Interview questions in AI & ML, page 18
Fine-tuning vs RAG for daily-updated docs
Choose RAG because docs change daily, embed and index chunks in a vector store, retrieve top matches and inject into the prompt.
Exploration versus exploitation: epsilon-greedy and UCB
Exploit current best to earn reward, explore to discover better options; epsilon-greedy explores randomly with probability epsilon; UCB explores by an uncertainty bonus…
Transfer learning from ResNet50 on small data
Replace the final classification head with one sized to your classes, freeze the pretrained convolutional backbone as a feature extractor, train the new head, then optionally fine-tune top blocks at a low…
Propose an architectural solution for contended GPU training resources
Tests multi-tenant GPU scheduling design at scale. Great answers tier jobs by checkpointability, apply quota-based preemption, mix spot and on-demand instances, and use MIG or time-slicing to bin-pack. Red flag: buying GPUs without scheduling logic.
Closed API vs open-weight models for production
APIs offer top quality and zero ops but recurring per-token cost and data-sharing concerns, open weights give control, privacy, and tuning at the price of hosting and MLOps burden.
Why Transformers use multi-head attention
A single head averages into one representation subspace; multiple heads attend in parallel to different subspaces, letting the model capture diverse relations like syntax and coreference at once, then…
Leveraging unlabeled data with limited labels
Pretrain a representation on the million unlabeled images via self-supervision, then fine-tune on the 1,000 labels; or use pseudo-labeling and consistency regularization.
How do you give read-only access to a shared cloud storage bucket?
Bind an IAM role with read permissions to the team at the bucket level, avoid object-level ACLs, and mount read-only on training VMs.
Hugging Face Hub, transformers, and datasets
The Hub hosts models and data, transformers loads models and tokenizers and provides the Trainer, datasets streams and maps preprocessing.
Mode collapse in GANs and how to fix it
Mode collapse is the generator producing few similar outputs, missing data diversity to fool the discriminator; fixes include Wasserstein loss, minibatch discrimination, unrolled GANs, and feature…
Contrastive self-supervised learning with SimCLR
Two augmentations of one image form a positive pair, other images in the batch are negatives; an encoder plus projection head and the NT-Xent loss pull positives together and push negatives apart.
Why was this customer denied: global or local explanation?
This tests matching questions to explanation scope. Global methods show overall behavior; local methods explain one prediction. Specific denials need local methods like SHAP. A red flag is using global summaries like permutation importance or PDPs for a case.
Mixture of Experts architecture and routing
Many expert FFNs per layer, a router picks top-k experts per token, only those compute so active params are far fewer than total.
Spark transformations versus actions
Transformations like map and filter are lazy and build a lineage DAG returning a new RDD; actions like count or collect trigger execution and return a value to the driver.
Prototypical Networks for few-shot classification
An encoder embeds support examples, each class prototype is the mean embedding of its support examples, and a query is classified by nearest prototype using a distance like Euclidean via softmax.
Describe securing an automated ML pipeline and CI/CD integration points
Tests ML supply-chain depth versus bolt-on appsec. Strong answers stage checks across
Designing a production LLM summarization eval
A representative gold set, quality via human or LLM-as-judge plus faithfulness checks, and operational metrics like p95 latency and cost per request.
HDFS purpose and fault tolerance
HDFS stores huge files across many commodity machines as large blocks, replicating each block across nodes for fault tolerance; unlike NTFS or ext4 it is distributed, write-once, and optimized for…
How does MAML's inner and outer loop work?
Inner loop does task-specific gradient steps from shared init, outer loop updates the init for fast adaptability via second-order gradients.
How would you programmatically monitor a deployed model for demographic bias?
Tests operationalizing fairness beyond static audits. Track group metrics like parity and equalized odds; slice by protected attributes; alert on drift; route violations to review. Red flag: treating fairness as a one-time check versus continuous monitoring.
We are hiring for this. Every open role lists the topics its interview covers, so you can prepare for the real thing rather than guessing.
See open roles