text
stringlengths
3
8.33k
repo
stringclasses
52 values
path
stringlengths
6
141
language
stringclasses
35 values
sha
stringlengths
64
64
chunk_index
int32
0
273
n_tokens
int32
1
896
, answered: answeredQuestions.length, correct, missed: answeredQuestions.length - correct, unanswered: questions.length - answeredQuestions.length, accuracy: percent(correct, questions.length), } } export function getMockReviewItems( questions: Question[], session: MockExamSession | null, i...
hands-on-ai
lib/practice.ts
TypeScript
d9bcee5a3841388c9b3b4e678e154c87e521e0b995fd9ff7ad686165d48deea4
2
896
= progress.bookmarks.filter((questionId) => questionIds.has(questionId) ).length return { total: questions.length, answered: answeredQuestions.length, correct, missed: answeredQuestions.length - correct, bookmarked, unanswered: questions.length - answeredQuestions.length, coverage: ...
hands-on-ai
lib/practice.ts
TypeScript
c029b3a88ac94c232fac106381a7fa5f1fde592c2ceca1fdfac25f21f5ebdbee
3
896
: [] if (!selectedChoiceIds.length && answer.isCorrect !== false) { return null } return [ questionId, { selectedChoiceIds, isCorrect: answer.isCorrect === true, attempts: typeof answer.attempts === "number" && ans...
hands-on-ai
lib/practice.ts
TypeScript
91f1ae5fd1f1949e9262486012155f46e277553833568ec8ab6d3b83bea6c85b
4
896
unknown ): Record<PracticeMode, string | null> { const fallback = createDefaultProgress().currentByMode if (!isRecord(value)) { return fallback } return { exam: normalizeQuestionId(value.exam) ?? fallback.exam, practice: normalizeQuestionId(value.practice) ?? fallback.practice, review: normali...
hands-on-ai
lib/practice.ts
TypeScript
49b4edc8ee821e2e2163c00fd6e02705434b90aa2b58691f516c485f6b5b1b4a
5
740
import type { Question } from "./data" const sharedPredictionFields = { origin: "generated_from_slides", answerSource: "expert_verified", confidence: 0.86, examMode: "closed_book", sourcePriority: "primary", approvalStatus: "approved", interactionType: "single_choice", } as const export const PREDICTED_...
hands-on-ai
lib/predicted-mock-questions.ts
TypeScript
ad1af229e6e3bb6d85075300ae9000b3b765930eb6c2320948ba90066f142094
0
896
Counting invalid moves as additional actions instead of available action choices.", coreClosedBookLikely: true, cognitiveLevel: "calculation", stem: "A robot moves on a 4x4 grid. Each state has 4 possible actions: up, down, left, right. If the Q-table stores one Q-value per state-action pair, how many Q-val...
hands-on-ai
lib/predicted-mock-questions.ts
TypeScript
ec0e7fa033dda21af4ea303bc0affcf0cbedcfa1109d063e186c3fe3146aa1f4
1
896
lstm", "forget-gate"], retrievalText: "LSTM forget gate previous cell state kept discarded", }, { ...sharedPredictionFields, id: "mock-2026-pred-q006", sourceRefs: [ { sourceId: "exam-1-2026-pdf", note: "Predictive transformer question based on 2026 language-model coverage.", ...
hands-on-ai
lib/predicted-mock-questions.ts
TypeScript
8a49449fa6cbc661815b555159dc0bb84274f61cbac4285414adcca031329944
2
896
", text: "They can only process text tokens, not images." }, ], correctChoiceIds: ["c1"], explanation: "Convolutions exploit locality and weight sharing, which makes them effective for spatial image patterns.", tags: ["2026-prediction", "cnn", "convolution"], retrievalText: "convolution ...
hands-on-ai
lib/predicted-mock-questions.ts
TypeScript
f0b96fdc0220822c71e45817a4f7effaddc17901509c1c98403f26577568086b
3
896
: "c2", text: "Mean squared error is always required for classification.", }, { id: "c3", text: "The discount factor gamma." }, { id: "c4", text: "BLEU score as the training loss for every classifier.", }, ], correctChoiceIds: ["c1"], explanation: "Softm...
hands-on-ai
lib/predicted-mock-questions.ts
TypeScript
a9f7739f98ab706b21d01a447072c5bd14f22cfb4c7d89eb9e3d9fd37544329a
4
896
id: "c1", text: "It helps represent rare or unseen words using smaller recurring pieces.", }, { id: "c2", text: "It prevents the model from needing positional information.", }, { id: "c3", text: "It converts every language task into reinforcement learning....
hands-on-ai
lib/predicted-mock-questions.ts
TypeScript
9b2c986ddd9db477085e2481c95cd47ba7a863187756eb672bbe238eaf072bc4
5
477
import questionBankJson from "./question-bank.json" import slideIndexJson from "./slide-index.json" import type { QuestionBank, SlideIndex } from "./schema" export const questionBank = questionBankJson as QuestionBank export const slideIndex = slideIndexJson as SlideIndex export type { AnswerSource, Choice, Co...
hands-on-ai
lib/data/index.ts
TypeScript
5c8b93949d6c6a29c62293e8ad672fcd4a75af3febaa623ad6d5f349d18f8207
0
89
{ "version": "1.0.0", "generatedAt": "2026-07-07T20:06:48.206801+00:00", "course": { "code": "HOAI2", "title": "Hands On AI II", "university": "Johannes Kepler University Linz", "language": "en" }, "defaults": { "choicesPerMultipleChoice": 4, "generatedQuestionStyle": "Reuse the traine...
hands-on-ai
lib/data/question-bank.json
JSON
3caf7fa6411c6747fc683aa191208cec757e31b7eadea09f55232270b4dcef7d
0
896
unit-06", "kind": "lecture_slides", "title": "Unit 6: Introduction to Reinforcement Learning", "fileName": "Unit6.pdf", "pageCount": 36, "lectureNumber": 6 }, { "id": "exam-1-2026-page-01", "kind": "exam_image", "title": "Exam 1 2026 screenshot page 1", "fil...
hands-on-ai
lib/data/question-bank.json
JSON
a9c242d0696b8366725949cc2bafb63518e4458eda51483d72ad6acba710144f
1
896
page-08", "kind": "exam_image", "title": "Exam 1 2026 screenshot page 8", "fileName": "exam-1-2026-page-08.png", "assetPath": "/sources/exam-1-2026/exam-1-2026-page-08.png", "pageCount": 1, "year": 2026, "pageNumber": 8, "sourceQuality": "primary_capture" }, { ...
hands-on-ai
lib/data/question-bank.json
JSON
8d8f69fec7a2973a6e9634e9859aa10aa7844f6dfc5cfe9133c2259ca1c7f4e4
2
896
{ "id": "group-a-page-05", "kind": "exam_image", "title": "Past paper Group A photo page 5", "fileName": "group-a-page-05.png", "assetPath": "/sources/past-paper-group-a/group-a-page-05.png", "pageCount": 1, "pageNumber": 5, "sourceQuality": "primary_capture" }, {...
hands-on-ai
lib/data/question-bank.json
JSON
65fec019440ace60139cae466e0bae67e59b94e1f10ec182b101907768693805
3
896
": "exam-1-2026-pdf", "pages": [ 1 ], "questionNumber": "1" } ], "topicIds": [ "neural-networks" ], "lectureIds": [ "summary-unit-01", "summary-unit-02", "unit-01", "unit-02" ], "difficulty": "e...
hands-on-ai
lib/data/question-bank.json
JSON
b2142e41eaaf55fe4fc30164892ceabb50879fcf905ed9b8f6d36c6482e83fea
4
896
"During exploitation, the agent chooses actions that given a specific state of the environment seem best." } ], "correctChoiceIds": [ "c4" ], "answerSource": "expert_verified", "confidence": 0.96, "examMode": "closed_book", "sourcePriority": "primary", "ap...
hands-on-ai
lib/data/question-bank.json
JSON
434a31e4c433fa79c397b6a72d9bb26beb6ad4d2fe467ded9bf2014c1c85f259
5
896
learning scenario with a 2-dimensional grid of size 5x5. A robot is placed randomly on this grid and must move to a random target destination. The start and target location are guaranteed to be different. There are 4 possible movements: up, down, left and right. If the robot tries to go off the grid, it remains at the ...
hands-on-ai
lib/data/question-bank.json
JSON
3b0c68cf3c57c12ba2e8c21d970e2704dfaac7ee5bd9274a34b3f90d33c69d4f
6
896
real_closed_book_exam", "sourceRefs": [ { "sourceId": "exam-1-2026-page-02", "pages": [ 2 ], "questionNumber": "5" }, { "sourceId": "exam-1-2026-pdf", "pages": [ 2 ], "questionNumber": "...
hands-on-ai
lib/data/question-bank.json
JSON
36dc1deb546d2c6fb6c9154387b96e7bffb1b28aea5745b2f399d76df06fdeef
7
896
"correctChoiceIds": [ "c3" ], "answerSource": "expert_verified", "confidence": 0.96, "examMode": "closed_book", "sourcePriority": "primary", "approvalStatus": "approved", "selectionRationale": "Transcribed from the Exam 1 2026 screenshot and checked against the supplied...
hands-on-ai
lib/data/question-bank.json
JSON
2cf56c3f464144234cdb97d8807e9e92d912ab6fec1c47b2a87fd7f88b5569fd
8
896
], "difficulty": "medium", "examStyleConfidence": 0.95, "commonTrap": "Separating convolution effects from pooling, normalization, and general image preprocessing.", "coreClosedBookLikely": true, "cognitiveLevel": "conceptual", "interactionType": "single_choice", "stem": "In co...
hands-on-ai
lib/data/question-bank.json
JSON
96f4606b086302958392b622a1be5fc5cb7a062ec6df22f539f027adffcc1a1a
9
896
"selectionRationale": "Transcribed from the Exam 1 2026 screenshot and checked against the supplied course material.", "explanation": "Projecting to fewer dimensions generally discards information, even when the lower-dimensional representation is useful.", "tags": [ "ai-recap-data", "exam-2...
hands-on-ai
lib/data/question-bank.json
JSON
199f637f1975573166294652d148331970ccae141bb5dd944a8bcdab0508688c
10
896
up input length, output length, hidden state size, and feature dimension.", "coreClosedBookLikely": true, "cognitiveLevel": "conceptual", "interactionType": "single_choice", "stem": "Which of the following statements is true about sequence data input and standard, feed-forward neural networks?",...
hands-on-ai
lib/data/question-bank.json
JSON
b6b7108adcd707cb465633669c7883d6349e6b69842201801032b3efc584d391
11
896
and checked against the supplied course material.", "explanation": "Back-propagation through time trains recurrent networks by unrolling the recurrent computation across sequence steps.", "tags": [ "sequence-models", "exam-2026", "closed-book" ], "retrievalText": "Back-pr...
hands-on-ai
lib/data/question-bank.json
JSON
4c72b9de0dff429f3d4546867126fe559d78bf1821ca62c398c390cbdd5528c2
12
896
: "14" } ], "topicIds": [ "language-models" ], "lectureIds": [ "summary-unit-02", "unit-04" ], "difficulty": "medium", "examStyleConfidence": 0.95, "commonTrap": "Confusing embeddings, encoders, decoders, and sampling settings.", "cor...
hands-on-ai
lib/data/question-bank.json
JSON
1236efccdc31f2094e3c0f0a97d2d56d25d243c97d2f8808c325af186c8bc112
13
896
course material.", "explanation": "Q(s, a) estimates the expected return after taking action a in state s and then following a policy.", "tags": [ "reinforcement-learning", "exam-2026", "closed-book" ], "retrievalText": "Which of the following statements is true about Q-v...
hands-on-ai
lib/data/question-bank.json
JSON
de8d5233a92807802a08cb9bbc14e284b073dc7e35700a83d8ff7c3998f2a7c1
14
896
and policy choice.", "coreClosedBookLikely": true, "cognitiveLevel": "conceptual", "interactionType": "single_choice", "stem": "Which of the following statements is true about deep Q-learning?", "choices": [ { "id": "c1", "text": "Deep Q-learning can only be use...
hands-on-ai
lib/data/question-bank.json
JSON
ce824b450654933541103f7b5de757ce2a2efe07a0ae4228a427f8956e0e38db
15
896
layers or time steps can shrink gradients.", "tags": [ "neural-networks", "exam-2026", "closed-book" ], "retrievalText": "Which of the following statements is true about the vanishing gradient problem? The vanishing gradient problem can be avoided by rescaling the input values....
hands-on-ai
lib/data/question-bank.json
JSON
40fa62dea2b7d8500e91acaff069c983454dcd87d6d625daf9ec33a7328eb85f
16
896
decoders, and sampling settings.", "coreClosedBookLikely": true, "cognitiveLevel": "conceptual", "interactionType": "single_choice", "stem": "What is the general data flow through a Transformer language model with a full encoder-decoder stack?", "choices": [ { "id": "c1",...
hands-on-ai
lib/data/question-bank.json
JSON
cd79c3a75453daf9ff7e86e4e1b00769123c6514fad916e3abd1dc9c81dffb2a
17
896
, "tags": [ "reinforcement-learning", "exam-2026", "closed-book" ], "retrievalText": "Which of these is not part of a Markov decision process? States Rewards Actions Losses MDPs are described through states, actions, transition dynamics, and rewards. A loss is an optimization c...
hands-on-ai
lib/data/question-bank.json
JSON
8b023fb3296cfe7244758119ad9e16dcc3161af73530bcb8f89c980b9eba7d06
18
896
: "passed through an element-wise non-linearity, multiplied by a weight matrix and added to bias weights." }, { "id": "c2", "text": "multiplied by a weight matrix, added to bias weights and passed through an element-wise non-linearity." }, { "id": "c3", ...
hands-on-ai
lib/data/question-bank.json
JSON
027bd39c18b99326fc019f70d3ae3090eaa3f57c3120388f0915b40fe03a9273
19
896
2026 screenshot and checked against the supplied course material.", "explanation": "The sigmoid derivative is positive and at most 1/4. At input 45 it is also very close to zero, but still positive.", "tags": [ "neural-networks", "exam-2026", "closed-book" ], "retrievalTe...
hands-on-ai
lib/data/question-bank.json
JSON
15614dd7ec8b2539e009d9f89d44158ae28278f11c1c5d61fe110e97c0ac0ff9
20
896
origin": "real_closed_book_exam", "sourceRefs": [ { "sourceId": "exam-1-2026-page-09", "pages": [ 9 ], "questionNumber": "26" }, { "sourceId": "exam-1-2026-pdf", "pages": [ 9 ], "questio...
hands-on-ai
lib/data/question-bank.json
JSON
10f2d8f68a78c8f3c47f1dff5413e911e23af77cf3ba714e873fa221394b00b8
21
896
the same as regular feed-forward neural networks in case the sequence length is fixed." }, { "id": "c3", "text": "They cannot process sequences of fixed length." }, { "id": "c4", "text": "They are suitable for processing sequences of variable lengt...
hands-on-ai
lib/data/question-bank.json
JSON
cd99f1b3fde648439d24ac0309996bae1e25f75c10e14e7d3fa3040f747b67f3
22
896
], "questionNumber": "29" } ], "topicIds": [ "language-models" ], "lectureIds": [ "summary-unit-02", "unit-04" ], "difficulty": "medium", "examStyleConfidence": 0.95, "commonTrap": "Confusing embeddings, encoders, decoders, and sa...
hands-on-ai
lib/data/question-bank.json
JSON
a97be9b534d79c6eaa832f1e15147d986a23e6a4977cefce26b0a4aecb03a840
23
896
": "primary", "approvalStatus": "approved", "selectionRationale": "Transcribed from the Exam 1 2026 screenshot and checked against the supplied course material.", "explanation": "Dropout randomly disables units during training. At inference time, the full network is used with the appropriate scaling b...
hands-on-ai
lib/data/question-bank.json
JSON
1449858fe3ed2752266527b4ce00d98561c7fd36172b79dc4ea106f7b686480a
24
896
, "topicIds": [ "drug-discovery" ], "lectureIds": [ "summary-unit-02", "unit-05" ], "difficulty": "medium", "examStyleConfidence": 0.95, "commonTrap": "Distinguishing structure/activity prediction from direct experimental molecule or protein measurement....
hands-on-ai
lib/data/question-bank.json
JSON
484a8006136c4c10e4cde2dacb6ae9cb79e71efdd49556ab3ec0167b695b7489
25
896
.", "explanation": "Alpha is the learning-rate factor. Values between zero and one mix the old estimate with the new target.", "tags": [ "reinforcement-learning", "exam-2026", "closed-book" ], "retrievalText": "Consider the update function used during Q-learning: $Q(s_t,a...
hands-on-ai
lib/data/question-bank.json
JSON
48e6da410f222a607460ea8d0dcf892f56c81ced1ca002a4b04a36ca40281f24
26
896
of the loss with respect to an RNN’s weight. is commonly used to train an RNN. is the method used to train recurrent neural networks. generates (potentially) very deep networks. to optimize the parameters of an RNN. Correct according to the supplied historical answer bank: is commonly used to train an RNN.; to optimize...
hands-on-ai
lib/data/question-bank.json
JSON
efe2cf22cf36fed0bf1d84dcf339202a759004580f234f02353ae14af3e0a61d
27
896
"interactionType": "multi_select", "stem": "Which of the following statements is/are true about the vanishing gradient problem?", "choices": [ { "id": "c1", "text": "Repeated multiplication of gradients smaller than 1 leads to a vanishing gradient." }, { ...
hands-on-ai
lib/data/question-bank.json
JSON
f84ec6705d2302991b6c9969dd4375b29a18ef0a3a2e54e2eef0b12d46554b36
28
896
"id": "c2", "text": "rewards can be negative." }, { "id": "c3", "text": "rewards can be positive." }, { "id": "c4", "text": "rewards can be 0." }, { "id": "c5", "text": "reward is associated with the tr...
hands-on-ai
lib/data/question-bank.json
JSON
2d346b07de01ccd108751b691a23a2dc663c48fa3c04b74595fef706d3a46e3f
29
896
exam marker(s): (24), (24 R2), (24 R1), (23), (23 R1), (22 R1).", "explanation": "Correct according to the supplied historical answer bank: SMILES is not the only form of representing a molecule.; A molecule can have different SMILES representations.; 3D structure cannot be fully represented using SMILES.; A mole...
hands-on-ai
lib/data/question-bank.json
JSON
5e64160cb164655d00c5c97012cc0f49acd54352d83e43c08961d868c472bebb
30
896
position of a sequence.; They are suitable for processing sequences of constant dimensionality (i.e. number of features)..", "tags": [ "sequence-models", "historical-bank", "repeat-6" ], "retrievalText": "Which of the following statements is/are true about recurrent neural netw...
hands-on-ai
lib/data/question-bank.json
JSON
a1f63617ad2fbe11b514aad4c84c570955d86545716ee4c8cb882da60e777af8
31
896
can be multiple optimal policies. It maximizes the sum of rewards. Correct according to the supplied historical answer bank: There can be multiple optimal policies.; The optimal policy may be obtained via Q-learning.; It maximizes the cumulative reward.; Continuously choosing the actions with the highest immediate rewa...
hands-on-ai
lib/data/question-bank.json
JSON
b455ec4549ca979426229fba6a92b95a3adc706db2aa3a6e2418c423baa78f64
32
896
id": "hist-q080", "origin": "real_open_book_exam_approved", "sourceRefs": [ { "sourceId": "docx-answer-bank", "questionNumber": "80", "note": "Section: WORD EMBEDDING; prior markers: (24 R2), (24 R1), (23 R1), (22), (22 R1)" } ], "topicIds": [ ...
hands-on-ai
lib/data/question-bank.json
JSON
53d73d0e447f95d103f61c1a330953744a997ac2dacf11c70c1f78d231ab4ef3
33
896
24 R1), (23), (23 R1), (22 R1)" } ], "topicIds": [ "neural-networks" ], "lectureIds": [ "summary-unit-01", "summary-unit-02", "unit-01", "unit-02" ], "difficulty": "medium", "examStyleConfidence": 0.72, "commonTrap": "Rememb...
hands-on-ai
lib/data/question-bank.json
JSON
7616f15580789f24a132bbc5ce8283c94ee8d71d6eda5bba83e68696da45df17
34
896
"interactionType": "multi_select", "stem": "The 3D structure of a molecule...", "choices": [ { "id": "c1", "text": "is ambiguous determined by a molecular graph." }, { "id": "c2", "text": "determines the functionality to a high degree." ...
hands-on-ai
lib/data/question-bank.json
JSON
4b705d7c16798e09bacb4261357d527c75ad6b428f68df4890fd4e8167ee13c1
35
896
bind/react with the target of interest..", "tags": [ "drug-discovery", "historical-bank", "repeat-5" ], "retrievalText": "Which of the following statements is true about virtual screening (VS)? In VS, a trained neural network predicts whether a molecule in the database is likel...
hands-on-ai
lib/data/question-bank.json
JSON
877bc4e622d8ab5ec8661eb230889b8222439368e4faba59759fed4bf47ca2e1
36
896
", "coreClosedBookLikely": false, "cognitiveLevel": "conceptual", "interactionType": "multi_select", "stem": "Which of the following statements are true about gating/gates in an LSTM?", "choices": [ { "id": "c1", "text": "The parameters of the gates are optimize...
hands-on-ai
lib/data/question-bank.json
JSON
96088597f9276163ea2e72b9fcad23290417f69b734a1b93fa560c08adfdf6df
37
896
"correctChoiceIds": [ "c1", "c2", "c3", "c4", "c5" ], "answerSource": "explicit_pdf", "confidence": 0.88, "examMode": "open_book_reference", "sourcePriority": "supporting", "approvalStatus": "approved", "selectionRationale": "Converted fr...
hands-on-ai
lib/data/question-bank.json
JSON
4d142f72a716b1c3c615d50b17d76712e6ab4d62d15dbc81fa3a13b87fcae46f
38
896
the identity function.; For positive inputs, ReLU is equivalent to Leaky ReLU.; The derivative is easy to compute..", "tags": [ "neural-networks", "historical-bank", "repeat-4" ], "retrievalText": "Which of the following statements are true about the ReLU activation function? F...
hands-on-ai
lib/data/question-bank.json
JSON
3ef7f696112c9d4fd03e26c94527b009aa5f335565dd9a623d7b9f513eaf0e8e
39
896
": "single_choice", "stem": "Which of the following formulae is correct regarding the backward pass in a neural network layer, given some input vector x, weight matrix W, bias b and non-linear function f? You can assume that all vectors and matrices are in correct shape.", "choices": [ { "...
hands-on-ai
lib/data/question-bank.json
JSON
896a2d3026429fea9f361bcb4cb2616c7835ad98cd7abf6a35117fce4272f5e0
40
896
responsible for going from the old cell state to the new cell state.; is responsible for countering the vanishing gradient problem.; is, together with the gating, the most important part of an LSTM..", "tags": [ "sequence-models", "historical-bank", "repeat-4" ], "retrievalText...
hands-on-ai
lib/data/question-bank.json
JSON
2f1dc35466d8aefeace6e1a70ddc5732725a5fae046f504d4fa9d11fb7e8a32e
41
896
"Consider the update function used duringQ-learning: Q(s , a ) ← (1 - α) * Q(s , a ) + α * (r + γ * max Q(s , a ))Which of the following statements are true? γ = 1 means to put a strong emphasis on future rewards. α is the learning rate factor. γ = 0 means to only take the immediate reward into account. α = 0 means tha...
hands-on-ai
lib/data/question-bank.json
JSON
340d104d751093640260dd010edb55222f7c5f86bc7ec7c3011122a916e20878
42
896
. Language modeling is the task of predicting a word given a context. In order to generate text using an already trained RNN language model, one can sample from the output probability distribution. The sample word from this distribution is used as the next word to generate the word afterwards. Correct according to the ...
hands-on-ai
lib/data/question-bank.json
JSON
151407fdd8f3a6974ae7a8e3b917519de4469f053bfd05913bf1ba889d549cfc
43
896
": "The size of the weight matrix is independent of the sequence length." } ], "correctChoiceIds": [ "c1", "c2" ], "answerSource": "explicit_pdf", "confidence": 0.88, "examMode": "open_book_reference", "sourcePriority": "supporting", "approvalStatu...
hands-on-ai
lib/data/question-bank.json
JSON
14aa1ca1fd33ea162ab9b5519de0d8abe207f2b6ca292eee70533c522e104cea
44
896
-01", "unit-02" ], "difficulty": "easy", "examStyleConfidence": 0.72, "commonTrap": "Remember whether a statement is about the activation, the gradient, or the optimization objective.", "coreClosedBookLikely": false, "cognitiveLevel": "conceptual", "interactionType": "m...
hands-on-ai
lib/data/question-bank.json
JSON
1b49260837ec1cd6664a72d08098a4883ecf7c5f359769b5c3c5ede79c93a4e4
45
896
"text": "might have several local minima." }, { "id": "c5", "text": "is common in deep learning." } ], "correctChoiceIds": [ "c1", "c2", "c3", "c4", "c5" ], "answerSource": "explicit_pdf", "confidence": 0.8...
hands-on-ai
lib/data/question-bank.json
JSON
79b419140ad8598e3b9ac68655b1b77185bc63eb77933b99ad9a5c577d0bd3a3
46
896
: SUPERVISED MACHINE LEARNING; prior markers: (24), (23), (22 R1)" } ], "topicIds": [ "ai-recap-data" ], "lectureIds": [ "summary-unit-01", "unit-01" ], "difficulty": "easy", "examStyleConfidence": 0.72, "commonTrap": "Confusing a useful pr...
hands-on-ai
lib/data/question-bank.json
JSON
1c24df05afba0e512ff01bad32f12d9b7deb403356d61ee08f54ca3829477cd3
47
896
", "repeat-3" ], "retrievalText": "The goal of Q-learning is to... approximate the Q-values of the state-action pairs of an MDP (Markov decision process). Correct according to the supplied historical answer bank: approximate the Q-values of the state-action pairs of an MDP (Markov decision process)....
hands-on-ai
lib/data/question-bank.json
JSON
9d0eee0d339b072e44f6436d3fedc3c87b0921504300f9ab5daf685d38c5e179
48
896
24 R2), (22), (22 R1).", "explanation": "Correct according to the supplied historical answer bank: multiplied by a weight matrix, added to bias weights and passed through an element-wise non-linearity..", "tags": [ "neural-networks", "historical-bank", "repeat-3" ], "retr...
hands-on-ai
lib/data/question-bank.json
JSON
e54d48088d0f6a5fa72043d5a6a1d19b85b9d6027fe186feeaf392398e5c5b85
49
896
", "text": "input -> encoder -> Self-Attention -> Feed-Forward -> decoder" } ], "correctChoiceIds": [ "c1", "c2" ], "answerSource": "explicit_pdf", "confidence": 0.88, "examMode": "open_book_reference", "sourcePriority": "supporting", "ap...
hands-on-ai
lib/data/question-bank.json
JSON
45fb62444ce6c4a66d6e4ff10b204db66e895c3f28f34f0cac70af390136b226
50
896
displayed in line plots. Every data can theoretically be represented as a sequence. There are dedicated machine learning models that deal with sequence data. Typically, there is some sort of order in the data (time, position, etc.). Correct according to the supplied historical answer bank: Sequence data is typically di...
hands-on-ai
lib/data/question-bank.json
JSON
da66d4985ba66caf1418d32bd2614967dcb78f9d465c6649a9ca6d6068a44aed
51
896
"id": "hist-q074", "origin": "real_open_book_exam_approved", "sourceRefs": [ { "sourceId": "docx-answer-bank", "questionNumber": "74", "note": "Section: RNN TYPES; prior markers: (24), (22)" } ], "topicIds": [ "sequence-models" ], ...
hands-on-ai
lib/data/question-bank.json
JSON
7f82cb0b0fc2932112b45b7f11995705d8369435cab94e01a98bcc2e3c42bb22
52
896
"summary-unit-02", "unit-06" ], "difficulty": "medium", "examStyleConfidence": 0.72, "commonTrap": "Separating immediate reward, cumulative return, Q-values, and policy choice.", "coreClosedBookLikely": false, "cognitiveLevel": "conceptual", "interactionType": "multi_se...
hands-on-ai
lib/data/question-bank.json
JSON
ea55b5ec91e2dcef907be0253cc952154857e5be72c276b7df11f70aea3a428f
53
896
current highest Q-value." }, { "id": "c4", "text": "Depending on the environment, choosing a suboptimal action might lead to Q-learning getting stuck." } ], "correctChoiceIds": [ "c1", "c2", "c3", "c4" ], "answerSource":...
hands-on-ai
lib/data/question-bank.json
JSON
d1b97ab60cbf2501c07e9c3cdd660ab6fc19fe9c353a91aaff4c3468cf9d9109
54
896
historical-bank", "repeat-2" ], "retrievalText": "Which of the following statements are correct regarding generative models in the area of drug discovery? Generative models generate new molecules according to the target or desired bioactivity. Generative models are a promising addition since the che...
hands-on-ai
lib/data/question-bank.json
JSON
5df668d81568936a7645e8931e3068c79e6a38bb3513dc37ba3a0379a927507a
55
896
"unit-02" ], "difficulty": "medium", "examStyleConfidence": 0.72, "commonTrap": "Remember whether a statement is about the activation, the gradient, or the optimization objective.", "coreClosedBookLikely": false, "cognitiveLevel": "conceptual", "interactionType": "multi_select"...
hands-on-ai
lib/data/question-bank.json
JSON
637fd9518514d1cf7f513be79217c53b8928ce8440044e47ddb1d501dfaf8b03
56
896
": [ "ai-recap-data", "historical-bank", "repeat-2" ], "retrievalText": "Which of the following statements are correct regarding tabular data?... Columns in the table are called features. Rows in the table are called samples. Tabular data might contain duplicate rows. There can be mo...
hands-on-ai
lib/data/question-bank.json
JSON
7c764e49b91180d4807649abdb24015d340f81635d17df3d2174d032f5542c04
57
896
-q009", "origin": "real_open_book_exam_approved", "sourceRefs": [ { "sourceId": "docx-answer-bank", "questionNumber": "9", "note": "Section: SUPERVISED MACHINE LEARNING; prior markers: (24 R1), (23 R1)" } ], "topicIds": [ "ai-recap-data" ...
hands-on-ai
lib/data/question-bank.json
JSON
15c27cdcaaf650167821efa52361480413ec2ecbd3c312b38147f42f3c57a777
58
896
"explanation": "Correct according to the supplied historical answer bank: It allows to get a better estimate of the generalization error.; It allows to detect overfitting..", "tags": [ "ai-recap-data", "historical-bank", "repeat-2" ], "retrievalText": "Why is it a good idea to ...
hands-on-ai
lib/data/question-bank.json
JSON
4f996283f193ff80e2e49be2fceb418275560adf92881e184bb9e140ea9c7d13
59
896
Section: ACTIVATION FUNCTIONS; prior markers: (24), (23)" } ], "topicIds": [ "neural-networks" ], "lectureIds": [ "summary-unit-01", "summary-unit-02", "unit-01", "unit-02" ], "difficulty": "medium", "examStyleConfidence": 0.72, ...
hands-on-ai
lib/data/question-bank.json
JSON
23560b199ae86870d54c975170510d45fd7848e8630b91e3c5caa7aa58d11845
60
896
hidden state (h0) is typically set to 0. The input sequence must be fed timestep by timestep to the RNN. Correct according to the supplied historical answer bank: The input sequence must be fed timestep by timestep to the RNN.; The initial hidden state (h0) is typically set to 0.; There is only one weight matrix that i...
hands-on-ai
lib/data/question-bank.json
JSON
09dc5763b3f766e0532d0badff72f48c1f4cd3f764360d698011c1e97408fd52
61
896
layers." } ], "correctChoiceIds": [ "c1", "c2" ], "answerSource": "explicit_pdf", "confidence": 0.88, "examMode": "open_book_reference", "sourcePriority": "supporting", "approvalStatus": "approved", "selectionRationale": "Converted from the s...
hands-on-ai
lib/data/question-bank.json
JSON
50744a64e6c1554f992fe92b9e9945afdd9e84bd9c6ef3e043c63c80c6afd526
62
896
eases the computational demand.; can reduce the computational load and memory requirements.; is common in convolutional neural networks.." }, { "id": "hist-q002", "origin": "real_open_book_exam_approved", "sourceRefs": [ { "sourceId": "docx-answer-bank", "questionNu...
hands-on-ai
lib/data/question-bank.json
JSON
34da3cf0fb59dcfb0255cf811df6bbceaf7dec71668a3cf26eb7e5e3a45544d0
63
896
the one-to-many RNN type?", "choices": [ { "id": "c1", "text": "Given an input of length Tx = 1, the output/prediction length is Ty > 1." } ], "correctChoiceIds": [ "c1" ], "answerSource": "explicit_pdf", "confidence": 0.88, "examMode...
hands-on-ai
lib/data/question-bank.json
JSON
98a65a1991d413527d6b440d70291cabb99f2eaf255c98bee88b00090f891acf
64
278
export type QuestionOrigin = | "real_closed_book_exam" | "real_open_book_exam_approved" | "generated_from_slides" export type AnswerSource = | "explicit_pdf" | "expert_verified" | "inferred_from_slides" | "inferred_from_exam_context" | "generated" export type InteractionType = | "true_false" | "si...
hands-on-ai
lib/data/schema.ts
TypeScript
1b33f9006f93393eb27d900cac4489102082330fa6430d0bdd529dd61209cf54
0
475
{ "version": "1.0.0", "generatedAt": "2026-07-07T20:06:48.206801+00:00", "chunks": [ { "id": "summary-pdf-p002", "sourceId": "summary-pdf", "lectureId": "summary-unit-02", "page": 2, "title": "Summary page 2", "text": "Unit 1 – Recap of Hands-on AI I Tabular Data • Data org...
hands-on-ai
lib/data/slide-index.json
JSON
2505e171a87e3d96d044734b1b4768385aedfc68df3d8607509a0db32314125d
0
896
an. • Every dataset can theoretically be represented as a sequence. • Typically has an order: time, position, word order, etc. • Commonly visualized in line plots. • Dedicated ML models (RNNs, LSTMs) handle sequence input. • Feed-forward NNs: sequence relations are lost. • Variable-length sequences must be preprocessed...
hands-on-ai
lib/data/slide-index.json
JSON
19ff0e19061237d0ea5c126d00aadd6f245a99993c5c1be510f12c2024c189e0
1
896
}, { "id": "summary-pdf-p006", "sourceId": "summary-pdf", "lectureId": "summary-unit-02", "page": 6, "title": "Summary page 6", "text": "Gradient Descent • Iterative optimization method. • Update rule: • Learning rate η: controls step size. • Moves in direction of negative gradie...
hands-on-ai
lib/data/slide-index.json
JSON
bad4c3657d7a0588d28ca778792b0d33be502eb0d51a3e4e27bafa03b568bb70
2
896
not shown in exam statement, but true). 2. 50×50 image, 4 channels, 2 kernels (3×3) • Input: 50×50, depth=4. • Kernel: 3×3×4.", "keywords": [ "state", "kernel", "feature map", "image" ], "retrievalText": "Convolutional Networks. Let’s apply these rules: 1. 30×30 image, ...
hands-on-ai
lib/data/slide-index.json
JSON
740ac553dd45291101d2928e08bf6c276dc1721c33348a61fe86522d36c68bdf
3
896
. 5. 8×8 image, 1 channel, 2 kernels (4×4) • Input: 8×8, depth=1. • Kernel: 4×4×1. • 2 kernels → 2 feature maps. • Resolution: 8−4+1=58 - 4 + 1 = 58−4+1=5. • So → 2 maps of size 5×5. 6. 6×6 image, 3 channels, 1 kernel (3×3) • Input: 6×6, depth=3." }, { "id": "summary-pdf-p010", "sourceId": "summary-...
hands-on-ai
lib/data/slide-index.json
JSON
62ca1358c8bdf17680aaada5d9a20ba45c093be318f48487208b72aa1b671929
4
896
= 2.4 Hidden activations: [1.8, 2.4] Step 2: Calculate output y=h1∗w1+h2∗w2+bias Output neuron weights: w1=0.7, w2=0.5, bias=0.1 y = 1.8*0.7 + 2.4*0.5 + 0.1 = 2.56 • Predicted score = 2.56 Compare to true score (calculate error) • True score = 3 • Error = 3 - 2.56 = 0.44 • Loss function: Mean Squared Error (MSE) → (0.4...
hands-on-ai
lib/data/slide-index.json
JSON
b4cc3332e58dd3f985e3798a902643d9a9fe5e9d80084c9d2eec13297c568ba1
5
896
(since distributions change). o At test time, use statistics collected from the training set. o Noise introduced by batch statistics can help prevent overfitting. Regularization • Which of the following statements are true about L1 and L2 regularization? o L1 regularization: adds the sum of absolute weights to the loss...
hands-on-ai
lib/data/slide-index.json
JSON
0b66e5564b46a4d0a8e51f43b35595b9573ae2082e8cadee8b2422f289f636b8
6
896
, "title": "Summary page 16", "text": "o A large learning rate helps exploration; a small one helps fine-tuning near minima. o Starting large and decaying the rate balances exploration vs. exploitation. o Common schemes include step decay, exponential decay, and linear decay. Convex vs. Non-convex Functions...
hands-on-ai
lib/data/slide-index.json
JSON
97b2d95423d6b11336f5ac45ebe4bfb21786683ac4fb5c986e8b115d2ab451d9
7
896
p018", "sourceId": "summary-pdf", "lectureId": "summary-unit-02", "page": 18, "title": "Summary page 18", "text": "UNIT 3 — Recurrent Neural Networks Sequence data • A sample s is a sequence of length T with D features at every timestep. • D (feature dimensionality) is constant across samp...
hands-on-ai
lib/data/slide-index.json
JSON
f00f842cd756fe25ca39834c2b3ca144de3d7207975a5ac203e783ac89fae18e
8
896
be unrolled in time into a deep feed-forward graph with shared weights. • Training commonly uses Backpropagation Through Time (BPTT): compute gradients on the unrolled graph and update shared weights. • BPTT makes the effective depth proportional to sequence length, which can cause vanishing or exploding gradients. Van...
hands-on-ai
lib/data/slide-index.json
JSON
c696aac2902e455ef6857586b24f9b779263d2ba7062120df97cdaff13c37d64
9
896
1) — e.g., image captioning, music generation (single input → sequence output). • Many-to-One (Tx>1, Ty=1) — e.g., sentiment classification (sequence → single label).", "keywords": [ "state", "sequence", "rnn", "lstm", "gate", "gating", "hidden state", ...
hands-on-ai
lib/data/slide-index.json
JSON
0487e52b5a7e1977a1b0b2f429046a08b93e1535aa943faf3476054d674e0189
10
896
, "transformer", "sequence", "rnn", "lstm", "hidden state", "gradient" ], "retrievalText": "Reinforcement Learning and Q-Learning. • Many-to-Many equal (Tx=Ty) — e.g., POS tagging, Named Entity Recognition (sequence → sequence, same length). • Many-to-Many dif...
hands-on-ai
lib/data/slide-index.json
JSON
ef56b9fb38a9c482859d83dfeb3952bf2541e67a014ab46dc80349ca052655ad
11
896
losses over sequence/batch. Q: What is perplexity? • Perplexity = exp(L). • Lower = better model. • Interpreted as geometric mean of token probabilities. • Automatically normalized for sequence length. • Not a measure of semantic quality. Q: Which statements about text generation are true? • Sampling uses predicted pro...
hands-on-ai
lib/data/slide-index.json
JSON
fb8702d4e547bb73fd22d18c92699a2be96af3d1f3537be97c741bdead40f834
12
896
Which statements about word2vec are true? • Similar words map to similar regions. • Similarity measured via cosine similarity. Q: What is tokenization in modern LMs? • Tokens can be subwords/characters. • Handles out-of-vocabulary words. • Examples: BPE, SentencePiece. RNN Language Models Q: What is the data flow in an...
hands-on-ai
lib/data/slide-index.json
JSON
8c58e755e9fc7060d1f8a4555d2e246014671a643a6d68915b42a98a86cae309
13
896
.g., translation). Q: Masked vs unmasked attention? • Unmasked: encoder, all tokens can attend to all others. • Masked: decoder, tokens attend only to previous ones (for autoregressive generation). Scaling & Energy (could be asked) Q: How do model parameters scale?", "keywords": [ "state", "embedd...
hands-on-ai
lib/data/slide-index.json
JSON
e0efbfdd7d1babb39d34e2a85c7cdb83d512e034b5a7b632f8fa6806099e7074
14
896
many years. • is an extremely extensive process.", "keywords": [ "drug", "molecule", "state" ], "retrievalText": "AI for Drug Discovery. Unit 5 - DRUG DISCOVERY Q: Which of the following statements are correct regarding generative models in the area of drug discovery? • Generat...
hands-on-ai
lib/data/slide-index.json
JSON
666f454f7f8ec96a5ee827dca4fbf0b85df5c48f0397b6c53efdbbb51df37193
15
896
the order of 10^60. TARGET PREDICTION (HTS vs VS) Q: High-Throughput Screening (HTS) • HTS is experimental testing of many compounds in assays; it is expensive and time-consuming. • HTS provides the raw experimental data that populates public assay databases. Q: Virtual Screening (VS) • VS uses computational models (e....
hands-on-ai
lib/data/slide-index.json
JSON
bd5c2fafce2f0a11639af53f0cdf960cdb088144be0b0abfd40f00f1bb1f865f
16
896
encode 3D conformation. • A molecular graph can be transformed into a SMILES string and vice versa. Q: MOL / SDF files • MOL / SDF are file formats that encode atoms, coordinates and bond information." }, { "id": "summary-pdf-p032", "sourceId": "summary-pdf", "lectureId": "summary-unit-02", ...
hands-on-ai
lib/data/slide-index.json
JSON
d1f6573034f4c6c4b0d6757b6264f154ba0594845d521eb0b0510e78b7794499
17
896
made of amino-acid sequences (typically from ~50 up to ~2000 residues) performing diverse biological functions (enzymes, signaling, structure, transport, etc.). Q: Protein folding problem • The protein folding problem: predict a protein’s 3D structure from its amino-acid sequence. • The number of possible folds/configu...
hands-on-ai
lib/data/slide-index.json
JSON
2084bdbe45c6882fa93eafce5dbd61234f8ad109830ab31083dfa07a6061536f
18
896
Key ready facts to memorize (short bullets) • Drug development costs ≈ $0.5–2B; discovery stage ≈ 30% of cost; typical success rate ≈ 10%. • Eroom’s law: drug-discovery costs have been increasing (roughly doubling every ~10 years). • Chemical space for drug-like molecules is astronomically large (≈ 10^60 for ~40- atom ...
hands-on-ai
lib/data/slide-index.json
JSON
7bb61f01c9939e0e20c7901311d761ce033ef9588994a3836fdabd4b14f3fc91
19
896
Find a policy π(s) mapping states → actions that maximizes cumulative reward. Q: What is a common challenge in RL? • Reward signals are often sparse (only some state transitions give feedback). The Goal of RL Q: What is a policy? • The behavior function of an agent, specifying actions in each state. Q: What is the goal...
hands-on-ai
lib/data/slide-index.json
JSON
ed0be69891c99e5ffd3e4ee1cd89d978f3e717982bdcb2c0a63cb39e0e5b5eb7
20
896
4. Update Q(s,a) using the update rule. 5. Repeat until convergence. Q: How to balance exploration and exploitation? • Use ε-greedy: with probability ε pick random action (exploration), else pick best action (exploitation). Limits of Q-Learning Q: Why doesn’t Q-learning scale well? • Large state/action spaces → Q-table...
hands-on-ai
lib/data/slide-index.json
JSON
55176af3f4cce57a6cd8f5d24f95a512b16ea0c0d39008cce94ac55bcdf8d733
21
896
RL Q: What is the difference between narrow and general RL systems? • Narrow AI: works well in games/environments with clear rules, rewards, and full observability. • General AI: real-world problems are messy (uncertainty, sparse/ambiguous rewards, long time horizons). Q: Why is real-world RL harder? • Sparse or ambigu...
hands-on-ai
lib/data/slide-index.json
JSON
651799fa48c542c65d491aefc308335e5e5dd616e8fe996fe76f41a096a2cbe9
22
896