include the baseline so an apparent improvement has meaning.

Make rollback part of the design. Identify which input, model, schema, configuration, or binary remains available; which metric triggers a stop; and who can act. A rollback plan does not imply the change is unsafe. It demonstrates that the team can learn from a bounded experiment without turning a local defect into unrecoverable history. For a non-reversible boundary, such as schema deletion or a public API contract, replace rollback with a staged migration and an explicit deprecation window.

Finally, communicate uncertainty precisely. Say “the documentation establishes X,” “the trace or benchmark establishes Y in this build,” and “I infer Z, so I would validate it with W.” That separation is especially important for Machine Learning Engineer work, where a mechanism may be correct while the proposed operational response is still wrong for the product’s constraints. It also keeps an interview answer honest when the interviewer has intentionally withheld a critical fact.

A five-minute whiteboard drill

Use this repeatable drill before an interview. Spend the first minute drawing the data, object, update, or control-flow boundary relevant to the prompt. Spend the second minute labeling the state that crosses the boundary and the state that must remain local, immutable, compatible, bounded, or trusted. Spend the third minute naming the symptom and listing two competing hypotheses. Spend the fourth minute choosing the cheapest discriminating observation. Spend the final minute proposing a small fix, an acceptance metric, and a rollback or migration path.

This format avoids a common failure mode: changing a configuration while several causes remain plausible. It also prevents a different failure mode: producing a technically elegant design with no evidence it works in the environment. If a panel asks for code, implement a narrow probe that exposes one invariant. If it asks for system design, write the observability and ownership fields next to the component diagram. If it asks for debugging, lead with the evidence you would gather before naming the fix.

Score your own response on five questions. Did it name the boundary? Did it distinguish a documented fact from a local observation? Did it identify an assumption that could reverse the recommendation? Did it specify a measurement that falsifies the current theory? Did it include a safe next action? A “yes” to all five usually signals stronger engineering judgment than adding another unsupported optimization detail.

When the prompt gives incomplete information, say so without stopping. Choose a provisional assumption that minimizes risk, explain why it is provisional, and state the first observation you would request. For example, a performance question may need a profile and workload distribution; a model question may need a protected evaluation slice; a security question may need the actual sink API and route reachability. This is not hedging. It is a way to turn uncertainty into an ordered investigation. A candidate who can do that will usually make better trade-offs after the interview too.

How to answer follow-up questions well

1. State what the model chooses and what the human supplies.
2. Separate uncertainty from diversity and random exploration.
3. Define the budget in operational terms, not only a row count.
4. Explain why selected labels cannot be the only evaluation data.
5. Mention non-response and annotation quality as data-generating processes.

Avoid two common interview failures. First, do not make a tool guarantee that its documentation does not make; qualify compiler, hardware, library, and deployment assumptions. Second, do not answer with a bag of keywords. A compact causal chain—input or state, mechanism, observable symptom, verification, reversible fix—lets the interviewer assess your judgment.

Sources and Further Reading

• A survey of active learning methods and challenges
• Active learning with biased non-response to label requests
• The Label Complexity of Active Learning from Observational Data