Core Machine Learning and AI Knowledge

NCA-GENM · 60 questions

• A cooperage trains a network on stave photos paired with stave notes. An apprentice says the run succeeded because the preview “looks nice.” What quantity actually drives the weight updates during training?
• A lighthouse lens room walks through one training step on a cracked-prism still plus a short keeper note. Which statement correctly separates the two halves of that step?
• A honey-house desk trains on comb photos with harvest notes. One volunteer calls every single comb “an epoch,” and another calls a shuffled tray of sixteen pairs “the whole dataset.” Which assignment is correct?
• A milliner’s loft fits every training still and label perfectly, then misses next season’s new hat blocks. What pattern does that describe?
• A blacksmith shop trains a tiny network on quench stills plus temperature notes, and both the training loss and the held-out loss stay high. What does that pattern indicate?
• A rum rickhouse intern wants to tune every knob on the same barrel-end photos that will later be called “the test.” Why keep training, validation, and a final untouched split separate?
• A canal-lock keeper’s training curve either jumps wildly after each update or barely descends over many passes. What learning-rate idea best explains those two behaviors?
• A stained-glass studio watches training loss keep falling while the held-out loss turns upward. What training-loop idea fits ending the run when the unused split no longer improves?
• A charcoal-kiln camp stacks a very deep stills-and-notes network. In one recipe the error signal becomes tiny after a few layers; in another the numbers explode. What failure modes are those?
• A letterpress shop is ready to train on plate photos plus ink notes. A volunteer offers a hand-written update loop on a spreadsheet. What is the appropriate associate-level training tool?
• A sled-dog kennel hears that training the photo-plus-log model “uses the GPU.” An intern offers to write a custom CUDA kernel “so training counts.” What is the correct associate stance?
• A seed-vault has a few hundred packet photos with species tags and many more untagged packets. How do supervised labeled-pair training and a self-supervised pretext differ?
• A tannery hide loft must choose an architecture that reads grain photos and later a short grade note. Which pairing best matches common architecture strengths?
• A violin shop wants one network that compresses a belly photo into a code, one that writes a condition note from that code, and one that does both. Which role names match those jobs?
• A bookbindery stacks many layers on cover stills plus binder notes and early layers stop receiving a useful error signal. Which architectural idea adds the block’s input back to its output?
• An ice-cutter pond house needs English lines and pond stills to live in one comparable space. How does CLIP’s dual-encoder contrastive design work?
• A hop-kiln office hears that pictures are grown by starting from noise and removing it in stages. Which generative architecture is that?
• A hop-kiln diagram of a diffusion model shows a shrinking path, a growing path, and bridges between matching sizes. What common denoiser backbone is that?
• A map-archive wants the same transformer family that reads catalog sentences to also read a sheet by cutting it into patches. What architecture idea is that?
• An organ-pipe loft can wire every layer strictly after the last, or allow a parallel / skip graph so air and scale notes meet later. What contrast is that?
• A telescope-grinding shop hears that high-resolution synthesis often works in a compressed stage and then decodes to pixels. What architecture awareness does that describe?
• A carousel restoration barn wants a talking figure. Official additional material points to ACE with Riva speech, Audio2Face, and a NeMo LLM. What kind of architecture is that stack?
• A spice-mill captioner asks what is actually stacked inside the language model that will later sit beside stills. What is the reusable transformer unit?
• A cobbler bench places a care sentence next to a sole photo. Before the language stack can run on that sentence, what must happen to the text?
• A mustard-mill note says “add seed then grind,” which means the opposite if the words swap. Why does a transformer need positional information?
• A city flax-retting pond office needs three text jobs: a stack that only reads a note, a stack that only writes a new note, and a stack that reads one note then writes another. Which transformer layouts match those three roles?
• A municipal pigeon-loft log runs for many pages. Staff ask why a modern language model uses a transformer instead of stepping through the log with a long recurrent chain. What is the architectural reason?
• A wool-carding mill is offered “a modern pretrained language model” to sit beside fleece stills in a multimodal pilot. How should the team describe what that model is?
• A municipal kite-maker loft wants the language stack to look at sail-cloth stills while it writes repair notes. Which transformer mechanism hooks another modality into the text generation path?
• A mushroom-cave contractor starts listing prompt catalogs, retrieval chains, and text-alignment recipes from a text-only associate exam. For this multimodal exam, how should transformers be treated?
• A net-mending shed’s care manual is longer than the language stack can hold at once. What does the context window represent for that transformer LLM?
• A brick-kiln yard hears, “We bought a transformer, so it already sees firings and hears the roar.” What is the correct response?
• A copper-roofing shop has JPEG stills of seams ready for a vision model. What readiness step makes those stills neural-network ready?
• A linotype shop’s repair sentence must be multiplied by weight matrices inside a language model. What conversion makes that text neural-network ready?
• A soap-boiling shed records the kettle and a volunteer passes only the recording’s filename string into a speech model. What must happen before the model can run?
• A dune-grass nursery records a short planting clip and will later ask a typed question about it. Official materials allow video through an NVIDIA AI Blueprint with VIA. At data-readiness depth, how should that video be treated?
• A wax-seal archive has a seal photo in pixels and a motto in tokens. What readiness goal must both sides meet before a joint model can score a match?
• A piano-action shop’s parts catalog is a scanned PDF page. Which readiness step turns that page into text a network can use?
• A pearl-boat locker mixes 8 kHz radio clips with 48 kHz deck recordings and stills of three different sizes. Why is this collection not yet network-ready?
• An ice-fishing shack tags each hole photo “safe,” “thin,” or “open water.” How should those categorical tags be made neural-network ready for a training loss?
• A thatcher’s yard has notes of three words and notes of three pages next to roof stills. Why must variable-length text or audio be padded or packed before a batched forward pass?
• A silversmith dumps whole file blobs—PDF bytes, WAV bytes, camera bytes—into one array and calls it multimodal input. What is wrong with that approach?
• A bone-china studio concatenates raw pixel values next to raw token ids and feeds that concat into one stack. Which fusion type is that?
• A basket-weaving shed runs a stills-only grader and a notes-only grader, then averages or votes at the end. Which fusion type is that?
• A felt-hat factory runs a vision encoder and a text encoder to mid-level codes, then lets those codes interact—often by cross-attention—before a joint head. Which fusion type is that?
• A windmill flour loft concatenates millions of pixel numbers with a twelve-token note, and the stills drown the sentence. What known early-fusion weakness does that illustrate?
• An ink-making shop already trusts a stills grader and a smell-note grader and only needs a final mix. Updating one specialist must not retrain the other. Which fusion choice best fits?
• A coral-nursery log only makes sense if a bleaching patch in the photo is tied to a phrase in the diver note before any decision. Which fusion type should the team pick?
• A tea-tasting room has a spoken comment, a liquor photo, and a tasting slip. As a design concept, how do the three fusion types join those modalities?
• A pulley-block shop intern calls “late fusion” the name of a sharper picture. What correction is needed?
• A typewriter-repair bench claims “late fusion of audio and stills,” but the audio was never turned into features. What is wrong?
• An ice-house must choose a fusion type. Leadership either wants to “swap the stills model next month without touching text” or needs “the caption to point at a region.” How should those constraints map?
• A ferry-ticket office has a still of a pass, a spoken request, and a typed note. Something must decide which encoder each stream enters and when those streams may meet. What is that routing of data types called?
• A sled-maker loft wants the system to greet, look up a runner part, speak a reply, then optionally move a face. What kind of orchestration is that sequencing of jobs?
• An ice-climbing hut sometimes has only a spoken condition report and sometimes only a slope photo. A router sends the available type to the matching encoder. What is that pattern?
• A calliope workshop script says: acknowledge the visitor, fetch the tune card, then speak the title. Even if every step uses the same text encoder, what is this script?
• A lantern-maker intern points at an intermediate concat layer and says, “That is the agent.” What is the correct distinction?
• A canal-barge office has a planner that picks “ask a question” versus “show a still.” Someone files the planner under “late fusion.” What should happen?
• A tidal-pool lab uses an NVIDIA AI Blueprint with VIA on a short clip plus a typed question. At Domain 2 depth, how should that situation be read?
• A bell-foundry visitor desk can stream speech and drive a face (ACE / Riva / Audio2Face / NeMo LLM from official additional material) and can also follow a “listen, answer, smile” script. How should those concerns be split?