ISRO Scientist/Engineer 'SC' Computer Science 2025 question paper, solved with the final answer key

ISRO's written test for Scientist/Engineer 'SC' (Computer Science), held on 26 October 2025, question by question. Every answer follows ISRO's final key of 15 May 2026. That key reinstated Q7, excluded Q75 instead, and kept five questions where ISRO accepts either of two answers. Every solution is worked out in full by Skill-Lync, and the answers ISRO changed are marked.

Question paper and answer keys published by ISRO's Centralised Recruitment Board; solutions by Skill-Lync · Updated 18 Nov 2026 · 20 of 80 questions on this page

How many questions were in the ISRO Scientist/Engineer Computer Science 2025 paper?

95 multiple-choice questions in two parts, written on 26 Oct 2025. Part A had 80 computer science questions of 1 mark each, with 0.33 deducted for a wrong answer. Part B had 15 aptitude questions worth 20 marks, with no negative marking; they were the same in every discipline and are solved on the Mechanical paper page. ISRO later excluded Part A Q3, Q50 and Q75 (and aptitude Q82) from evaluation and revised 5 answers after objections; on 5 more (Q13, Q19, Q20, Q54 and Q72) it accepted either of two options.

Which subjects carried the most marks?

Counting the 80 Part A questions by subject: databases 12, information security 10, operating systems 10. Together that is 32 of 80 technical marks. The remaining questions spread over 8 more subjects; the table below links every question.

What was the cut-off for this paper?

In January 2026 ISRO evaluated the Computer Science paper out of 96 (77 + 19) under its December key: the last candidate short-listed for interview, other than PwBD, scored 57.41, and every candidate also needed 38.5 in Part A and 9.5 in Part B; 198 were short-listed. ISRO then issued a final key on 15 May 2026 that reinstated Q7 and excluded Q75 instead (still out of 96) and re-issued the results on 22 May 2026 with 199 short-listed; it did not publish a new cut-off. The panel of 14 Aug 2026 selected 36 plus 3 PwBD, with 18 wait-listed.

Questions by subject, ISRO Computer Science 2025 (Set A), counted by Skill-Lync
SubjectQuestionsQuestion numbers
Databases12232425262742434445464748
Information security10891011121314153638
Operating systems1021226364656667686970
Data structures and algorithms91234717181920
Discrete mathematics and probability95656575859606162
Software engineering81671727374757677
Theory of computation and compilers82830313233787980
Computer networks53537394041
Digital logic55052535455
Programming and languages22934
Computer organisation and architecture24951

ISRO 2025 · Computer Science Set A · Q.1Part A · 1 mark · -0.33 if wrongData structures and algorithms

Inorder traversal of the given binary tree (a to k)

When a tree given in the diagram is traversed inorder, the order would be
abcdefghijk
  1. (a)a, b, d, h, e, i, j, c, f, g, k
  2. (b)h, d, i, j, e, b, f, k, g, c, a
  3. (c)h, d, b, i, e, j, a, f, c, k, g
  4. (d)h, d, b, e, i, j, a, f, c, k, g

Answer: (c) h, d, b, i, e, j, a, f, c, k, g Official final key: (c)

Inorder means: traverse the left subtree, visit the node, then traverse the right subtree.

Read the tree from the figure: has children . has . has only a left child . has . has . has only a left child .

  • Left subtree of (rooted at ): 's left is , so ; then ; then 's subtree . This gives .
  • Then the root .
  • Right subtree of (rooted at ): , then , then 's subtree, where is the left child, so . This gives .

Joined together: . Option (d) is the trap: it visits before , visiting before its left subtree , which is preorder-style order for that node, not inorder.

Where this shows up in industry

Inorder walks are how a sorted index or binary search tree is read out in key order, and the same recursion evaluates and prints expression trees inside compilers.

ISRO 2025 · Computer Science Set A · Q.2Part A · 1 mark · -0.33 if wrongData structures and algorithms

Principle on which Huffman coding works

Huffman coding algorithm works on the principle of
  1. (a)Randomization of symbols
  2. (b)Key based portioning of symbols
  3. (c)Hashing of symbols
  4. (d)Frequencies of input symbols

Answer: (d) Frequencies of input symbols Official final key: (d)

Huffman coding is a lossless compression method. It first counts how often each input symbol occurs. It then repeatedly merges the two least frequent symbols (or merged groups) into one node, until a single binary tree is left.

Reading the path from the root to each symbol gives its code. Frequent symbols end up close to the root and get short codes, rare symbols get long codes, so the average code length is minimal among prefix codes. Everything is driven by the symbol frequencies: nothing is random, nothing is keyed and nothing is hashed.

Frequencies of input symbols

Where this shows up in industry

Huffman coding (usually combined with other steps) sits inside DEFLATE, JPEG and many telemetry and image compression chains, where it trims downlink or storage volume without losing data.

ISRO 2025 · Computer Science Set A · Q.3Part A · 1 mark · -0.33 if wrongData structures and algorithms

Match shortest-travel problems with Floyd, multi-stage graph and Dijkstra algorithms

Match the following table showing algorithm and its application.
(P)A person wants to visit from one place to another place in shortest period(i)Floyd's algorithm
(Q)A person wants to visit M places in shortest period(ii)Multi-stage graph algorithm
(R)All N persons wants to visit all M places in shortest period(iii)Dijkstra's algorithm
  1. (a)P - (ii), Q - (iii), R - (i)
  2. (b)P - (i), Q - (iii), R - (ii)
  3. (c)P - (ii), Q - (i), R - (iii)
  4. (d)P - (i), Q - (ii), R - (iii)

Answer: No option is correct (the textbook matching P - (iii), Q - (ii), R - (i) is not listed) Excluded from evaluation by ISRO

Each algorithm solves a particular shortest-path problem:

  • Dijkstra's algorithm finds the shortest path from one source to one (or every) destination. That is (P) travelling from one place to another in the shortest time, so P matches (iii).
  • The multi-stage graph algorithm finds a shortest route through a graph arranged in stages, visiting one place per stage. That fits (Q), a person visiting M places, so Q matches (ii).
  • Floyd's algorithm (Floyd-Warshall) finds shortest paths between all pairs of places. That fits (R), all N persons going to all M places, so R matches (i).

The matching is not among the four options. Option (a), the provisional key, gets R right but swaps P and Q. Options (b), (c) and (d) each get at most one pair right. So no option is correct on the standard reading, which fits ISRO's decision to withdraw the question.

No option is correct.

ISRO excluded this question from evaluation "for administrative reasons" in its December 2025 key, and the final key of 15 May 2026 kept it excluded. The provisional key (Oct 2025) had given (a).

Where this shows up in industry

Choosing between single-source Dijkstra and all-pairs Floyd-Warshall is a routine decision in routing, scheduling and network planning tools, where graph size decides which one is affordable.

ISRO 2025 · Computer Science Set A · Q.4Part A · 1 mark · -0.33 if wrongData structures and algorithms

Which statement is true about a greedy algorithm

Which of the following is true about Greedy algorithm?
  1. (a)does not make any optimization
  2. (b)makes local optimal decisions based on the selected criterion
  3. (c)requires exhaustive search of all possible solutions
  4. (d)is obvious in its correctness

Answer: (b) makes local optimal decisions based on the selected criterion Official final key: (b)

A greedy algorithm builds a solution step by step. At each step it picks the option that looks best right now, according to a fixed selection criterion (smallest edge weight, earliest finish time, highest value per weight and so on), and never reconsiders it.

So it does optimise, but locally (a is wrong). It avoids exhaustive search, which is what makes it fast (c is wrong). Its correctness is not obvious: for problems such as the fractional knapsack or Kruskal's spanning tree it needs a proof (greedy-choice property plus optimal substructure), and for the 0/1 knapsack it simply fails (d is wrong).

Makes local optimal decisions based on the selected criterion

Where this shows up in industry

Greedy heuristics (earliest deadline first, nearest-neighbour routing, Huffman and Kruskal) are used wherever a fast, good-enough decision beats an exact but slow search, such as task scheduling in real-time systems.

ISRO 2025 · Computer Science Set A · Q.5Part A · 1 mark · -0.33 if wrongDiscrete mathematics and probability

Maximum parks in regions enclosed by roads joining 20 malls, 3 roads each

There are 20 shopping malls in a city interconnected by roads. Each shopping mall can be reached through 3 different roads. It is assumed that
  • only one park is planned to be constructed in a region
  • roads do not cross each other
  • road(s) do not pass through a park.
What is the maximum number of parks that can be constructed in the regions enclosed by roads?
  1. (a)12
  2. (b)30
  3. (c)60
  4. (d)11

Answer: (a) 12 Official final key: (a) · provisional key was (d)

Model the city as a planar graph: malls are vertices, roads are edges, and roads do not cross. Every mall has degree 3, so .

With : roads.

Euler's formula for a connected planar graph is , where counts all regions (faces), including the unbounded outer one:

Each region can hold one park, so the maximum is 12 regions. Strictly, only 11 of these are completely enclosed by roads; the twelfth is the outer region around the road network. ISRO's final key counts all regions, which is the Euler-formula answer.

ISRO's provisional key (Oct 2025) gave (d); the December 2025 key changed it to (a), and the final key of 15 May 2026 kept (a).

Where this shows up in industry

Euler's formula for planar graphs is a quick sanity check on road, circuit-board and mesh layouts: it ties the number of nodes, links and regions together without drawing anything.

ISRO 2025 · Computer Science Set A · Q.6Part A · 1 mark · -0.33 if wrongDiscrete mathematics and probability

Chromatic number of the 8-vertex graph a to h

What is the chromatic number of the graph shown below?
afbehcdg
  1. (a)3
  2. (b)4
  3. (c)5
  4. (d)6

Answer: (a) 3 Official final key: (a) · provisional key was (b)

The chromatic number is the smallest number of colours that can colour the vertices so that no two adjacent vertices share a colour.

Edges read from the figure: , , , , , , , , , , , , (the line from to passes through , so it is and ).

Lower bound: are pairwise adjacent (a triangle), so at least 3 colours are needed. There is no 4-clique ( and are not adjacent), so nothing forces a fourth colour.

A 3-colouring: , , , , , , , . Check each edge: (3,1), (3,2), (1,3), (1,2), (1,3), (1,2), (3,2), (2,1), (2,3), (2,1), (1,3), (3,2), (3,1). No edge has equal colours.

ISRO's provisional key (Oct 2025) gave (b); the December 2025 key changed it to (a), and the final key of 15 May 2026 kept (a).

Where this shows up in industry

Graph colouring is the model behind register allocation in compilers and frequency or time-slot assignment, where conflicting items must get different values and fewer values is cheaper.

ISRO 2025 · Computer Science Set A · Q.7Part A · 1 mark · -0.33 if wrongData structures and algorithms

Which sort turns {10, 5, 16, 2, 8, 9} into {2, 5, 10, 16, 8, 9} in three iterations

Which of the following sorting technique was applied for an input array which gives array after three iterations?
  1. (a)Bubble sort
  2. (b)Selection sort
  3. (c)Insertion sort
  4. (d)All of the above

Answer: (c) Insertion sort Official final key: (c)

Run each sort on for three iterations.

  • Bubble sort: pass 1 gives , and after three passes the array is . The large values move right, so has not reached the front. No match.
  • Selection sort: pass 1 swaps the minimum to the front: ; pass 2 keeps ; pass 3 swaps to position 3. After three passes: . No match.
  • Insertion sort (iterations insert the 2nd, 3rd and 4th elements into the sorted prefix): , then , then . This matches.

Insertion sort leaves the unsorted tail untouched, and the front part sorted but not final, exactly as in the target array.

Insertion sort

ISRO excluded this question in its December 2025 key but reinstated it in the final key of 15 May 2026 with answer (c), the same as the provisional key.

Where this shows up in industry

Insertion sort's habit of keeping a sorted prefix is why it is used for small or nearly sorted buffers, for example as the final step inside hybrid sorts in standard libraries.

ISRO 2025 · Computer Science Set A · Q.8Part A · 1 mark · -0.33 if wrongInformation security

Number of unique AES keys for 16 persons communicating pairwise

How many keys are needed for encrypted communication among 16 persons using AES encryption assuming unique key is used between any two persons:
  1. (a)120 keys
  2. (b)256 keys
  3. (c)32 keys
  4. (d) keys

Answer: (a) 120 keys Official final key: (a)

AES is a symmetric cipher, so each pair of persons needs its own shared secret key. The number of keys is the number of distinct pairs among persons:

For : . The wrong options are (counts ordered pairs and self-pairs), and .

This quadratic growth is the key-distribution problem of symmetric cryptography.

keys

Where this shows up in industry

The growth is why large systems use a key distribution centre or public-key exchange instead of provisioning a separate secret for every pair of devices or ground stations.

ISRO 2025 · Computer Science Set A · Q.9Part A · 1 mark · -0.33 if wrongInformation security

Which of these does not use a key: message digest generation

Which of the following does not use key(s)?
  1. (a)Symmetric Ciphers
  2. (b)Generation of Message Digest
  3. (c)Generation of Message Authentication Code
  4. (d)Key exchange algorithms

Answer: (b) Generation of Message Digest Official final key: (b)

  • Symmetric ciphers (AES, DES) encrypt and decrypt with a shared secret key.
  • A message digest is the output of a plain hash function (SHA-256, for example). It takes only the message: anyone can recompute it, which is why a digest alone proves nothing about who sent the data.
  • A message authentication code is a keyed hash (such as HMAC). The key is what makes it authenticate the sender.
  • Key exchange algorithms (Diffie-Hellman and similar) work with secret or private values to produce a shared key.

Generation of Message Digest

Where this shows up in industry

Plain hashes are used to check file integrity (download checksums, firmware images), while keyed MACs are used on commands and telemetry where the receiver must also trust the sender.

ISRO 2025 · Computer Science Set A · Q.10Part A · 1 mark · -0.33 if wrongInformation security

Match information security terms with their definitions

Match the following in the context of Information Security.
(i)Non-repudiation(P)Protection of data from unauthorised disclosure
(ii)Data Integrity(Q)Prevention of unauthorised use of a resource
(iii)Data Confidentiality(R)Protection against denial by one of the entities involved in a communication having participated in all or part of communication
(iv)Access Control(S)Assurance that data arrived are exactly as sent by authorised entity
  1. (a)(i) - (R), (ii) - (P), (iii) - (Q), (iv) - (S)
  2. (b)(i) - (R), (ii) - (S), (iii) - (P), (iv) - (Q)
  3. (c)(i) - (P), (ii) - (R), (iii) - (S), (iv) - (Q)
  4. (d)(i) - (R), (ii) - (Q), (iii) - (P), (iv) - (S)

Answer: (b) (i) - (R), (ii) - (S), (iii) - (P), (iv) - (Q) Official final key: (b)

  • Non-repudiation: protection against one party in a communication denying that it took part. That is (R).
  • Data integrity: assurance that data arrived exactly as sent by the authorised entity. That is (S).
  • Data confidentiality: protection of data from unauthorised disclosure. That is (P).
  • Access control: preventing unauthorised use of a resource. That is (Q).

So the matching is (i)-(R), (ii)-(S), (iii)-(P), (iv)-(Q).

(i) - (R), (ii) - (S), (iii) - (P), (iv) - (Q)

Where this shows up in industry

These four services (alongside authentication and availability) are the standard checklist for reviewing a system's security design, for example for command uplinks and ground-station access.

ISRO 2025 · Computer Science Set A · Q.11Part A · 1 mark · -0.33 if wrongInformation security

Attack that persuades a user to reveal personal data

In which of the following cases, a user is persuaded to reveal personal data by hackers/cyber criminals?
  1. (a)Eavesdropping attack
  2. (b)SQL Injection attack
  3. (c)Phishing
  4. (d)Website Defacement

Answer: (c) Phishing Official final key: (c)

Phishing is a social-engineering attack: the attacker sends a fake email, message or website that looks genuine and tricks the user into handing over passwords, card numbers or other personal data.

The others do not depend on persuading the user. Eavesdropping silently listens to traffic. SQL injection abuses an application's database query through crafted input. Website defacement changes the content of a site.

Phishing

Where this shows up in industry

Phishing is the usual first step in breaches of organisations of every size, which is why mail filtering, multi-factor authentication and staff awareness training are standard controls.

ISRO 2025 · Computer Science Set A · Q.12Part A · 1 mark · -0.33 if wrongInformation security

Which protocol does not encrypt data: DNS using DNSSEC

Which of the following protocols/applications does not facilitate data encryption when data is sent over computer network using that protocol/application?
  1. (a)HTTPS
  2. (b)SNMPv3 (Version 3)
  3. (c)DNS using DNSSEC
  4. (d)SFTP

Answer: (c) DNS using DNSSEC Official final key: (c)

  • HTTPS is HTTP over TLS, so the data is encrypted.
  • SNMPv3 supports a privacy option that encrypts the message contents (alongside authentication).
  • SFTP runs over an SSH connection, which encrypts everything.
  • DNSSEC adds digital signatures to DNS records so that a resolver can verify they are authentic and unaltered. The records themselves still travel in clear text: DNSSEC gives authenticity and integrity, not confidentiality.

DNS using DNSSEC

Where this shows up in industry

Knowing which protocols only authenticate and which also encrypt decides what an observer on the network path can still read, which matters when choosing DNS, management and file-transfer protocols for a secure network.

ISRO 2025 · Computer Science Set A · Q.13Part A · 1 mark · -0.33 if wrongInformation security

Number of distinct encryption keys possible with an n-bit key

If an encryption key is bits long, then how many distinct encryption keys are possible?
  1. (a)
  2. (b)
  3. (c)
  4. (d)

Answer: (a) or (b) Official final key: (a or b), either accepted · provisional key was (a)

Each of the bits can be 0 or 1 independently, so the number of distinct bit patterns is

Check with a small case: gives 16 patterns, to . This is the standard textbook reading, so is the answer, and it is the size of the key space an attacker must search in brute force.

Option (b), , equals the number of keys if the all-zero key is excluded as unusable (15 for ). The question does not say that, so (b) is not the usual reading. ISRO's key nevertheless accepts both (a) and (b) after objections. Options (c) and (d) grow only quadratically and linearly in and are clearly wrong.

ISRO's provisional key (Oct 2025) gave (a); the December 2025 key accepted (a) or (b), and the final key of 15 May 2026 kept (a) or (b).

Where this shows up in industry

Key length is the first number quoted when sizing a cipher: each extra bit doubles the brute-force search space, which is why 128-bit and 256-bit keys are the norm for AES.

ISRO 2025 · Computer Science Set A · Q.14Part A · 1 mark · -0.33 if wrongInformation security

Threat a firewall can protect against: network scanning

Which of the following can be protected using a firewall?
  1. (a)Dumpster Diving
  2. (b)Physical Theft
  3. (c)Shoulder Surfing
  4. (d)Network Scanning

Answer: (d) Network Scanning Official final key: (d)

A firewall filters network traffic by rules (addresses, ports, protocol state). It can only stop threats that travel over the network.

Network scanning (probing hosts and open ports) is such a threat, so a firewall can block or limit it. Dumpster diving (digging through discarded papers and media), physical theft and shoulder surfing (watching someone type a password) happen in the physical world, and no packet filter can see or stop them.

Network Scanning

Where this shows up in industry

Perimeter firewalls and host firewalls are the first control between a ground-segment or office network and the outside world, while physical access control and shredding cover the non-network threats listed here.

ISRO 2025 · Computer Science Set A · Q.15Part A · 1 mark · -0.33 if wrongInformation security

Standard that defines the format for Public Key Infrastructure

Which of the following defines format for Public Key Infrastructure?
  1. (a)X.25
  2. (b)X.509
  3. (c)IEEE 802.11
  4. (d)IEEE 802.3

Answer: (b) X.509 Official final key: (b)

X.509 is the ITU-T standard that defines the format of public-key certificates (and certificate revocation lists), which is the building block of a public key infrastructure. TLS certificates for HTTPS are X.509 certificates.

The others are unrelated: X.25 is an old packet-switching network protocol, IEEE 802.11 is Wi-Fi and IEEE 802.3 is Ethernet.

X.509

Where this shows up in industry

Every HTTPS site, code-signing scheme and mutual-TLS service-to-service link relies on X.509 certificates, so engineers meet this format when issuing, rotating and debugging certificates.

ISRO 2025 · Computer Science Set A · Q.16Part A · 1 mark · -0.33 if wrongSoftware engineering

Which of these is not a type of Artificial Intelligence agent

Which of the following is not a type of Artificial Intelligence agent?
  1. (a)Simple AI agent
  2. (b)Unity based AI agent
  3. (c)Learning AI agent
  4. (d)Goal based AI agent

Answer: (b) Unity based AI agent Official final key: (b)

The standard classification of AI agents is: simple reflex agents, model-based reflex agents, goal-based agents, utility-based agents and learning agents.

Simple (reflex), goal-based and learning agents are all in this list. Unity based is not a recognised agent type (Unity is a game engine, not a kind of agent).

Unity based AI agent

Where this shows up in industry

Knowing the agent types (reflex, goal-based, learning) helps pick the right control architecture, from simple rule-based autonomy to systems that adapt from data.

ISRO 2025 · Computer Science Set A · Q.17Part A · 1 mark · -0.33 if wrongData structures and algorithms

Which of the following is an advantage of Artificial Intelligence

Which of the following is advantage of Artificial Intelligence?
  1. (a)Identifying a business trend to help in informed decision making
  2. (b)Helps in analysis of cyber security incidents
  3. (c)Based on learning capability, solves a pattern recognition problem
  4. (d)All of the above

Answer: (d) All of the above Official final key: (d)

AI and machine learning are used to find patterns in large data that people cannot scan by hand. Each of (a), (b) and (c) is a standard benefit: spotting business trends to support decisions, analysing security incidents (anomaly and intrusion detection), and learning from examples to solve pattern recognition problems.

Since all three statements are true, the correct choice is (d).

(d) All of the above

Where this shows up in industry

Telemetry anomaly detection, log analysis for security operations and image pattern recognition on satellite data all use exactly these capabilities.

ISRO 2025 · Computer Science Set A · Q.18Part A · 1 mark · -0.33 if wrongData structures and algorithms

What a decision tree in AI can be used for

A decision tree in AI can be used for
  1. (a)Only classification tasks
  2. (b)Only regression tasks
  3. (c)Both classification and regression tasks
  4. (d)Neither classification nor regression tasks

Answer: (c) Both classification and regression tasks Official final key: (c)

A decision tree repeatedly splits the data on a feature threshold. If the target is a class label, each split is chosen to make the groups purer (Gini impurity or entropy) and a leaf predicts the majority class. If the target is a number, each split is chosen to reduce variance and a leaf predicts the mean.

This is the CART idea (Classification And Regression Trees), so one structure serves both tasks.

(c) Both classification and regression tasks

Where this shows up in industry

Tree models are a common first choice for tabular engineering data: fault classification from sensor flags and regression of a temperature or load from other channels.

ISRO 2025 · Computer Science Set A · Q.19Part A · 1 mark · -0.33 if wrongData structures and algorithms

Algorithm used for reinforcement learning

Which of the following algorithms is used for Reinforcement learning?
  1. (a)Naive Bayes
  2. (b)Principal Component Analysis
  3. (c)Fast Forest Quantile
  4. (d)Q-learning

Answer: (c) Fast Forest Quantile or (d) Q-learning Official final key: (c or d), either accepted · provisional key was (d)

Reinforcement learning trains an agent by rewards from an environment. Q-learning is the textbook example: it keeps a table of action values and updates it with Q(s,a) += alpha*(r + gamma*max Q(s',a') - Q(s,a)).

The others are not reinforcement learning. Naive Bayes is a supervised classifier, Principal Component Analysis is unsupervised dimensionality reduction, and Fast Forest Quantile (a quantile regression forest, as named in ML.NET) is a supervised regression method. So by the standard definition only (d) fits.

(d) Q-learning

ISRO's final key accepts either (c) or (d) after objections. We found no reading under which Fast Forest Quantile is a reinforcement learning algorithm, so (d) is the one the textbook answer supports.

ISRO's provisional key (Oct 2025) gave (d); the December 2025 key changed it to (c) or (d), and the final key of 15 May 2026 kept it.

Where this shows up in industry

Reinforcement learning is used for control and scheduling problems such as autonomous navigation, resource allocation and robotic control policies.

ISRO 2025 · Computer Science Set A · Q.20Part A · 1 mark · -0.33 if wrongData structures and algorithms

Algorithm used for search and optimization in AI

Which of the following algorithms is used for search/optimization in AI?
  1. (a)Greedy Algorithm
  2. (b)Recursive Algorithm
  3. (c)Hill Climbing Algorithm
  4. (d)Hot Potato Algorithm

Answer: (a) Greedy Algorithm or (c) Hill Climbing Algorithm Official final key: (a or c), either accepted · provisional key was (c)

Hill climbing is a local search method: from the current state it moves to the best neighbour until none is better. Greedy search strategies, which always take the locally best choice (greedy best-first search), are likewise used for search and optimization in AI.

Recursion is a way of writing a procedure, not a search or optimization algorithm. Hot potato is a routing idea from networking (hand the packet off as soon as possible), not an AI method.

So the strictly correct options are (a) and (c); ISRO accepted either after objections, and the textbook example of the family is hill climbing.

(a) Greedy Algorithm or (c) Hill Climbing Algorithm

ISRO's provisional key (Oct 2025) gave (c); the December 2025 key changed it to (a) or (c), and the final key of 15 May 2026 kept it.

Where this shows up in industry

Local search and greedy heuristics are used for scheduling observations and ground-station passes when exact optimisation is too slow.

From ISRO prep to a computer science career

Talk to a counsellor.

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.