CSCI 570 Homework 4: Dynamic Programming & Network Flow Tutorial

Introduction: Why Dynamic Programming and Network Flow Matter

Dynamic programming (DP) and network flow are foundational topics in algorithm design, often featured in advanced courses like CSCI 570. This tutorial breaks down key problems from Homework 4—balloon bursting, coin games, bridge crossing, and flow network challenges—using timely analogies from gaming, AI, and school life. By connecting abstract concepts to familiar scenarios, you'll gain intuition for solving complex optimization problems.

1. Balloon Bursting: Maximizing Coins with DP

Imagine you're playing a virtual reality game where you burst balloons to earn coins. Each balloon has a number, and bursting it gives you coins equal to the product of its number and the numbers of its adjacent balloons. After bursting, neighbors become adjacent. This is exactly the burst balloons problem from CSCI 570. The goal is to maximize total coins.

DP Formulation

We define dp[i][j] as the maximum coins obtainable from bursting all balloons between indices i and j (exclusive). The recurrence considers the last balloon k to burst in that range:

dp[i][j] = max(dp[i][k] + nums[i]*nums[k]*nums[j] + dp[k][j]) for i < k < j

Base case: dp[i][i+1] = 0. Time complexity: O(n³). This mirrors optimizing resource allocation in AI resource management, where decisions depend on adjacent constraints.

2. Coin Game: Winning Strategy with DP

You and a friend take turns picking either the first or last coin from a row. Each coin has a value. You move first. What's the maximum amount you can guarantee? This classic DP problem teaches optimal play, similar to e-sports tournament strategies where players choose resources to maximize advantage.

DP Solution

Let dp[i][j] be the maximum net advantage (your total minus opponent's total) over the subarray coins[i..j]. Recurrence:

dp[i][j] = max(coins[i] - dp[i+1][j], coins[j] - dp[i][j-1])

Base: dp[i][i] = coins[i]. Answer: dp[0][n-1]. Time: O(n²). This DP is used in financial trading bots to decide buy/sell timing under uncertainty.

3. Octopus Bridge: Counting Safe Paths

Jack must cross a 3×n bridge with weak tiles. He can move forward or diagonally. How many safe paths? This DP problem resembles pathfinding in autonomous vehicles (like Tesla's FSD) navigating obstacles.

DP Formulation

Let ways[j][i] be number of ways to reach tile (j,i). Recurrence:

ways[1][i] = (ways[1][i-1] + ways[2][i-1]) if tile strong
ways[2][i] = (ways[1][i-1] + ways[2][i-1] + ways[3][i-1]) if strong
ways[3][i] = (ways[2][i-1] + ways[3][i-1]) if strong

Base: ways[j][0] = 1 for starting positions. Time: O(n). This is a classic DP for counting paths under constraints.

4. Network Flow: Edge Deletion and Min-Cut

Given a flow network, does deleting the edge that reduces max flow the most imply it's in every min-cut? This is a theory question from CSCI 570. The answer is: true. Proof: If edge e is not in some min-cut, there exists another min-cut with same capacity. Deleting e would not reduce max flow as much as deleting an edge in all min-cuts. This concept is critical in network reliability for cloud services like AWS.

5. Increasing Capacities: Does Min-Cut Persist?

If you increase every edge capacity by 1, does a given min-cut remain minimum? False. Example: a graph with two parallel edges from s to t of capacities 1 and 2. Min-cut capacity = 1 (cut containing the 1-edge). After +1, capacities become 2 and 3; min-cut capacity = 2 (still the 1-edge cut), but another cut might become smaller? Actually, the cut containing the 1-edge remains min-cut? Let's analyze: original min-cut capacity = 1. After +1, that cut capacity = 2. Another cut (both edges) capacity = 5. So it remains min-cut. But consider a graph where min-cut is not unique. For a simple path s->a->t with capacities 1 and 1, min-cut = 1 (either edge). After +1, each edge capacity = 2, min-cut = 2 (any single edge). So the original min-cut still works. However, if original graph had a cut of capacity 2 and another of 3, after +1, both increase by number of edges in cut. A cut with more edges gains more capacity, so the ordering may change. Example: s connected to t via two disjoint paths: one direct edge capacity 2, another via two edges each capacity 1. Original min-cut = 2 (direct edge). After +1, direct edge capacity = 3, the two-edge path cut capacity = (1+1)+(1+1)=4? Wait, each edge +1: the two edges become 2 each, so cut capacity = 2+2=4. So min-cut still 3. But if we had a cut with three edges each capacity 1, original capacity 3, after +1 becomes 6, while a cut with one edge capacity 4 becomes 5. So min-cut changes. Thus statement is false.

6. Max Flow and Min-Cut in Practice

Finding max flow and min-cut is essential for data routing in CDNs like Netflix. The Ford-Fulkerson algorithm runs in O(E * max_flow). For unit capacities, it's O(E * sqrt(V)) using Dinic.

7. Client-Base Station Assignment

Given n clients and k base stations with range R and load L, can we connect each client to a base station? Model as bipartite graph: left side clients, right side base stations (each split into L copies). Add edges if distance ≤ R. Then check if max flow equals n. This is a matching problem used in Uber's rider-driver assignment.

8. Deleting k Edges to Minimize Max Flow

To reduce max flow as much as possible by deleting k edges, find the min s-t cut and delete its edges. Since all capacities are 1, the min-cut capacity equals the number of edges crossing the cut. Deleting those k edges reduces flow by at least k. This is used in cybersecurity to identify critical links.

9. Detecting SPY in a Grid

Find maximum number of disjoint 'SPY' words in a grid moving orthogonally. Model as flow: create source to each S, each S to adjacent P, each P to adjacent Y, each Y to sink. All capacities 1. Max flow gives answer. This is similar to matching in game grids like Candy Crush.

10. Class Enrollment as Full-Time Student

Students need at least m classes each, classes have capacity. Model as flow: source to each student with capacity m, student to classes they want (capacity 1), class to sink with capacity qi. Max flow = n*m if feasible. This is a scheduling problem used in universities.

Conclusion

Dynamic programming and network flow are powerful tools for optimization. By relating them to AI decision-making, gaming strategies, and real-world logistics, you can master CSCI 570 concepts. Practice with these analogies to ace your homework and interviews.