AI on GATE DA splits into three clean pillars — search, logic, and reasoning under uncertainty — and each pillar has its own best book. Russell & Norvig handles search and probabilistic reasoning. Kenneth Rosen handles logic far better than R&N for exam-style problems. The third pillar (conditional independence, variable elimination and sampling) is the one aspirants skip most often and regret most on exam day. If you treat AI as "just search algorithms", you are leaving an entire topic group unprepared.
Verify: Confirm the syllabus on the official GATE 2027 syllabus page (IIT Madras). The AI section is unchanged from GATE DA 2026.
Official Syllabus
- Search — informed, uninformed, adversarial
- Logic — propositional, predicate
- Reasoning under uncertainty — conditional independence representation, exact inference through variable elimination, and approximate inference through sampling
The three pillars do not overlap heavily with each other (except the probability link in pillar three), so phased study works naturally. Note: the official syllabus uses "predicate" logic (equivalent to first-order logic / FOL). Bayesian networks provide a useful framework for the listed conditional-independence and inference topics. Markov models are supplementary, not additional mandatory syllabus topics.
The Three-Pillar Map
| Pillar | Topics | Best book | Overlap |
|---|---|---|---|
| Search | Uninformed and informed search (BFS, DFS, UCS, greedy best-first, A*); adversarial search (minimax, alpha-beta); properties and complexity | Russell & Norvig | DSA — BFS, DFS, shortest path |
| Logic | Propositional logic, predicate (first-order) logic, inference, CNF/DNF, resolution | Kenneth Rosen — Discrete Mathematics (logic chapters) | Formal methods |
| Reasoning under uncertainty | Conditional independence, exact inference (variable elimination), approximate inference (sampling) | Russell & Norvig | Probability, ML (naive Bayes) |
Books and Resources
| Resource | Role | Use for | Skip |
|---|---|---|---|
| Russell & Norvig — AI: A Modern Approach | Primary (search + reasoning) | Informed/uninformed/adversarial search; conditional independence, variable elimination and sampling in Bayesian networks | Robotics, perception, NLP chapters |
| Kenneth Rosen — Discrete Mathematics and Its Applications | Primary (logic) | Propositional and FOL, inference, normal forms | Combinatorics beyond GATE scope |
| Official GATE DA PYQs (2024 onwards) | Practice | Search traces, logic conversion, Bayes-net inference | — |
Full book list: GATE DA books and resources guide.
Search — Pillar 1
Uninformed Search
BFS, DFS, depth-limited, iterative deepening, uniform-cost search. Be fluent on completeness, optimality, time and space complexity for each — common conceptual MSQ targets.
Informed Search
Greedy best-first uses only h(n). A* uses f(n) = g(n) + h(n). For the standard tree-search guarantees, assume finite branching, step costs bounded below by a positive constant and an admissible heuristic (one that never overestimates true remaining cost). For graph search without reopening closed nodes, consistency is the stronger condition used for optimality. Always check the algorithm variant and assumptions.
Adversarial Search
Adversarial search is explicitly in the syllabus. Learn minimax for alternating MAX/MIN decisions and alpha-beta pruning for avoiding branches that cannot change the minimax result. Trace backed-up values on small game trees, track alpha and beta bounds, and explain how move ordering changes pruning effort without changing the chosen minimax value.
Algorithm Comparison Card
Print this. It is the single most useful reference for search PYQs.
| Algorithm | Complete? | Optimal? | Time | Space |
|---|---|---|---|---|
| BFS | Yes (finite branching) | Yes (unit costs) | O(bd) | O(bd) |
| DFS | No (infinite spaces) | No | O(bm) | O(bm) |
| UCS | Yes (finite branching, costs ≥ ε > 0) | Yes under those assumptions | O(b1+⌊C*/ε⌋) | O(b1+⌊C*/ε⌋) |
| Greedy | No | No | O(bm) | O(bm) |
| A* | Yes under the assumptions above | Tree: admissible h; graph without reopening: consistent h | Exponential worst-case | Exponential worst-case |
Logic — Pillar 2 (Use Rosen)
Propositional Logic
Atomic propositions, connectives (¬, ∧, ∨, →, ↔), truth tables, tautologies, contradictions, satisfiability. Equivalence laws (De Morgan, distribution, contrapositive). CNF and DNF — converting an arbitrary formula to CNF is a common PYQ.
First-Order Logic
Quantifiers (∀, ∃), predicates, variables, constants, functions. Translating English ↔ FOL is the standard PYQ pattern. Be fluent on negation with quantifiers: ¬∀x P(x) ≡ ∃x ¬P(x).
Inference
Modus ponens, modus tollens, hypothetical syllogism, resolution. Resolution refutation: convert to CNF, negate the goal, derive the empty clause. For FOL, add unification.
Why Rosen over R&N here: Rosen's exposition of inference rules and CNF/DNF conversion reads more like an exam-style treatment. R&N's logic chapters lean toward AI-system implementation. Either is correct; Rosen is faster for prep.
Reasoning Under Uncertainty — Pillar 3
Bayesian Networks
DAG where nodes = random variables, edges = conditional dependence, each node has a CPT given its parents. D-separation lets you read conditional independence from the graph structure.
Inference: enumeration (exact, exponential), variable elimination (exact, more efficient), approximate inference through sampling (rejection, likelihood weighting, Gibbs). The prerequisite is Bayes' theorem from probability; naive Bayes from the ML syllabus is essentially a Bayes net with fixed structure.
Note on Markov models and MDPs: The official GATE DA syllabus does not explicitly list Markov chains, HMMs, or Markov decision processes under AI. However, Markov models are a natural extension of the conditional-independence framework the syllabus names, and some coaching material covers them. Focus your time on the explicitly listed topics (conditional independence representation, variable elimination, sampling-based inference) and treat Markov models as supplementary awareness only.
Practice Patterns for Official Topics
Use these exercise categories alongside official PYQs; this is a study checklist, not a claim about frequency or marks:
- Trace BFS / DFS / UCS / A* on a small graph, and minimax/alpha-beta on a game tree
- Identify completeness / optimality / complexity from algorithm definitions
- Translate English ↔ FOL or convert propositional formulas to CNF
- Resolution refutation
- Exact inference by variable elimination and approximate inference by sampling in a small Bayesian network
- Conditional independence from a Bayes-net structure
Solve every AI PYQ from GATE DA 2024 and 2025. Graph search overlaps with DSA — reuse that preparation.
What to Skip
- Robotics — motion planning, kinematics
- Perception — computer vision beyond conceptual exposure
- NLP — parsing, translation, LLMs (the R&N NLP chapters)
- Specialist game-playing systems — supplementary beyond the listed adversarial search topic; minimax and alpha-beta trace practice remain in the core plan.
- Reinforcement learning — MDPs, Q-learning, policy gradients (not in official syllabus)
- CSPs beyond basic exposure
Study Sequence
Weeks 1–2: Search
- Uninformed search — BFS, DFS, UCS, iterative deepening; properties and complexities.
- Informed search — greedy, A*; admissibility and consistency.
- Adversarial search — minimax values and alpha-beta pruning on game trees.
- Trace practice on small graphs.
- Build the comparison card (table above).
Weeks 3–4: Logic (Rosen)
- Propositional logic — connectives, truth tables, equivalences, CNF/DNF.
- FOL — quantifiers, English ↔ FOL translation.
- Inference — modus ponens/tollens, resolution.
Weeks 5–6: Reasoning Under Uncertainty
- Conditional independence representation using Bayesian-network structure and d-separation.
- Exact inference through variable elimination — factor products, marginalisation and elimination order.
- Approximate inference through sampling — estimate query probabilities and compare rejection sampling, likelihood weighting and Gibbs sampling as supporting methods.
Week 7: PYQs and Revision
- Every AI PYQ from GATE DA 2024 and 2025.
- 2–3 topic-wise tests from The ML Hub GATE DA test series.
- Finalise comparison card, logic-equivalence sheet, Bayes-net inference template.
Traps to Avoid
- Using R&N for everything including logic. Rosen is faster for exam prep; use each book in its strength zone.
- Skipping reasoning under uncertainty. Conditional independence representation, variable elimination and sampling-based inference are explicitly listed; Bayesian networks help you study them.
- Memorising search complexities without trace practice. PYQs test behaviour, not memorised tables.
- Confusing admissibility and consistency. Admissibility: h never overestimates. Consistency: adds triangle inequality.
- Missing the DSA overlap. Graph traversal can appear under AI or DSA.
- Over-studying Markov models / MDPs. Not explicitly in the official syllabus; keep to supplementary awareness.
Find your weak pillar
The AI module in our free GATE DA demo course includes lectures on search, logic, and Bayesian networks plus a benchmark test — useful for identifying which pillar needs the most work.
AI in The ML Hub's Course
The AI block in The ML Hub's GATE DA course follows the three-pillar map. Search uses the comparison table directly; logic uses Rosen as the primary reference; reasoning under uncertainty connects to the probability prerequisite. Topic-wise tests include search trace, FOL conversion, and Bayes-net inference problems.
Three pillars. Two books. One structured programme.
AI is structurally clean — but only if you respect the pillar split and use the right book for each.
- Mentor-led lectures: search (R&N), logic (Rosen), Bayesian networks (R&N)
- Search comparison card and Bayes-net inference templates from GATE DA rankers
- Topic-wise tests on search, logic, and reasoning in the test series
FAQs
Which book for GATE DA AI?
Russell & Norvig for search and reasoning under uncertainty. Kenneth Rosen's Discrete Mathematics (logic chapters) for propositional and first-order logic — Rosen's exposition is clearer for exam-style problems.
Is first-order logic in GATE DA?
Yes. Cover quantifiers, predicates, English ↔ FOL translation, negation with quantifiers, and resolution-based inference.
Are Bayesian networks in the GATE DA syllabus?
The official syllabus lists conditional independence representation, exact inference through variable elimination, and approximate inference through sampling — Bayesian networks are the standard framework for these topics. Cover graph structure, d-separation, variable elimination, and sampling-based inference.
Is A* search in GATE DA?
Yes. Cover f(n) = g(n) + h(n), admissibility, consistency, and the resulting completeness/optimality guarantees.
Related Guides
Subjects that share machinery with AI: Python & DSA (graph search overlap) and Machine Learning (naive Bayes, probabilistic reasoning). Full subject map: GATE DA books and resources · GATE DA syllabus 2027 guide.